Patent Application
20200371001
In a complex system, different components work together to function as the complex system. For example, an airplane may have electrical, mechanical and software components that work together for the airplane to land. An engineer may have different options for a given component in the system (e.g., different control systems or different settings for a control system for the landing gear of the airplane). An engineer testing a complex system can construct a test suite that represents different test cases for the system with selections for the different options for each of the components in the system. The test suite can be referred to as a combinatorial test suite in that it tests different combinations of configurable options for a complex system. If there are failures, the test engineer is faced with the task of identifying the option or combination of options that precipitated the failures (e.g., from a table of entries or summary statistics). When there are multiple components in the complex system, it can be difficult to visualize different options for each component and the results of testing those different options.
An engineer may design an experiment with test cases each test case specifying one of different options for each factor of the experiment (e.g., to test a complex system). A screening design, for instance, is useful for determining which active factors in the experiment affect the outcome. Data (e.g., input and response system data) can be used to generate a model (e.g., a machine learning algorithm model). Validations techniques can be used to validate data for generating the model (e.g., a K-fold cross-validation).
Experimenters in designing an experiment will sometimes employ a screening design to help identify what potential factors in the experiment may be active factors in determining an outcome of the experiment. A supersaturated design for an experiment is one in which the number of factors investigated in the experiment is larger than the number of test cases. Supersaturated designs are presently underused or avoided by experimenters because these designs cannot estimate all factors for an experiment simultaneously, and it can be difficult to identify active factors using current model selection techniques.
In an example embodiment, a computer-program product tangibly embodied in a non-transitory machine-readable storage medium is provided. The computer-program product includes instructions to cause a computing device to receive a request for a design of an experiment. The design of the experiment comprises a plurality of test cases. Each test case of the plurality of test cases comprises multiple test conditions for testing one of different factors for the experiment. A test condition of the multiple test conditions comprises one of different options for a given factor in the design of the experiment. The computer-program product includes instructions to cause a computing device to obtain for the design of the experiment: a run indication of a total number of the plurality of test cases for the design of the experiment; and a factor indication of a total number of the different factors for the design of the experiment. The total number of the different factors is greater than the total number of the plurality of test cases. The computer-program product includes instructions to cause a computing device to generate a data representation of a matrix Y. The matrix Y defines a supersaturated design for the design of the experiment. The generating the data representation is by: selecting, based on the run indication and factor indication, dimensions for the matrix Y; generating a data representation of a matrix X according to an obtained design; computing an indication of correlation between effects of factors of a matrix Y; and generating, based on the indication of correlation, the data representation of the matrix Y that is the transposition of the matrix X or is the transposition of a matrix X*. The matrix Y is a transposition of the matrix X or is a transposition of the matrix X*; and the matrix X* is a first subset of the matrix X such that the transposition of the matrix X* represents a same number of factors as the transposition of the matrix X. The computer-program product includes instructions to cause a computing device to output, to an output device, a setting for each test condition for test cases of the supersaturated design for the experiment according to the matrix Y.
In another example embodiment, a computing device is provided. The computing device includes, but is not limited to, a processor and memory. The memory contains instructions that, when executed by the processor, control the computing device to output a setting for each test condition for test cases of the supersaturated design for the experiment according to the matrix Y.
In another example embodiment, a method of outputting a setting for each test condition for test cases of the supersaturated design for the experiment according to the matrix Y is provided.
Other features and aspects of example embodiments are presented below in the Detailed Description when read in connection with the drawings presented with this application.
In the following description, for the purposes of explanation, specific details are set forth to provide a thorough understanding of embodiments of the technology. However, it will be apparent that various embodiments may be practiced without these specific details. The figures and description are not intended to be restrictive.
The ensuing description provides example embodiments only, and is not intended to limit the scope, applicability, or configuration of the disclosure. Rather, the ensuing description of the example embodiments will provide those skilled in the art with an enabling description for implementing an example embodiment. It should be understood that various changes may be made in the function and arrangement of elements without departing from the spirit and scope of the technology as set forth in the appended claims.
Specific details are given in the following description to provide a thorough understanding of the embodiments. However, it will be understood by one of ordinary skill in the art that the embodiments may be practiced without these specific details. For example, circuits, systems, networks, processes, and other components may be shown as components in block diagram form in order not to obscure the embodiments in unnecessary detail. In other instances, well-known circuits, processes, algorithms, structures, and techniques may be shown without unnecessary detail to avoid obscuring the embodiments.
Also, it is noted that individual embodiments may be described as a process which is depicted as a flowchart, a flow diagram, a data flow diagram, a structure diagram, or a block diagram. Although a flowchart may describe the operations as a sequential process, many of the operations can be performed in parallel or concurrently. In addition, the order of the operations may be re-arranged. A process is terminated when its operations are completed but could have additional operations not included in a figure. A process may correspond to a method, a function, a procedure, a subroutine, a subprogram, etc. When a process corresponds to a function, its termination can correspond to a return of the function to the calling function or the main function.
Systems depicted in some of the figures may be provided in various configurations. In some embodiments, the systems may be configured as a distributed system where one or more components of the system are distributed across one or more networks in a cloud computing system.
Data transmission network 100 may also include computing environment 114. Computing environment 114 may be a specialized computer or other machine that processes the data received within the data transmission network 100. Data transmission network 100 also includes one or more network devices 102. Network devices 102 may include client devices that attempt to communicate with computing environment 114. For example, network devices 102 may send data to the computing environment 114 to be processed, may send signals to the computing environment 114 to control different aspects of the computing environment or the data it is processing, among other reasons. Network devices 102 may interact with the computing environment 114 through a number of ways, such as, for example, over one or more networks 108. As shown in
In other embodiments, network devices may provide a large amount of data, either all at once or streaming over a period of time (e.g., using event stream processing (ESP), described further with respect to
Data transmission network 100 may also include one or more network-attached data stores 110. Network-attached data stores 110 are used to store data to be processed by the computing environment 114 as well as any intermediate or final data generated by the computing system in non-volatile memory. However in certain embodiments, the configuration of the computing environment 114 allows its operations to be performed such that intermediate and final data results can be stored solely in volatile memory (e.g., RAM), without a requirement that intermediate or final data results be stored to non-volatile types of memory (e.g., disk). This can be useful in certain situations, such as when the computing environment 114 receives ad hoc queries from a user and when responses, which are generated by processing large amounts of data, need to be generated on-the-fly. In this non-limiting situation, the computing environment 114 may be configured to retain the processed information within memory so that responses can be generated for the user at different levels of detail as well as allow a user to interactively query against this information.
Network-attached data stores may store a variety of different types of data organized in a variety of different ways and from a variety of different sources. For example, network-attached data storage may include storage other than primary storage located within computing environment 114 that is directly accessible by processors located therein. Network-attached data storage may include secondary, tertiary or auxiliary storage, such as large hard drives, servers, virtual memory, among other types. Storage devices may include portable or non-portable storage devices, optical storage devices, and various other mediums capable of storing, containing data. A machine-readable storage medium or computer-readable storage medium may include a non-transitory medium in which data can be stored and that does not include carrier waves and/or transitory electronic signals. Examples of a non-transitory medium may include, for example, a magnetic disk or tape, optical storage media such as compact disk or digital versatile disk, flash memory, memory or memory devices. A computer-program product may include code and/or machine-executable instructions that may represent a procedure, a function, a subprogram, a program, a routine, a subroutine, a module, a software package, a class, or any combination of instructions, data structures, or program statements. A code segment may be coupled to another code segment or a hardware circuit by passing and/or receiving information, data, arguments, parameters, or memory contents. Information, arguments, parameters, data, etc. may be passed, forwarded, or transmitted via any suitable means including memory sharing, message passing, token passing, network transmission, among others. Furthermore, the data stores may hold a variety of different types of data. For example, network-attached data stores 110 may hold unstructured (e.g., raw) data, such as manufacturing data (e.g., a database containing records identifying products being manufactured with parameter data for each product, such as colors and models) or product sales databases (e.g., a database containing individual data records identifying details of individual product sales).
The unstructured data may be presented to the computing environment 114 in different forms such as a flat file or a conglomerate of data records, and may have data values and accompanying time stamps. The computing environment 114 may be used to analyze the unstructured data in a variety of ways to determine the best way to structure (e.g., hierarchically) that data, such that the structured data is tailored to a type of further analysis that a user wishes to perform on the data. For example, after being processed, the unstructured time stamped data may be aggregated by time (e.g., into daily time period units) to generate time series data and/or structured hierarchically according to one or more dimensions (e.g., parameters, attributes, and/or variables). For example, data may be stored in a hierarchical data structure, such as a ROLAP OR MOLAP database, or may be stored in another tabular form, such as in a flat-hierarchy form.
Data transmission network 100 may also include one or more server farms 106. Computing environment 114 may route select communications or data to the one or more sever farms 106 or one or more servers within the server farms. Server farms 106 can be configured to provide information in a predetermined manner. For example, server farms 106 may access data to transmit in response to a communication. Server farms 106 may be separately housed from each other device within data transmission network 100, such as computing environment 114, and/or may be part of a device or system.
Server farms 106 may host a variety of different types of data processing as part of data transmission network 100. Server farms 106 may receive a variety of different data from network devices, from computing environment 114, from cloud network 116, or from other sources. The data may have been obtained or collected from one or more sensors, as inputs from a control database, or may have been received as inputs from an external system or device. Server farms 106 may assist in processing the data by turning raw data into processed data based on one or more rules implemented by the server farms. For example, sensor data may be analyzed to determine changes in an environment over time or in real-time.
Data transmission network 100 may also include one or more cloud networks 116. Cloud network 116 may include a cloud infrastructure system that provides cloud services. In certain embodiments, services provided by the cloud network 116 may include a host of services that are made available to users of the cloud infrastructure system on demand. Cloud network 116 is shown in
While each device, server and system in
Each communication within data transmission network 100 (e.g., between client devices, between a device and connection management system 150, between servers 106 and computing environment 114 or between a server and a device) may occur over one or more networks 108. Networks 108 may include one or more of a variety of different types of networks, including a wireless network, a wired network, or a combination of a wired and wireless network. Examples of suitable networks include the Internet, a personal area network, a local area network (LAN), a wide area network (WAN), or a wireless local area network (WLAN). A wireless network may include a wireless interface or combination of wireless interfaces. As an example, a network in the one or more networks 108 may include a short-range communication channel, such as a Bluetooth or a Bluetooth Low Energy channel. A wired network may include a wired interface. The wired and/or wireless networks may be implemented using routers, access points, bridges, gateways, or the like, to connect devices in the network 114, as will be further described with respect to
Some aspects may utilize the Internet of Things (IoT), where things (e.g., machines, devices, phones, sensors) can be connected to networks and the data from these things can be collected and processed within the things and/or external to the things. For example, the IoT can include sensors in many different devices, and high value analytics can be applied to identify hidden relationships and drive increased efficiencies. This can apply to both big data analytics and real-time (e.g., ESP) analytics. IoT may be implemented in various areas, such as for access (technologies that get data and move it), embed-ability (devices with embedded sensors), and services. Industries in the IoT space may automotive (connected car), manufacturing (connected factory), smart cities, energy and retail. This will be described further below with respect to
As noted, computing environment 114 may include a communications grid 120 and a transmission network database system 118. Communications grid 120 may be a grid-based computing system for processing large amounts of data. The transmission network database system 118 may be for managing, storing, and retrieving large amounts of data that are distributed to and stored in the one or more network-attached data stores 110 or other data stores that reside at different locations within the transmission network database system 118. The compute nodes in the grid-based computing system 120 and the transmission network database system 118 may share the same processor hardware, such as processors that are located within computing environment 114.
As shown in
Although network devices 204-209 are shown in
As noted, one type of system that may include various sensors that collect data to be processed and/or transmitted to a computing environment according to certain embodiments includes an oil drilling system. For example, the one or more drilling operation sensors may include surface sensors that measure a hook load, a fluid rate, a temperature and a density in and out of the wellbore, a standpipe pressure, a surface torque, a rotation speed of a drill pipe, a rate of penetration, a mechanical specific energy, etc. and downhole sensors that measure a rotation speed of a bit, fluid densities, downhole torque, downhole vibration (axial, tangential, lateral), a weight applied at a drill bit, an annular pressure, a differential pressure, an azimuth, an inclination, a dog leg severity, a measured depth, a vertical depth, a downhole temperature, etc. Besides the raw data collected directly by the sensors, other data may include parameters either developed by the sensors or assigned to the system by a client or other controlling device. For example, one or more drilling operation control parameters may control settings such as a mud motor speed to flow ratio, a bit diameter, a predicted formation top, seismic data, weather data, etc. Other data may be generated using physical models such as an earth model, a weather model, a seismic model, a bottom hole assembly model, a well plan model, an annular friction model, etc. In addition to sensor and control settings, predicted outputs, of for example, the rate of penetration, mechanical specific energy, hook load, flow in fluid rate, flow out fluid rate, pump pressure, surface torque, rotation speed of the drill pipe, annular pressure, annular friction pressure, annular temperature, equivalent circulating density, etc. may also be stored in the data warehouse.
In another example, another type of system that may include various sensors that collect data to be processed and/or transmitted to a computing environment according to certain embodiments includes a home automation or similar automated network in a different environment, such as an office space, school, public space, sports venue, or a variety of other locations. Network devices in such an automated network may include network devices that allow a user to access, control, and/or configure various home appliances located within the user's home (e.g., a television, radio, light, fan, humidifier, sensor, microwave, iron, and/or the like), or outside of the user's home (e.g., exterior motion sensors, exterior lighting, garage door openers, sprinkler systems, or the like). For example, network device 102 may include a home automation switch that may be coupled with a home appliance. In another embodiment, a network device can allow a user to access, control, and/or configure devices, such as office-related devices (e.g., copy machine, printer, or fax machine), audio and/or video related devices (e.g., a receiver, a speaker, a projector, a DVD player, or a television), media-playback devices (e.g., a compact disc player, a CD player, or the like), computing devices (e.g., a home computer, a laptop computer, a tablet, a personal digital assistant (PDA), a computing device, or a wearable device), lighting devices (e.g., a lamp or recessed lighting), devices associated with a security system, devices associated with an alarm system, devices that can be operated in an automobile (e.g., radio devices, navigation devices), and/or the like. Data may be collected from such various sensors in raw form, or data may be processed by the sensors to create parameters or other data either developed by the sensors based on the raw data or assigned to the system by a client or other controlling device.
In another example, another type of system that may include various sensors that collect data to be processed and/or transmitted to a computing environment according to certain embodiments includes a power or energy grid. A variety of different network devices may be included in an energy grid, such as various devices within one or more power plants, energy farms (e.g., wind farm, solar farm, among others) energy storage facilities, factories, homes and businesses of consumers, among others. One or more of such devices may include one or more sensors that detect energy gain or loss, electrical input or output or loss, and a variety of other efficiencies. These sensors may collect data to inform users of how the energy grid, and individual devices within the grid, may be functioning and how they may be made more efficient.
Network device sensors may also perform processing on data it collects before transmitting the data to the computing environment 114, or before deciding whether to transmit data to the computing environment 114. For example, network devices may determine whether data collected meets certain rules, for example by comparing data or values calculated from the data and comparing that data to one or more thresholds. The network device may use this data and/or comparisons to determine if the data should be transmitted to the computing environment 214 for further use or processing.
Computing environment 214 may include machines 220 and 240. Although computing environment 214 is shown in
Computing environment 214 can communicate with various devices via one or more routers 225 or other inter-network or intra-network connection components. For example, computing environment 214 may communicate with devices 230 via one or more routers 225. Computing environment 214 may collect, analyze and/or store data from or pertaining to communications, client device operations, client rules, and/or user-associated actions stored at one or more data stores 235. Such data may influence communication routing to the devices within computing environment 214, how data is stored or processed within computing environment 214, among other actions.
Notably, various other devices can further be used to influence communication routing and/or processing between devices within computing environment 214 and with devices outside of computing environment 214. For example, as shown in
In addition to computing environment 214 collecting data (e.g., as received from network devices, such as sensors, and client devices or other sources) to be processed as part of a big data analytics project, it may also receive data in real time as part of a streaming analytics environment. As noted, data may be collected using a variety of sources as communicated via different kinds of networks or locally. Such data may be received on a real-time streaming basis. For example, network devices may receive data periodically from network device sensors as the sensors continuously sense, monitor and track changes in their environments. Devices within computing environment 214 may also perform pre-analysis on data it receives to determine if the data received should be processed as part of an ongoing project. The data received and collected by computing environment 214, no matter what the source or method or timing of receipt, may be processed over a period of time for a client to determine results data based on the client's needs and rules.
The model can include layers 302-314. The layers are arranged in a stack. Each layer in the stack serves the layer one level higher than it (except for the application layer, which is the highest layer), and is served by the layer one level below it (except for the physical layer, which is the lowest layer). The physical layer is the lowest layer because it receives and transmits raw bites of data and is the farthest layer from the user in a communications system. On the other hand, the application layer is the highest layer because it interacts directly with a software application.
As noted, the model includes a physical layer 302. Physical layer 302 represents physical communication and can define parameters of that physical communication. For example, such physical communication may come in the form of electrical, optical, or electromagnetic signals. Physical layer 302 also defines protocols that may control communications within a data transmission network.
Link layer 304 defines links and mechanisms used to transmit (i.e., move) data across a network. The link layer manages node-to-node communications, such as within a grid computing environment. Link layer 304 can detect and correct errors (e.g., transmission errors in the physical layer 302). Link layer 304 can also include a media access control (MAC) layer and logical link control (LLC) layer.
Network layer 306 defines the protocol for routing within a network. In other words, the network layer coordinates transferring data across nodes in a same network (e.g., such as a grid computing environment). Network layer 306 can also define the processes used to structure local addressing within the network.
Transport layer 308 can manage the transmission of data and the quality of the transmission and/or receipt of that data. Transport layer 308 can provide a protocol for transferring data, such as, for example, a Transmission Control Protocol (TCP). Transport layer 308 can assemble and disassemble data frames for transmission. The transport layer can also detect transmission errors occurring in the layers below it.
Session layer 310 can establish, maintain, and manage communication connections between devices on a network. In other words, the session layer controls the dialogues or nature of communications between network devices on the network. The session layer may also establish checkpointing, adjournment, termination, and restart procedures.
Presentation layer 312 can provide translation for communications between the application and network layers. In other words, this layer may encrypt, decrypt and/or format data based on data types known to be accepted by an application or network layer.
Application layer 314 interacts directly with software applications and end users, and manages communications between them. Application layer 314 can identify destinations, local resource states or availability and/or communication content or formatting using the applications.
Intra-network connection components 322 and 324 are shown to operate in lower levels, such as physical layer 302 and link layer 304, respectively. For example, a hub can operate in the physical layer and a switch can operate in the link layer. Inter-network connection components 326 and 328 are shown to operate on higher levels, such as layers 306-314. For example, routers can operate in the network layer and network devices can operate in the transport, session, presentation, and application layers.
As noted, a computing environment 314 can interact with and/or operate on, in various embodiments, one, more, all or any of the various layers. For example, computing environment 314 can interact with a hub (e.g., via the link layer) so as to adjust which devices the hub communicates with. The physical layer may be served by the link layer, so it may implement such data from the link layer. For example, the computing environment 314 may control which devices it will receive data from. For example, if the computing environment 314 knows that a certain network device has turned off, broken, or otherwise become unavailable or unreliable, the computing environment 314 may instruct the hub to prevent any data from being transmitted to the computing environment 314 from that network device. Such a process may be beneficial to avoid receiving data that is inaccurate or that has been influenced by an uncontrolled environment. As another example, computing environment 314 can communicate with a bridge, switch, router or gateway and influence which device within the system (e.g., system 200) the component selects as a destination. In some embodiments, computing environment 314 can interact with various layers by exchanging communications with equipment operating on a particular layer by routing or modifying existing communications. In another embodiment, such as in a grid computing environment, a node may determine how data within the environment should be routed (e.g., which node should receive certain data) based on certain parameters or information provided by other layers within the model.
As noted, the computing environment 314 may be a part of a communications grid environment, the communications of which may be implemented as shown in the protocol of
Communications grid computing system (or just “communications grid”) 400 also includes one or more worker nodes. Shown in
A control node may connect with an external device with which the control node may communicate (e.g., a grid user, such as a server or computer, may connect to a controller of the grid). For example, a server or computer may connect to control nodes and may transmit a project or job to the node. The project may include a data set. The data set may be of any size. Once the control node receives such a project including a large data set, the control node may distribute the data set or projects related to the data set to be performed by worker nodes. Alternatively, for a project including a large data set, the data set may be receive or stored by a machine other than a control node (e.g., a Hadoop data node).
Control nodes may maintain knowledge of the status of the nodes in the grid (i.e., grid status information), accept work requests from clients, subdivide the work across worker nodes, coordinate the worker nodes, among other responsibilities. Worker nodes may accept work requests from a control node and provide the control node with results of the work performed by the worker node. A grid may be started from a single node (e.g., a machine, computer, server, etc.). This first node may be assigned or may start as the primary control node that will control any additional nodes that enter the grid.
When a project is submitted for execution (e.g., by a client or a controller of the grid) it may be assigned to a set of nodes. After the nodes are assigned to a project, a data structure (i.e., a communicator) may be created. The communicator may be used by the project for information to be shared between the project code running on each node. A communication handle may be created on each node. A handle, for example, is a reference to the communicator that is valid within a single process on a single node, and the handle may be used when requesting communications between nodes.
A control node, such as control node 402, may be designated as the primary control node. A server, computer or other external device may connect to the primary control node. Once the control node receives a project, the primary control node may distribute portions of the project to its worker nodes for execution. For example, when a project is initiated on communications grid 400, primary control node 402 controls the work to be performed for the project to complete the project as requested or instructed. The primary control node may distribute work to the worker nodes based on various factors, such as which subsets or portions of projects may be completed most efficiently and in the correct amount of time. For example, a worker node may perform analysis on a portion of data that is already local (e.g., stored on) the worker node. The primary control node also coordinates and processes the results of the work performed by each worker node after each worker node executes and completes its job. For example, the primary control node may receive a result from one or more worker nodes, and the control node may organize (e.g., collect and assemble) the results received and compile them to produce a complete result for the project received from the end user.
Any remaining control nodes, such as control nodes 404 and 406, may be assigned as backup control nodes for the project. In an embodiment, backup control nodes may not control any portion of the project. Instead, backup control nodes may serve as a backup for the primary control node and take over as primary control node if the primary control node were to fail. If a communications grid were to include only a single control node, and the control node were to fail (e.g., the control node is shut off or breaks) then the communications grid as a whole may fail and any project or job being run on the communications grid may fail and may not complete. While the project may be run again, such a failure may cause a delay (severe delay in some cases, such as overnight delay) in completion of the project Therefore, a grid with multiple control nodes, including a backup control node, may be beneficial.
To add another node or machine to the grid, the primary control node may open a pair of listening sockets, for example. A socket may be used to accept work requests from clients, and the second socket may be used to accept connections from other grid nodes). The primary control node may be provided with a list of other nodes (e.g., other machines, computers, servers) that will participate in the grid, and the role that each node will fill in the grid. Upon startup of the primary control node (e.g., the first node on the grid), the primary control node may use a network protocol to start the server process on every other node in the grid. Command line parameters, for example, may inform each node of one or more pieces of information, such as: the role that the node will have in the grid, the host name of the primary control node, the port number on which the primary control node is accepting connections from peer nodes, among others. The information may also be provided in a configuration file, transmitted over a secure shell tunnel, recovered from a configuration server, among others. While the other machines in the grid may not initially know about the configuration of the grid, that information may also be sent to each other node by the primary control node. Updates of the grid information may also be subsequently sent to those nodes.
For any control node other than the primary control node added to the grid, the control node may open three sockets. The first socket may accept work requests from clients, the second socket may accept connections from other grid members, and the third socket may connect (e.g., permanently) to the primary control node. When a control node (e.g., primary control node) receives a connection from another control node, it first checks to see if the peer node is in the list of configured nodes in the grid. If it is not on the list, the control node may clear the connection. If it is on the list, it may then attempt to authenticate the connection. If authentication is successful, the authenticating node may transmit information to its peer, such as the port number on which a node is listening for connections, the host name of the node, information about how to authenticate the node, among other information. When a node, such as the new control node, receives information about another active node, it will check to see if it already has a connection to that other node. If it does not have a connection to that node, it may then establish a connection to that control node.
Any worker node added to the grid may establish a connection to the primary control node and any other control nodes on the grid. After establishing the connection, it may authenticate itself to the grid (e.g., any control nodes, including both primary and backup, or a server or user controlling the grid). After successful authentication, the worker node may accept configuration information from the control node.
When a node joins a communications grid (e.g., when the node is powered on or connected to an existing node on the grid or both), the node is assigned (e.g., by an operating system of the grid) a universally unique identifier (UUID). This unique identifier may help other nodes and external entities (devices, users, etc.) to identify the node and distinguish it from other nodes. When a node is connected to the grid, the node may share its unique identifier with the other nodes in the grid. Since each node may share its unique identifier, each node may know the unique identifier of every other node on the grid. Unique identifiers may also designate a hierarchy of each of the nodes (e.g., backup control nodes) within the grid. For example, the unique identifiers of each of the backup control nodes may be stored in a list of backup control nodes to indicate an order in which the backup control nodes will take over for a failed primary control node to become a new primary control node. However, a hierarchy of nodes may also be determined using methods other than using the unique identifiers of the nodes. For example, the hierarchy may be predetermined, or may be assigned based on other predetermined factors.
The grid may add new machines at any time (e.g., initiated from any control node). Upon adding a new node to the grid, the control node may first add the new node to its table of grid nodes. The control node may also then notify every other control node about the new node. The nodes receiving the notification may acknowledge that they have updated their configuration information.
Primary control node 402 may, for example, transmit one or more communications to backup control nodes 404 and 406 (and, for example, to other control or worker nodes within the communications grid). Such communications may be sent periodically, at fixed time intervals, between known fixed stages of the project's execution, among other protocols. The communications transmitted by primary control node 402 may be of varied types and may include a variety of types of information. For example, primary control node 402 may transmit snapshots (e.g., status information) of the communications grid so that backup control node 404 always has a recent snapshot of the communications grid. The snapshot or grid status may include, for example, the structure of the grid (including, for example, the worker nodes in the grid, unique identifiers of the nodes, or their relationships with the primary control node) and the status of a project (including, for example, the status of each worker node's portion of the project). The snapshot may also include analysis or results received from worker nodes in the communications grid. The backup control nodes may receive and store the backup data received from the primary control node. The backup control nodes may transmit a request for such a snapshot (or other information) from the primary control node, or the primary control node may send such information periodically to the backup control nodes.
As noted, the backup data may allow the backup control node to take over as primary control node if the primary control node fails without requiring the grid to start the project over from scratch. If the primary control node fails, the backup control node that will take over as primary control node may retrieve the most recent version of the snapshot received from the primary control node and use the snapshot to continue the project from the stage of the project indicated by the backup data. This may prevent failure of the project as a whole.
A backup control node may use various methods to determine that the primary control node has failed. In one example of such a method, the primary control node may transmit (e.g., periodically) a communication to the backup control node that indicates that the primary control node is working and has not failed, such as a heartbeat communication. The backup control node may determine that the primary control node has failed if the backup control node has not received a heartbeat communication for a certain predetermined period of time. Alternatively, a backup control node may also receive a communication from the primary control node itself (before it failed) or from a worker node that the primary control node has failed, for example because the primary control node has failed to communicate with the worker node.
Different methods may be performed to determine which backup control node of a set of backup control nodes (e.g., backup control nodes 404 and 406) will take over for failed primary control node 402 and become the new primary control node. For example, the new primary control node may be chosen based on a ranking or “hierarchy” of backup control nodes based on their unique identifiers. In an alternative embodiment, a backup control node may be assigned to be the new primary control node by another device in the communications grid or from an external device (e.g., a system infrastructure or an end user, such as a server or computer, controlling the communications grid). In another alternative embodiment, the backup control node that takes over as the new primary control node may be designated based on bandwidth or other statistics about the communications grid.
A worker node within the communications grid may also fail. If a worker node fails, work being performed by the failed worker node may be redistributed amongst the operational worker nodes. In an alternative embodiment, the primary control node may transmit a communication to each of the operable worker nodes still on the communications grid that each of the worker nodes should purposefully fail also. After each of the worker nodes fail, they may each retrieve their most recent saved checkpoint of their status and re-start the project from that checkpoint to minimize lost progress on the project being executed.
The process may also include receiving a failure communication corresponding to a node in the communications grid in operation 506. For example, a node may receive a failure communication including an indication that the primary control node has failed, prompting a backup control node to take over for the primary control node. In an alternative embodiment, a node may receive a failure that a worker node has failed, prompting a control node to reassign the work being performed by the worker node. The process may also include reassigning a node or a portion of the project being executed by the failed node, as described in operation 508. For example, a control node may designate the backup control node as a new primary control node based on the failure communication upon receiving the failure communication. If the failed node is a worker node, a control node may identify a project status of the failed worker node using the snapshot of the communications grid, where the project status of the failed worker node includes a status of a portion of the project being executed by the failed worker node at the failure time.
The process may also include receiving updated grid status information based on the reassignment, as described in operation 510, and transmitting a set of instructions based on the updated grid status information to one or more nodes in the communications grid, as described in operation 512. The updated grid status information may include an updated project status of the primary control node or an updated project status of the worker node. The updated information may be transmitted to the other nodes in the grid to update their stale stored information.
Similar to in
Each node also includes a data store 624. Data stores 624, similar to network-attached data stores 110 in
Each node also includes a user-defined function (UDF) 626. The UDF provides a mechanism for the DMBS 628 to transfer data to or receive data from the database stored in the data stores 624 that are managed by the DBMS. For example, UDF 626 can be invoked by the DBMS to provide data to the GESC for processing. The UDF 626 may establish a socket connection (not shown) with the GESC to transfer the data. Alternatively, the UDF 626 can transfer data to the GESC by writing data to shared memory accessible by both the UDF and the GESC.
The GESC 620 at the nodes 602 and 620 may be connected via a network, such as network 108 shown in
DMBS 628 may control the creation, maintenance, and use of database or data structure (not shown) within nodes 602 or 610. The database may organize data stored in data stores 624. The DMBS 628 at control node 602 may accept requests for data and transfer the appropriate data for the request. With such a process, collections of data may be distributed across multiple physical locations. In this example, each node 602 and 610 stores a portion of the total data managed by the management system in its associated data store 624.
Furthermore, the DBMS may be responsible for protecting against data loss using replication techniques. Replication includes providing a backup copy of data stored on one node on one or more other nodes. Therefore, if one node fails, the data from the failed node can be recovered from a replicated copy residing at another node. However, as described herein with respect to
To initiate the project, the control node may determine if the query requests use of the grid-based computing environment to execute the project. If the determination is no, then the control node initiates execution of the project in a solo environment (e.g., at the control node), as described in operation 710. If the determination is yes, the control node may initiate execution of the project in the grid-based computing environment, as described in operation 706. In such a situation, the request may include a requested configuration of the grid. For example, the request may include a number of control nodes and a number of worker nodes to be used in the grid when executing the project After the project has been completed, the control node may transmit results of the analysis yielded by the grid, as described in operation 708. Whether the project is executed in a solo or grid-based environment, the control node provides the results of the project in operation 712.
As noted with respect to
The ESPE may receive streaming data over a period of time related to certain events, such as events or other data sensed by one or more network devices. The ESPE may perform operations associated with processing data created by the one or more devices. For example, the ESPE may receive data from the one or more network devices 204-209 shown in
The engine container is the top-level container in a model that manages the resources of the one or more projects 802. In an illustrative embodiment, for example, there may be only one ESPE 800 for each instance of the ESP application, and ESPE 800 may have a unique engine name. Additionally, the one or more projects 802 may each have unique project names, and each query may have a unique continuous query name and begin with a uniquely named source window of the one or more source windows 806. ESPE 800 may or may not be persistent.
Continuous query modeling involves defining directed graphs of windows for event stream manipulation and transformation. A window in the context of event stream manipulation and transformation is a processing node in an event stream processing model. A window in a continuous query can perform aggregations, computations, pattern-matching, and other operations on data flowing through the window. A continuous query may be described as a directed graph of source, relational, pattern matching, and procedural windows. The one or more source windows 806 and the one or more derived windows 808 represent continuously executing queries that generate updates to a query result set as new event blocks stream through ESPE 800. A directed graph, for example, is a set of nodes connected by edges, where the edges have a direction associated with them.
An event object may be described as a packet of data accessible as a collection of fields, with at least one of the fields defined as a key or unique identifier (ID). The event object may be created using a variety of formats including binary, alphanumeric, XML, etc. Each event object may include one or more fields designated as a primary identifier (ID) for the event so ESPE 800 can support operation codes (opcodes) for events including insert, update, upsert, and delete. Upsert opcodes update the event if the key field already exists; otherwise, the event is inserted. For illustration, an event object may be a packed binary representation of a set of field values and include both metadata and field data associated with an event. The metadata may include an opcode indicating if the event represents an insert, update, delete, or upsert, a set of flags indicating if the event is a normal, partial-update, or a retention generated event from retention policy management, and a set of microsecond timestamps that can be used for latency measurements.
An event block object may be described as a grouping or package of event objects. An event stream may be described as a flow of event block objects. A continuous query of the one or more continuous queries 804 transforms a source event stream made up of streaming event block objects published into ESPE 800 into one or more output event streams using the one or more source windows 806 and the one or more derived windows 808. A continuous query can also be thought of as data flow modeling.
The one or more source windows 806 are at the top of the directed graph and have no windows feeding into them. Event streams are published into the one or more source windows 806, and from there, the event streams may be directed to the next set of connected windows as defined by the directed graph. The one or more derived windows 808 are all instantiated windows that are not source windows and that have other windows streaming events into them. The one or more derived windows 808 may perform computations or transformations on the incoming event streams. The one or more derived windows 808 transform event streams based on the window type (that is operators such as join, filter, compute, aggregate, copy, pattern match, procedural, union, etc.) and window settings. As event streams are published into ESPE 800, they are continuously queried, and the resulting sets of derived windows in these queries are continuously updated.
Within the application, a user may interact with one or more user interface windows presented to the user in a display under control of the ESPE independently or through a browser application in an order selectable by the user. For example, a user may execute an ESP application, which causes presentation of a first user interface window, which may include a plurality of menus and selectors such as drop down menus, buttons, text boxes, hyperlinks, etc. associated with the ESP application as understood by a person of skill in the art. As further understood by a person of skill in the art, various operations may be performed in parallel, for example, using a plurality of threads.
At operation 900, an ESP application may define and start an ESPE, thereby instantiating an ESPE at a device, such as machine 220 and/or 240. In an operation 902, the engine container is created. For illustration, ESPE 800 may be instantiated using a function call that specifies the engine container as a manager for the model.
In an operation 904, the one or more continuous queries 804 are instantiated by ESPE 800 as a model. The one or more continuous queries 804 may be instantiated with a dedicated thread pool or pools that generate updates as new events stream through ESPE 800. For illustration, the one or more continuous queries 804 may be created to model business processing logic within ESPE 800, to predict events within ESPE 800, to model a physical system within ESPE 800, to predict the physical system state within ESPE 800, etc. For example, as noted, ESPE 800 may be used to support sensor data monitoring and management (e.g., sensing may include force, torque, load, strain, position, temperature, air pressure, fluid flow, chemical properties, resistance, electromagnetic fields, radiation, irradiance, proximity, acoustics, moisture, distance, speed, vibrations, acceleration, electrical potential, or electrical current, etc.).
ESPE 800 may analyze and process events in motion or “event streams.” Instead of storing data and running queries against the stored data, ESPE 800 may store queries and stream data through them to allow continuous analysis of data as it is received. The one or more source windows 806 and the one or more derived windows 808 may be created based on the relational, pattern matching, and procedural algorithms that transform the input event streams into the output event streams to model, simulate, score, test, predict, etc. based on the continuous query model defined and application to the streamed data.
In an operation 906, a publish/subscribe (pub/sub) capability is initialized for ESPE 800. In an illustrative embodiment, a pub/sub capability is initialized for each project of the one or more projects 802. To initialize and enable pub/sub capability for ESPE 800, a port number may be provided. Pub/sub clients can use a host name of an ESP device running the ESPE and the port number to establish pub/sub connections to ESPE 800.
Publish-subscribe is a message-oriented interaction paradigm based on indirect addressing. Processed data recipients specify their interest in receiving information from ESPE 800 by subscribing to specific classes of events, while information sources publish events to ESPE 800 without directly addressing the receiving parties. ESPE 800 coordinates the interactions and processes the data. In some cases, the data source receives confirmation that the published information has been received by a data recipient.
A publish/subscribe API may be described as a library that enables an event publisher, such as publishing device 1022, to publish event streams into ESPE 800 or an event subscriber, such as event subscribing device A 1024a, event subscribing device B 1024b, and event subscribing device C 1024c, to subscribe to event streams from ESPE 800. For illustration, one or more publish/subscribe APIs may be defined. Using the publish/subscribe API, an event publishing application may publish event streams into a running event stream processor project source window of ESPE 800, and the event subscription application may subscribe to an event stream processor project source window of ESPE 800.
The publish/subscribe API provides cross-platform connectivity and endianness compatibility between ESP application and other networked applications, such as event publishing applications instantiated at publishing device 1022, and event subscription applications instantiated at one or more of event subscribing device A 1024a, event subscribing device B 1024b, and event subscribing device C 1024c.
Referring back to
ESP subsystem 800 may include a publishing client 1002, ESPE 800, a subscribing client A 1004, a subscribing client B 1006, and a subscribing client C 1008. Publishing client 1002 may be started by an event publishing application executing at publishing device 1022 using the publish/subscribe API. Subscribing client A 1004 may be started by an event subscription application A, executing at event subscribing device A 1024a using the publish/subscribe API. Subscribing client B 1006 may be started by an event subscription application B executing at event subscribing device B 1024b using the publish/subscribe API. Subscribing client C 1008 may be started by an event subscription application C executing at event subscribing device C 1024c using the publish/subscribe API.
An event block object containing one or more event objects is injected into a source window of the one or more source windows 806 from an instance of an event publishing application on event publishing device 1022. The event block object may be generated, for example, by the event publishing application and may be received by publishing client 1002. A unique ID may be maintained as the event block object is passed between the one or more source windows 806 and/or the one or more derived windows 808 of ESPE 800, and to subscribing client A 1004, subscribing client B 806, and subscribing client C 808 and to event subscription device A 1024a, event subscription device B 1024b, and event subscription device C 1024c. Publishing client 1002 may further generate and include a unique embedded transaction ID in the event block object as the event block object is processed by a continuous query, as well as the unique ID that publishing device 1022 assigned to the event block object.
In an operation 912, the event block object is processed through the one or more continuous queries 804. In an operation 914, the processed event block object is output to one or more computing devices of the event subscribing devices 1024a-c. For example, subscribing client A 1004, subscribing client B 1006, and subscribing client C 1008 may send the received event block object to event subscription device A 1024a, event subscription device B 1024b, and event subscription device C 1024c, respectively.
ESPE 800 maintains the event block containership aspect of the received event blocks from when the event block is published into a source window and works its way through the directed graph defined by the one or more continuous queries 804 with the various event translations before being output to subscribers. Subscribers can correlate a group of subscribed events back to a group of published events by comparing the unique ID of the event block object that a publisher, such as publishing device 1022, attached to the event block object with the event block ID received by the subscriber.
In an operation 916, a determination is made concerning whether or not processing is stopped. If processing is not stopped, processing continues in operation 910 to continue receiving the one or more event streams containing event block objects from the, for example, one or more network devices. If processing is stopped, processing continues in an operation 918. In operation 918, the started projects are stopped. In operation 920, the ESPE is shutdown.
As noted, in some embodiments, big data is processed for an analytics project after the data is received and stored. In other embodiments, distributed applications process continuously flowing data in real-time from distributed sources by applying queries to the data before distributing the data to geographically distributed recipients. As noted, an event stream processing engine (ESPE) may continuously apply the queries to the data as it is received and determines which entities receive the processed data. This allows for large amounts of data being received and/or collected in a variety of environments to be processed and distributed in real time. For example, as shown with respect to
Aspects of the current disclosure provide technical solutions to technical problems, such as computing problems that arise when an ESP device fails which results in a complete service interruption and potentially significant data loss. The data loss can be catastrophic when the streamed data is supporting mission critical operations such as those in support of an ongoing manufacturing or drilling operation. An embodiment of an ESP system achieves a rapid and seamless failover of ESPE running at the plurality of ESP devices without service interruption or data loss, thus significantly improving the reliability of an operational system that relies on the live or real-time processing of the data streams. The event publishing systems, the event subscribing systems, and each ESPE not executing ata failed ESP device are not aware of or effected by the failed ESP device. The ESP system may include thousands of event publishing systems and event subscribing systems. The ESP system keeps the failover logic and awareness within the boundaries of out-messaging network connector and out-messaging network device.
In one example embodiment, a system is provided to support a failover when event stream processing (ESP) event blocks. The system includes, but is not limited to, an out-messaging network device and a computing device. The computing device includes, but is not limited to, a processor and a computer-readable medium operably coupled to the processor. The processor is configured to execute an ESP engine (ESPE). The computer-readable medium has instructions stored thereon that, when executed by the processor, cause the computing device to support the failover. An event block object is received from the ESPE that includes a unique identifier. A first status of the computing device as active or standby is determined. When the first status is active, a second status of the computing device as newly active or not newly active is determined. Newly active is determined when the computing device is switched from a standby status to an active status. When the second status is newly active, a last published event block object identifier that uniquely identifies a last published event block object is determined. A next event block object is selected from a non-transitory computer-readable medium accessible by the computing device. The next event block object has an event block object identifier that is greater than the determined last published event block object identifier. The selected next event block object is published to an out-messaging network device. When the second status of the computing device is not newly active, the received event block object is published to the out-messaging network device. When the first status of the computing device is standby, the received event block object is stored in the non-transitory computer-readable medium.
Different machine-learning models may be used interchangeably to perform a task. Examples of tasks that can be performed at least partially using machine-learning models include various types of scoring; bioinformatics; cheminformatics; software engineering; fraud detection; customer segmentation; generating online recommendations; adaptive websites; determining customer lifetime value; search engines; placing advertisements in real time or near real time; classifying DNA sequences; affective computing; performing natural language processing and understanding; object recognition and computer vision; robotic locomotion; playing games; optimization and metaheuristics; detecting network intrusions; medical diagnosis and monitoring; or predicting when an asset, such as a machine, will need maintenance.
Any number and combination of tools can be used to create machine-learning models. Examples of tools for creating and managing machine-learning models can include SAS® Enterprise Miner, SAS® Rapid Predictive Modeler, and SAS® Model Manager, SAS Cloud Analytic Services (CAS)®, SAS Viya® of all which are by SAS Institute Inc. of Cary, N.C.
Machine-learning models can be constructed through an at least partially automated (e.g., with little or no human involvement) process called training. During training, input data can be iteratively supplied to a machine-learning model to enable the machine-learning model to identify patterns related to the input data or to identify relationships between the input data and output data. With training, the machine-learning model can be transformed from an untrained state to a trained state. Input data can be split into one or more training sets and one or more validation sets, and the training process may be repeated multiple times. The splitting may follow a k-fold cross-validation rule, a leave-one-out-rule, a leave-p-out rule, or a holdout rule. An overview of training and using a machine-learning model is described below with respect to the flow chart of
In block 1104, training data is received. In some examples, the training data is received from a remote database or a local database, constructed from various subsets of data, or input by a user. The training data can be used in its raw form for training a machine-learning model or pre-processed into another form, which can then be used for training the machine-learning model. For example, the raw form of the training data can be smoothed, truncated, aggregated, clustered, or otherwise manipulated into another form, which can then be used for training the machine-learning model.
In block 1106, a machine-learning model is trained using the training data. The machine-learning model can be trained in a supervised, unsupervised, or semi-supervised manner. In supervised training, each input in the training data is correlated to a desired output. This desired output may be a scalar, a vector, or a different type of data structure such as text or an image. This may enable the machine-learning model to learn a mapping between the inputs and desired outputs. In unsupervised training, the training data includes inputs, but not desired outputs, so that the machine-learning model has to find structure in the inputs on its own. In semi-supervised training, only some of the inputs in the training data are correlated to desired outputs.
In block 1108, the machine-learning model is evaluated. For example, an evaluation dataset can be obtained, for example, via user input or from a database. The evaluation dataset can include inputs correlated to desired outputs. The inputs can be provided to the machine-learning model and the outputs from the machine-learning model can be compared to the desired outputs. If the outputs from the machine-learning model closely correspond with the desired outputs, the machine-learning model may have a high degree of accuracy. For example, if 90% or more of the outputs from the machine-learning model are the same as the desired outputs in the evaluation dataset, the machine-learning model may have a high degree of accuracy. Otherwise, the machine-learning model may have a low degree of accuracy. The 90% number is an example only. A realistic and desirable accuracy percentage is dependent on the problem and the data.
In some examples, if the machine-learning model has an inadequate degree of accuracy for a particular task, the process can return to block 1106, where the machine-learning model can be further trained using additional training data or otherwise modified to improve accuracy. If the machine-learning model has an adequate degree of accuracy for the particular task, the process can continue to block 1110.
In block 1110, new data is received. In some examples, the new data is received from a remote database or a local database, constructed from various subsets of data, or input by a user. The new data may be unknown to the machine-learning model. For example, the machine-learning model may not have previously processed or analyzed the new data.
In block 1112, the trained machine-learning model is used to analyze the new data and provide a result. For example, the new data can be provided as input to the trained machine-learning model. The trained machine-learning model can analyze the new data and provide a result that includes a classification of the new data into a particular class, a clustering of the new data into a particular group, a prediction based on the new data, or any combination of these.
In block 1114, the result is post-processed. For example, the result can be added to, multiplied with, or otherwise combined with other data as part of a job. As another example, the result can be transformed from a first format, such as a time series format, into another format, such as a count series format. Any number and combination of operations can be performed on the result during post-processing.
A more specific example of a machine-learning model is the neural network 1200 shown in
The neurons and connections between the neurons can have numeric weights, which can be tuned during training. For example, training data can be provided to the input layer 1202 of the neural network 1200, and the neural network 1200 can use the training data to tune one or more numeric weights of the neural network 1200. In some examples, the neural network 1200 can be trained using backpropagation. Backpropagation can include determining a gradient of a particular numeric weight based on a difference between an actual output of the neural network 1200 and a desired output of the neural network 1200. Based on the gradient, one or more numeric weights of the neural network 1200 can be updated to reduce the difference, thereby increasing the accuracy of the neural network 1200. This process can be repeated multiple times to train the neural network 1200. For example, this process can be repeated hundreds or thousands of times to train the neural network 1200.
In some examples, the neural network 1200 is a feed-forward neural network. In a feed-forward neural network, every neuron only propagates an output value to a subsequent layer of the neural network 1200. For example, data may only move one direction (forward) from one neuron to the next neuron in a feed-forward neural network.
In other examples, the neural network 1200 is a recurrent neural network. A recurrent neural network can include one or more feedback loops, allowing data to propagate in both forward and backward through the neural network 1200. This can allow for information to persist within the recurrent neural network. For example, a recurrent neural network can determine an output based at least partially on information that the recurrent neural network has seen before, giving the recurrent neural network the ability to use previous input to inform the output.
In some examples, the neural network 1200 operates by receiving a vector of numbers from one layer; transforming the vector of numbers into a new vector of numbers using a matrix of numeric weights, a nonlinearity, or both; and providing the new vector of numbers to a subsequent layer of the neural network 1200. Each subsequent layer of the neural network 1200 can repeat this process until the neural network 1200 outputs a final result at the output layer 1206. For example, the neural network 1200 can receive a vector of numbers as an input at the input layer 1202. The neural network 1200 can multiply the vector of numbers by a matrix of numeric weights to determine a weighted vector. The matrix of numeric weights can be tuned during the training of the neural network 1200. The neural network 1200 can transform the weighted vector using a nonlinearity, such as a sigmoid tangent or the hyperbolic tangent. In some examples, the nonlinearity can include a rectified linear unit, which can be expressed using the following equation:
y=max(x,0)
where y is the output and x is an input value from the weighted vector. The transformed output can be supplied to a subsequent layer, such as the hidden layer 1204, of the neural network 1200. The subsequent layer of the neural network 1200 can receive the transformed output, multiply the transformed output by a matrix of numeric weights and a nonlinearity, and provide the result to yet another layer of the neural network 1200. This process continues until the neural network 1200 outputs a final result at the output layer 1206.
Other examples of the present disclosure may include any number and combination of machine-learning models having any number and combination of characteristics. The machine-learning model(s) can be trained in a supervised, semi-supervised, or unsupervised manner, or any combination of these. The machine-learning model(s) can be implemented using a single computing device or multiple computing devices, such as the communications grid computing system 400 discussed above.
Implementing some examples of the present disclosure at least in part by using machine-learning models can reduce the total number of processing iterations, time, memory, electrical power, or any combination of these consumed by a computing device when analyzing data. For example, a neural network may more readily identify patterns in data than other approaches. This may enable the neural network to analyze the data using fewer processing cycles and less memory than other approaches, while obtaining a similar or greater level of accuracy.
Some machine-learning approaches may be more efficiently and speedily executed and processed with machine-learning specific processors (e.g., not a generic CPU). Such processors may also provide an energy savings when compared to generic CPUs. For example, some of these processors can include a graphical processing unit (GPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), an artificial intelligence (AI) accelerator, a neural computing core, a neural computing engine, a neural processing unit, a purpose-built chip architecture for deep learning, and/or some other machine-learning specific processor that implements a machine learning approach or one or more neural networks using semiconductor (e.g., silicon (Si), gallium arsenide (GaAs)) devices. Furthermore, these processors may also be employed in heterogeneous computing architectures with a number of and a variety of different types of cores, engines, nodes, and/or layers to achieve various energy efficiencies, processing speed improvements, data communication speed improvements, and/or data efficiency targets and improvements throughout various parts of the system when compared to a homogeneous computing architecture that employs CPUs for general purpose computing.
The computing device 1302 has a computer-readable medium 1312 and a processor 1314. Computer-readable medium 1312 is an electronic holding place or storage for information so the information can be accessed by processor 1314. Computer-readable medium 1312 can include, but is not limited to, any type of random access memory (RAM), any type of read only memory (ROM), any type of flash memory, etc. such as magnetic storage devices (e.g., hard disk, floppy disk, magnetic strips), optical disks (e.g., compact disc (CD), digital versatile disc (DVD)), smart cards, flash memory devices, etc.
Processor 1314 executes instructions (e.g., stored at the computer readable medium 1312). The instructions can be carried out by a special purpose computer, logic circuits, or hardware circuits. In one or more embodiments, processor 1314 is implemented in hardware and/or firmware. Processor 1314 executes an instruction, meaning it performs or controls the operations called for by that instruction. The term “execution” is the process of running an application or the carrying out of the operation called for by an instruction. The instructions can be written using one or more programming language, scripting language, assembly language, etc. Processor 1314 in one or more embodiments can retrieve a set of instructions from a permanent memory device and copy the instructions in an executable form to a temporary memory device that is generally some form of RAM, for example.
In one or more embodiments computer-readable medium 1312 stores instructions for execution by processor 1314. For example, computer-readable medium 1312 comprises instructions for a test suite application 1360 and cause indicators application 1362.
In one or more embodiments, the test suite application 1360 determines or generates a test suite that provides test cases for testing a complex system. This is useful, for instance, to identify a most likely potential cause 1350 for a potential failure of the system. Complex systems can include software, hardware, mechanical, chemical, biological components, combinations thereof, or computer-simulated models thereof. For example, a complex system could include an airplane as discussed in the background. As another example the test cases could be used to test a mobile phone with different hardware components (e.g., camera types) and different software configurations (e.g., display options and ringtone configurations). Test cases are used to test the mobile phone or a computer-simulation of the mobile phone to determine whether a particular display configuration with a particular camera could cause the mobile phone to fail (e.g., in taking a picture). In another example, the complex system could be a microwave oven and test cases are used to test whether the microwave oven fails (e.g., a control module of the microwave oven fails) when the power is set to one configuration (e.g. “High” or “Low”) and the microwave oven receives a certain time input (e.g. 20 minutes or 30 minutes). In yet another example, the complex system could be a sprinkler fire system with different biological agents or chemical agents for corrosion prevention in the system with different piping and mechanical and/or electrical valve options for transporting the water in the sprinkler system. Test cases are used to test whether a particular chemical agent in combination with a particular piping could cause a particular valve to malfunction and the sprinkler system to fail (e.g., fail to initiate in the presence of a fire).
In one or more embodiments, the cause indictors application 1362 determines or generates cause indicators that represent a likelihood that a test condition or combination of test conditions of a test case of the test suite caused a failed test case. For example, in the case of a complex system that is a mobile phone as described above, if the mobile phone fails when taking a picture, a combination of the display option and camera type used when the mobile phone failed could have caused the mobile phone to fail. Alternatively, a combination of the display option and a ringtone configuration used when the mobile phone failed could have caused the mobile phone to fail, or a combination of a camera type and ringtone configuration. Cause indicators would represent the likelihood that it was the display option and camera type versus other possible combinations that caused the failed test case. For example, the cause indicator could be a probability or percentage.
In one or more embodiments, the computer readable medium 1312 also includes a testing application 1364 for testing a complex system according to the test suite and generating an indication of a failed test case. Testing could include testing the actual complex system or a computer-simulated environment modeling the complex system. The complex system is also referred to as a tested system or simply system. For simplicity, these terms to describe the complex system are used interchangeably herein. However, one of ordinary skill in the art will appreciate that a test system in one or more embodiments could be a model (e.g., a computer-simulated model) of the complex system.
In one or more embodiments, one or more applications stored on computer-readable medium 1312 are implemented in software (e.g., computer-readable and/or computer-executable instructions) stored in computer-readable medium 1312 and accessible by processor 1314 for execution of the instructions. The applications can be written using one or more programming languages, assembly languages, scripting languages, etc. The one or more application can be integrated with other analytic tools. As an example, test suite application 1360 and cause indicators 1362 are integrated data analytics software application and/or software architecture such as that offered by SAS Institute Inc. of Cary, N.C., USA. Merely for illustration, the applications are implemented using or integrated with one or more SAS software tools such as IMP®, Base SAS, SAS® Enterprise Miner™, SAS/STAT®, SAS® High Performance Analytics Server, SAS® Visual Data Mining and Machine Learning, SAS® LASR™ SAS® In-Database Products, SAS® Scalable Performance Data Engine, SAS® Cloud Analytic Services, SAS/OR®, SAS/ETS®, SAS® Inventory Optimization, SAS® Inventory Optimization Workbench, SAS® Visual Analytics, SAS® Viya™, SAS In-Memory Statistics for Hadoop®, SAS® Forecast Server, and SAS/IML® all of which are developed and provided by SAS Institute Inc. of Cary, N.C., USA.
One or more applications stored on computer-readable medium 1312 can be implemented as a Web application. For example, an application can be configured to receive hypertext transport protocol (HTTP) responses and to send HTTP requests. The HTTP responses may include web pages such as hypertext markup language (HTML) documents and linked objects generated in response to the HTTP requests. Each web page may be identified by a uniform resource locator (URL) that includes the location or address of the computing device that contains the resource to be accessed in addition to the location of the resource on that computing device. The type of file or resource depends on the Internet application protocol such as the file transfer protocol, HTTP, H.323, etc. The file accessed may be a simple text file, an image file, an audio file, a video file, an executable, a common gateway interface application, a Java applet, an extensible markup language (XML) file, or any other type of file supported by HTTP.
In one or more embodiments, fewer, different, and additional components can be incorporated into computing device 1302. For instance, in one or more embodiments, computing device 1302 further includes an input interface 1308. Processor 1314 operably couples with components of computing device 1302 (e.g., input interface 1308, with output interface 1310 and with computer readable medium 1312) to receive, to send, and to process information.
In one or more embodiments, the computing device 1302 receives information from input device 1304 via input interface 1308. In one or more embodiments, the input device 1304 is one or more devices for user entry (e.g. input weights 1354 and failure indication 1352) into the system 1300. For instance the input device 1304 could include one or more of a mouse 1340 or a keyboard 1342. Alternatively or additionally the input device 1304 includes a display, a track ball, a keypad, one or more buttons, a sensor, a phone, etc. For instance, one or more test engineers tests a physical complex system according to the test suite and inputs failure indication 1352 of a failed test case to the computing device 1302 (e.g., using mouse 1340 or keyboard 1342). In one or more embodiments, failure indication 1352 is multiple failure indications. In the same or different example, one or more test engineers inputs weights for the complex system (e.g., weights representing how likely it is that a given option for a given categorical factor will cause a failure of the complex system).
Alternatively, or additionally, input interface 1308 further provides an interface for receiving information from another device or machine such as a computing system 1344. For instance, in one or more embodiments, another computing system 1344 tests the complex system or a computer simulation of the complex system and generates one or more failure indications of a failed test case and sends failure indication 1352 to computing device 1302.
The computing device 1302 outputs information to output device 1306 via output interface 1310. Output interface 1310 provides an interface for outputting information (e.g., information representing a most likely potential cause 1350) for review by a user and/or for use by another application or device or multiple applications or devices (e.g., a display, printer, data storage).
In an alternative embodiment, the same interface supports both input interface 1308 and output interface 1310. For example, a touch screen provides a mechanism for user input and for presentation of output to the user. Alternatively, the input interface 1308 has more than one input interface that uses the same or different interface technology. Alternatively or additionally, the output interface 1310 has more than one output interface that uses the same or different interface technology.
In one or more embodiments, the computing device 1302 generates one or more input weights 1354 or one or more failure indications 1352. For instance, the testing of the complex system via the testing application 1364 can result in a failure indication 1352 or can be used to generate one or more input weights 1354. Alternatively or additionally, the computing device 1302 can generate default weights of the input weights 1354 (e.g., if an input weight is not received for a particular option via input interface 1308).
In one or more embodiments, the system 1300 implements a method as described herein (e.g., a method shown in
The method 1400 illustrates an operation 1410, in which a test suite is generated. In one or more embodiments, the test suite provides test cases for testing a system that has different components. The different components can be represented or referred to as categorical factors for the system. In one or more embodiments herein categorical factors are referred to as factors or inputs interchangeably. Different options for a component can be represented by different levels assigned to the categorical factor. In one or more embodiments, each element of a test case of the test suite is a test condition for testing one of categorical factors for the system. Each of the categorical factors represents one of the different components. A test condition in the test suite comprises one of different levels representing different options assigned to a categorical factor for the system. Typically, a categorical factor has different levels that represent discrete values or options for a component of the complex system in contrast to a continuous variable that describes a range of possible values or options. However, in one or more embodiments, levels of a categorical factor for the system are equivalence classes derived from continuous variables. Discrete values from a particular partitioned range of the possible values for the continuous variable are considered “equivalent” testing for other values in the range. For example, if the system is a software system, equivalence partitioning can be applied to input or output data of the software system to derive categorical factors for the software system.
The method 1400 illustrates an operation 1420 for receiving information. A set of input weights for one or more levels of the test suite is received. For example, a user in one or more embodiments provides the set of input weights. A failure indication is received. The failure indication indicates a test conducted according to the test cases failed.
The method 1400 illustrates an operation 1430 for determining cause indicators. In one or more embodiments, the cause indicators are determined in response to receiving the failure indication. The determined cause indicators are based on the set of input weights and any commonalities between test conditions of any failed test cases of the test suite that resulted in a respective failed test outcome. For example, in one or more embodiments, a same test condition is involved in multiple failed test cases. Each cause indicator represents a likelihood that a test condition or combination of test conditions of the any failed test cases caused the respective failed test outcome. In the same or different embodiments, cause indicators in a test case that received a failure indication are eliminated if they did not cause a failure in a different test case.
The method 1400 illustrates an operation 1440 for identifying, based on comparing the cause indicators, a most likely potential cause for a potential failure of the system. The method 1400 illustrates an operation 1450 in which an indication of the most likely potential cause for a potential failure of the system is output. In one or more embodiments, the indication is part of an ordered ranking of potential causes for a potential failure of the system. In one or more embodiments, one or more of the potential causes precipitated an actual failure of the tested system or a model of the test system. In one or more embodiments, this indication is used for further testing, which may cause the potential failure, and/or for refraining from design a complex system, which may cause the potential failure.
One reason covering arrays are attractive for this purpose is that they are an efficient way to construct test cases that are effective at precipitating failures that are due to the combination of several inputs. Testing complex systems is a difficult undertaking, where test engineers are tasked with constructing a set of test cases that can efficiently identify faults in the system. The engineers are usually expected to work within tight time, computational and resource constraints. Covering arrays ensure that all possible combinations among t categorical factors or inputs appear in a test suite, where t is known as the strength of the covering array.
Alternatively or additionally, a test suite is another type of array. For example, the array in one or more embodiments is an orthogonal array in which each possible combination occurs the same amount of times (e.g., only once). In one or more embodiments, the array is a randomly generated array with random combinations. In one or more embodiments, the array is generated based on previous failed test cases. In one or more embodiments, a test suite can be referred to as an array D with n rows and l columns. Let column i have si levels for i=1, . . . , l. D is said to be a covering array of strength t if any subset oft columns has the property that all Π si level combinations occur at least once.
Covering arrays have been shown to be effective as a means of identifying suites of test cases for testing complex engineered systems. A software system is an example of such a complex engineered system. In one or more embodiments, each row of the array represents one of the test cases and each column of the array represents one of the categorical factors. For instance, as shown in
Given a test suite derived from a covering array, if all the test cases result in a pass (e.g., produce a success indication and/or the expected result), then the test engineer can ascertain that there are no faults due to combinations of inputs involving t or fewer inputs. However, if there are failures, the test engineer is faced with the task of identifying the particular inputs and their level combinations that precipitated the failures. For instance, there is empirical evidence that nearly all faults are due to some combination of values for five or fewer inputs.1 For many systems, devising a test suite for the system is only the beginning. If a failure is discovered, the test engineer typically wants to know which combination of inputs and associated levels induced the failure. This is known as the fault location problem. This problem can also be referred to as a failure localization problem. The number of combinations makes it difficult or even infeasible to determine a list of potential causes without using a computing approach. 1D. R. Kuhn, D. R. Wallace, and A. M. Gallo, “Software fault interactions and implications for software testing,” IEEE Transactions on Software Engineering, vol. 30, no. 6, pp. 4 I 8-42 I, June 2004.
There are two categories of computing approaches to the fault location problem: exact approaches and stochastic (or probabilistic) approaches. In the simplest form in an exact approach, for combinations involving f inputs, the exact computing method starts with all input combinations involved in test cases that induce a failure, and removes input combinations that appear in the test suite for test cases in which a failure did not occur. The value of f investigated is the smallest value for which the set of potential causes is non-empty. A stochastic computing approach instead uses modeling techniques and statistics to investigate and determine a set of potential causes.
One or more embodiments described herein are an improvement on these traditional computing approaches to the fault location problem. Further this improved computing approach allows modification of the computing based on a test engineer's knowledge of the system. For example, a test engineer typically has prior knowledge of the system, through particular inputs, or combinations of inputs, that may have been problematic in the past when testing a system, or with recent changes to the complex system that may not have been rigorously tested. One or more embodiments provides a method for specifying prior knowledge of inputs and levels in a computing approach to the analysis of a test suite when failures are discovered. This analysis provides a ranking of potential causes for investigating potential causes.
Based on previous testing, level A of factor e is known to be problematic, as is level B of input a and level B of factor d, but to a lesser extent. The test engineer assigns the highest weight of 3 to level A of factor e, and higher weights of 2 to level B of input a and level B of factor d. The remaining weights are set at the baseline value of 1.
In this example shown in
w
i
i
(j1,j2)=wi
where j1∈1, . . . , si
Equation (1) does not preclude the test engineer from changing the value of wi
Cause indicators in Cause Indicator column 1666 can be compared (e.g., to provide an ordered ranking). Alternatively or additionally a normalized weight is computed as shown in Normalized Weight column 1668 for a cause indicator in order to provide an ordered ranking. As shown in
Based on this analysis, instead of treating all combinations equally likely to have caused the failure, the most likely candidate based on either the Cause Indicator column 1666 or Normalized Weight column 1668 is combination 1670 followed by combination 1672, followed by combination 1674. The remaining potential causes are equally likely. This allows the test engineer to focus on combination 1670 and combination 1672 (e.g., by further testing of combination 1670 and combination 1672) before needing to consider any other combinations (e.g., further testing of combination 1674).
The example shown in
In the simplest case, assume that test case m results in a failure and all potential causes in ψk(m) only occur in test case m. Denoting Ci
Simply put, if there is a single failure, the likelihood of a particular combination being the cause is the weight for that combination divided by the sum of the weights for all combinations in ψk(m). There are two underlying assumptions in using Equation (2): i) that the failure is not caused by a combination of greater than k inputs, and ii) if it is due to a combination of k inputs, there is only one cause for the failure in test case m.
The example presented in
Analyzing the failures, for each test case there is a different list of potential causes as shown in
When the results of a test suite have more than one test case with a failure, and a potential cause occurs in multiple test cases, additional care is taken in determining the relative likelihood a test condition will cause a failure (e.g., considering any commonalities between test conditions of failed test cases). Intuitively, if there are multiple failures and there are potential causes that are common to the multiple test cases, it is more likely that the failures are due to the common cause than distinct causes from the sets of potential causes for each failure. To account for this in one or more embodiments, a joint probability mass function, treating the test cases in which failures occur as random variables, is used to determine a value assigned to a cause indicator.
In general, a set of potential causes involving k inputs given a failure in test case m is denoted by ψk(m). In the case of two failures as shown in
If failures have been observed for two test cases m1 and m2, and failures are due to two-input combinations, there are two different possibilities to consider to determine the cause indicators (e.g., via a joint probability mass function):
In
In one or more embodiments, the cause indicators for individual test cases are computed to form a basis for comparison for outputting a most likely potential cause for a potential failure of the system. To find the most likely potential causes for further investigation, the computed probabilities account for the input weights 1700 and commonalities between test conditions of failed test cases (i.e. combination 1770).
To provide a method for comparison, a probability is defined that one would expect to see a failure for any given combination, prior to knowing that any failure has occurred. As an example, the probability is computed as shown in Equation 3:
Assuming that a two-input combination caused the failure, to find the probability of observing failures for m1 and m2, one considers the combination of configurations in ψk that would lead to failures for m1 and m2. The probability of observing failures for m1 and m2, P[ψk], can be calculated as the probability that a single combination caused the failure, P[single cause], plus the probability that there are two independent failures, P[two causes]. For each single cause, the probability that it causes a failure can be calculated from Equation (3). Then the probability that a single cause leads to the failures is simply
P[single cause]=ΣCi
In the case of two different causes leading to failures on separate test cases, these occur independently of one another, so to calculate the probability of a particular combination of two different causes requires multiplying the two individual probabilities. Taking the summation of those different combinations,
P[two causes]=ΣθP[Ci
where
θ={Ci
P[ψk]=P[single cause]+P[two causes]. (6)
For analysis, the probability that a given combination or set of combinations precipitated the failures is determined, given the set of potential causes in which there were observed failures in the particular test cases that generated the set of potential causes. For a potential cause that is common to the two test cases, the probability by Bayes' Theorem that the combination precipitated the failure is then
since P[ψk|Ci
Ci
Ci
Through (7) and (8), there is a joint probability mass function for the two failures. If a test engineer wants to focus on the cause for a particular test case, the marginal probability can be calculated by taking the sum of the probabilities in Equation (8) involving the possible combinations from the other test case where a failure occurred.
As an example, the calculation for the cause indicators for combination 1770 and combination 1780 in
P[single cause]=P[c(B),d(A)]=1/146=146/21316,
while for the remaining potential causes, using Equation (5),
P[two causes]=208/21316.
Combining these, from Equation (6),
P[ψk]=P[single cause]+P[two causes]=354/21316.
Each of the cause indicators in
Each of the other combinations in
P[a(A),c(B)]∪P[a(B),d(A)]=2/146*2/146=4/21316
The conditional probability then that combination 1780 and combination 1790 caused a failure is then computed using Equation (8):
Each of the remaining probabilities of a failure involving combination 1780 and each of the other independent causes of failure in
P[a(A),c(B)]]=4/354+12/354+4/354+4/354+6/354+2/354=32/354
As demonstrated in
From a practical standpoint, it is often preferable to present potential causes from multiples test cases separately from the unique occurrences. That is, do a weighted comparison of the potential causes appearing in multiple test cases first. If it is discovered that none of the multiple-test case potential causes are a cause, then the test engineer can examine a weighted analysis on the remaining potential causes unique to each test case. If there are more than two failures, this approach is more tractable than generating the joint probability mass function for the multiple failure test cases.
As shown in
In one or more embodiments, a component models an operation of the software program (e.g., modeling the operation of an operating system 1802). A computing device (e.g., computing device 1302) is used to test each of the test cases of the test suite by executing the software program on computer hardware using respective test conditions of the respective test case (e.g., executing a Windows® or Linux® operating system). In one or more embodiments, a failure indication indicates that executing the software program using the respective test conditions did not operate in accordance with a set of predefined behaviors for the software program. For instance, it may take beyond a predefined threshold for processing time for the operating system 1802 to manage a particular resource of the complex system 1800 or the operating system 1802 fails entirely in an operational task. In other examples, the output from a software program may not produce expected outcomes (e.g., a statistical program using RAM 1806 may produce an unexpected model or mathematical numbers).
In one or more embodiments, the complex system 1800 is a computer-simulated model that represents behavior of computer hardware (e.g., keyboard 1810) for testing a computer-simulated environment. In one or more embodiments, the computer-simulated model is a deterministic model in which the output of the model is fully determined by the parameter values and initial conditions. Other computer-simulated models include stochastic models in which a same set of parameter values and initial conditions will lead to an ensemble of different outputs. Alternatively or additionally, aspects of the system are physically tested. For instance, a categorical factor represents operation of a circuit of the keyboard 1810 in a physical test.
Regardless of the particular testing method performed, as shown in
As shown in
While example embodiments focused on causes due to combinations involving two inputs, the same methodology holds if all combinations due to two inputs have been ruled out. In addition, if one were to assign a prior weight to interactions of each strength, analysis could include interactions of different strengths that are potential causes. For example, if one believed that two-input combinations are four times more likely than three-input combinations to cause a failure, each wi
In one or more example embodiments that extend to three inputs, for distinct inputs i1i2, and i3 at levels j1, j2, and j3(ja ∈1, . . . , sa), respectively, the weight of the three-input combination, denoted by wi
w
i
i
i
(j1,j2,j3)=wi
assuming that Equation (1) was used for the weights of two-input combinations. As before, individual weights can be assigned to wi
1) wi
where max refers to the maximum value among the set of values, or
2) wi
where ave refers to the average of the values.
Equation (10) places greater emphasis on the prior knowledge that has been specified using Equation (1).
Weights for combinations involving more than three inputs can be defined in a similar fashion. In general for f>2 inputs, the weight of a combination is the product of the weights of the individual inputs:
w
i
i
, . . . , i
(j1,j2, . . . jf)=wi
Equations (10) and (11) can be expressed similarly, with adjustments to the number of ways to group inputs as m increases.
Embodiments herein allows the failures resulting from executing a test suite to be analyzed so that the potential causes of the failures can be ranked. One of ordinary skill in the art will appreciate that embodiments herein are useful for other applications and purposes that those expressly described herein. For example, the use of weights as a criterion for test suites can also be particularly useful when a test engineer has a testing budget that allows more test cases than a covering array of optimal size. Weights can help determine which additional test cases to add to the test suite. These additional test cases can have the beneficial effect of reducing the number of potential causes and can therefore provide a happy medium between optimal size covering arrays and error-locating arrays. Alternatively, one or more embodiments of computing weights could be used by algorithms that construct biased covering arrays or to encode prior information for a test suite prioritization technique. In addition, use of weights could instead represent costs for algorithms that construct cost-aware covering arrays.
Further, in one or more embodiments, the results of testing in accordance with the ranking can be used to augment the weights for future test cases. For example, a test engineer can learn that a combination of test conditions probably should not be weighted as problematic or should be weighted to downgrade the likelihood of it being a potential cause of failure in future test suites if testing indicates it is not a source of failure in a current test suite. In one or more embodiments, a test suite is generated in a repeating pattern (e.g., every day) and the test cases in the test suite or the weights applied can be augmented based on the performance of testing in a previous period of the repeating pattern.
Further, in one or more embodiments, the results of testing in accordance with the ranking can be used to augment the weights for future test cases. For example, a test engineer can learn that a combination of test conditions probably should not be weighted as problematic or should be weighted to downgrade the likelihood of it being a potential cause of failure in future test suites if testing indicates it is not a source of failure in a current test suite. In one or more embodiments, a test suite is generated in a repeating schedule (e.g., every day) and the test cases in the test suite or the weights applied can be augmented based on the performance of testing in a previous period of the repeating pattern.
In one or more embodiments, a graphical user interface is provided to allow users to visualize a test suite in a graphical representation and interact with key aspects of the design of the test suite (e.g., in an interactive report or viewer of the graphical user interface). This is particularly useful when there are more than three factors to test cases of a test suite. For instance, the user of the graphical user interface can interact with the graphical user interface to change levels associated with factors of the test suite and explore projections into fewer factors needed for a test suite. This allows test planners to explore and select different design options for a best design based on a visual assessment rather than solely on a table of entries or summary of statistics of test results resulting in a more complete understanding of their design. In particular, by providing a visual representation of a design, a user of the graphical user interface can identify missing pieces or gaps in a design of an experiment and rearrange levels and factors in the design to see the impact of the changes on the design in real time (e.g., in improving design efficiency or design coverage).
The system 2100 includes a computing device 2102. In one or more embodiments, the computing device 2102 is the same or different from computing devices described herein (e.g., computing device 1302). In one or more embodiments, the system 1300 includes other devices (e.g., input device 1304 described herein) for receiving information related to the design of an experiment 2104 (e.g., a test suite with test cases as described herein). The system is configured to exchange information between devices in the system (e.g., via wired and/or wireless transmission) and devices in other systems described herein (e.g., system 1300). For example, a network (not shown) can connect one or more devices of system 2100 to one or more other devices of system 2100 or system 1300. In one or more embodiments, fewer, different, and additional components than shown can be incorporated into the system 2100 (e.g., components of system 1300).
The computing device 2102 has a computer-readable medium 2112 (e.g., computer-readable medium 1312) and a processor 2114 (e.g., processor 1314). For instance, computer-readable medium 2112 comprises one or more features of computer-readable medium 1312 or is computer-readable medium 1312. Additionally, or alternatively, processor 2114 comprises one or more features of processor 1314 or is processor 1314. For instance, in one or more embodiments computer-readable medium 2112 stores instructions for execution by processor 2114. For example, computer-readable medium 2112 comprises instructions for a generation application 2160 and a manipulation application 2164.
In one or more embodiments, the generation application 2160 determines or generates a graphical representation 2106 representing the design of an experiment (e.g., test cases of experiment or results of an experiment designed according to the design of an experiment). This is particularly useful for generating a graphical representation 2106 for a design of an experiment that comprises a plurality of test cases, where each element of a test case is a test condition for testing one of factors for an experiment. A test condition comprises one of different levels representing different options for a given factor in the design (e.g., different options for components of a tested complex system described herein). For instance, as shown in
In one or more embodiments, the generation application 2160 plots data 2140 on the graphical representation 2106 corresponding to data assigned to one or more test cases of the plurality of test cases on the graphical representation. For instance, data 2140 could include the factors of the test cases, a test case number, a result of a test case, etc. such that the graphical representation provides a visual representation of information to a user of the graphical user interface 2110. In one or more embodiments, the informational value of data is derived or otherwise based on the proximity of data 2140 to one or more patterns of the graphical representation.
In one or more embodiments, the generation application 2160 generates a graphical representation 2106 that comprises a fractal-like representation. For instance, the graphical representation 2106 comprises a first pattern (e.g., first pattern 2134) and a second pattern (e.g., second pattern 2136) of a fractal sequence. For instance, a type of fractal sequence is a fractal in which a curve or geometric figure has a similar pattern recurring at progressively smaller scales. As shown in
In the example shown in
In one or more embodiments, the generation application 2160 generates other shapes or selects a shape of a plurality of shape options for a fractal sequence as described in more detail with respect to other embodiments (e.g., a grid view).
In one or more embodiments, the graphical representation 2106 comprises a plurality of axes used in a fractal sequence for respective ones of factors received, indicated, or otherwise represented in the design of experiment information 2104. For instance, an axis (e.g., axis 2132A) of the plurality of axes has two points comprising a first point (i.e. endpoint 2133A) that corresponds to a first level (e.g., level 2112A) of a respective factor (e.g., factor 2120A) and a second point (endpoint 2133B) that corresponds to a second level (e.g., level 2124A) of the respective factor. In one or more embodiments, data 2140 labeling a factor assigned or otherwise corresponding to one or more axes is plotted in proximity to an axis assigned or otherwise corresponding to a particular factor received in the design of experiment information 2104. For instance, each free endpoint terminating one of the axis of the graphical representation 2106 corresponds to a potential test case. Data 2140A corresponds to one test case and data 2140B corresponds to another test case in an example where there are multiple different test cases. The location of the plotted data 2140A and 2140B relative to a free end point of the graphical representation is selected based on the factors and levels assigned to the test case represented by the axes of the graphical representation 2106. Endpoints that do not have plotted data, for instance, correspond to a potential test case that was not selected for an experiment design.
In one or more embodiments, the graphical representation 2106 comprises a tree fractal sequence comprising a first pattern 2134 and a second pattern 2136. The first pattern 2134 is generated by selecting a main axis (e.g., main axis 2130) of the plurality of axis and joining a secondary axis (e.g., secondary axis 2132) of the plurality of axes to the main axis forming the first pattern. In this case, the first pattern 2134 combines two secondary axes 2132A and 2132B to the main axis. The first pattern 2134 comprises at least four endpoints (2133A-2133D) that are endpoints of one or more axes of the first pattern 2134.
In one or more embodiments, the graphical representation 2106 comprises a set of second patterns 2136. Each pattern of the set of the second patterns 2136 is a structure geometrically similar to the first pattern 2134 and proportionally smaller than the first pattern 2134. The structure is formed from axes of the plurality of axes representing factors 2120 of the experiment Each pattern of the set of the second patterns is joined to the first pattern at each of at least four endpoints 2133A-D of the first pattern. In one or more embodiments, data 2140 is plotted at locations according to a fractal sequence of the graphical representation 2106 as a function of factors and levels represented by the fractal sequence. For instance, as shown in
In one or more embodiments, the manipulation application 2164 manipulates the graphical representation 2106 according to one or more manipulations. For instance, a manipulation may be generated by the computing device 2102. As explained, the computing device 2102 could change the designated main axis (e.g., in response to a changed prioritization based on results of an experiment using the factors). Additionally or alternatively, a manipulation is in response to user input 2166 by a user of the graphical user interface 2110. For instance, the user input 2166 could include a manipulation of the display of the graphical representation 2106 (e.g., by selection of one of different structure options for the fractal sequence). Alternatively, or additionally, the user input 2166 could include a manipulation of data represented by the graphical 2106 (e.g., a manipulation of a value corresponding to a level for a factor). The user input 2166 could be received via one or more input devices as described herein (e.g., input device 1304).
In one or more embodiments, the computing device displays or otherwise outputs to a graphical user interface 2110 a graphical representation 2106 of the design of an experiment (e.g., test cases of experiment or results of an experiment designed according to the design of an experiment). Alternatively, as shown in
In one or more embodiments, the graphical user interface 2110 can also be referred to as an interactive graphical user interface that allows manipulation of the graphical representation 2106 (e.g., manipulation in response to changed input regarding the experiment or user input 2166 to the computing device 2102). In one or more embodiments, manipulation of an interactive graphical user interface provides advantages to a user of the interactive graphical user interface for improving the visualization of test cases or changing test conditions of test cases as will be described in more details herein.
The method 2200 includes an operation 2202, that includes receiving information representing a design of an experiment (e.g., design of experiment information 2104). The design of the experiment comprises a plurality of test cases. Each element of a test case of the design is a test condition for testing one of factors for the experiment. A test condition comprises one of different levels, the different levels representing different options for a given factor in the design of the experiment. For instance, in one or more embodiments, the information represents or indicates unique numbers assigned to each test case within the design of the experiment, the number of factors in the design and their corresponding levels, etc. Alternatively or additionally, in one or more embodiments the information represents or indicates results of an experiment according to the design of the experiment. For instance, the results could include values corresponding to a result of each of the test cases for the experiment according to the design.
The method 2200 includes an operation 2204 that includes generating a graphical representation (e.g., graphical representation 2106) based on the received information. For instance, in one or more embodiments, the graphical representation represents levels of factors of the design.
The method 2200 includes an operation 2206, that includes plotting, on the graphical representation, data (e.g., data 2140) corresponding to one or more test cases of the plurality of test cases of the design of the experiment. For instance, the data comprises factor information (e.g., received in the design of experiment information 2104). Additionally or alternatively, the data comprises a value representing a test case or results of an experiment conducted according to the design.
The method 2200 includes an operation 2208, that includes displaying an interactive graphical user interface (e.g., graphical user interface 2110) comprising the graphical representation. The interactive graphical user interface allows manipulation of a representation of a factor of the experiment in the graphical representation.
Optionally, the method 2200 includes an operation 2210, that includes manipulating the representation of a factor of the experiment in the graphical representation of the interactive graphical user interface. For instance, in one or more embodiments, the graphical representation is manipulated by changing a prioritization of the plurality of axes in the fractal sequence (e.g., for a user of the graphical user interface to observe plotted data more relevant to an experiment outcome).
The example interactive graphical user interface 2300 displays a report involving mixing a catalyst into a chemical. In other words, the experiment tests a complex system of a chemical solution involving a catalyst and a chemical where different mechanisms of combining the catalyst and chemical solution provide different results for the resulting chemical solution. A first factor feed rate 2322 is related to how fast the catalyst is fed into the chemical. A second factor catalyst 2323 is related to the type of catalyst used. A third factor stir rate 2324 is related to how fast or how long the catalyst is stirred in the chemical once added. A fourth factor temperature 2325 is related to the temperature of the chemical when the catalyst is added or a reaction temperature. A fifth factor concentration 2326 is related to amount of catalyst per chemical solution. The scenario of the chemical solution is merely an example. The interactive graphical user interface 2300 could be used for the design of any experiment (e.g., for testing a complex system described herein).
The first pattern of the graphical representation 2312 has axes corresponding to feed rate 2322 (horizontal axis) and catalyst 2322 (vertical axes). The set of second patterns have axes corresponding to stir rate 2324 (horizontal axes) and temperature 2325 (vertical axes). The number of factors is an odd number, so the fractal sequence is terminated by axes corresponding to concentration 2326 (horizontal axes).
Each of the other columns in the data control 2320 denote a level assigned to the factor for each of the test cases. The levels are represented by a “1” and a “−1”. These are merely symbolic representations of two options for a factor. For example, in the context of temperature 2325 factor, a “−1” could represent or otherwise indicate a temperature of 90 degrees and a “1” could represent or otherwise indicate a temperature of 100 degrees. The actual value assigned to a level could instead be displayed in the data control 2320 or other symbols could be used to represent different options (e.g., a “+” and a “−”). A value (e.g., either −1 or 1) is assigned to each test condition of a given test case of the test cases displayed in data control 2320.
In one or more embodiments, a computing device (e.g., computing device 2102) displays interactive graphical user interface (graphical user interface) by displaying the graphical representation 2312 in proximity to a data control 2320 displaying data values assigned to each test condition of the design of the experiment. The graphical representation displays plotted values corresponding to the values assigned to each test case of the design of the experiment in the run column 2321 (i.e., runs 1-12). In the tree view shown, up is positive for the vertical branches and right is positive for the horizontal branches. Following run 1 from the main axis corresponding to feed rate, the number “1” is displayed in proximity to almost the furthest endpoint up and to the right. However, the last factor concentration 2326 is given a “−1” so the “1” is plotted just to the left side of the branch. Other orientations for a tree view could be used. For example, the main axis could be vertical rather than horizontal. Alternatively or additionally, up could be negative and/or right could be negative.
In one or more embodiments, displaying the runs graphically in the viewer 2310 enables a designer of an experiment to consider their coverage of various factors. For example, as shown in the graphical representation 2312, there are no runs plotted in the pattern in the left bottom corner. This may encourage a designer to consider having test cases that would depict a run in that pattern in the left bottom corner.
In one or more embodiments, the graphical representation and matrix of the data control 2320 are interdependent such that changing a value displayed in the matrix changes a display of a plotted value in the graphical representation in the interactive graphical user interface, or changing the display of a plotted value in the graphical representation in the interactive graphical user interface changes a value displayed in the matrix. For instance, the numbers displayed in the matrix could be “buttons” that toggle the value assigned when clicked by a user of the graphical user interface 2300 (e.g., with a mouse 1340). Clicking on a button would then change the location of the plotted run to the opposite side of the graph along the axis determined by the column where that button was located. For instance, toggling data 2350, which currently has a value of “1”, assigned to catalyst 2323 in run 12 to “−1” would move the data 2316L representing plotted run “12” to the opposite side of the catalyst axis to a region 2314. As another example, toggling the “1” for feed rate 2322 of run 7 would move the plotted “7” to the region 2314. If both changes were made, a value of “7” and “12” would both be displayed in the region 2314. Alternatively or additionally, a value displayed can be pulled to a different region of a graphical representation (e.g., using a mouse 1340), and the corresponding value displayed in the data control 2320 will change.
One of ordinary skill in the art will appreciate other ways to augment the values displayed in the data control 2320. For example, the “buttons” could instead be text boxes for typing a different level value. This would be more useful in situations where there are more than two possible levels for a given factor (e.g., in other examples described herein with multiple possible levels for a given factor). Further, the run values could be text boxes for assigning a run. Alternatively, the numbers in the run 2321 column are static and the user could move test cases corresponding to a displayed run to a different row assigned to a different run to change the value assigned to that test case (e.g., using a mouse or arrow buttons).
In one or more embodiments, the interactive graphical user interface 2300 displays a factor control 2330 with each of the factors for the experiment. In one or more embodiments, a computing device (e.g., computing device 2102) manipulates the graphical representation by reducing or increasing the factors displayed in the graphical representation. For example, the computing device could receive user input (e.g., by checking a box associated with factors displayed in the factor control 2330).
In this case all the factors are checked so all the factors are displayed in the graphical representation 2312. In other examples, one or more boxes are unchecked which allows the user to project across the corresponding factors to a design in fewer factors. This would be of special interest in a screening design or application in which key factors are identified that impacted or affected the outcome or response in an experiment. In a screening application, it is expected that not all factors are significant, and so the graphical user interface 2300 allows a user to assess a design of an experiment and the significance of a factor or certain subsets of the factors in the experiment
Other methods could be used to indicate the display of a factor. For instance, a drop down box could be displayed in proximity to a factor with options to set a color for axes associated with a given factor or to exclude the factor from display.
In one or more embodiments, a computing device (e.g., computing device 2102) manipulates the graphical representation based on a changed prioritization of factors in the graphical representation. This could be accomplished by the user manipulating the factor control 2330, e.g., by selecting one or more factors in a plot hierarchy 2332 and using up and down arrows 2324 to move the factors in the plot hierarchy 2332. For instance, as shown feed rate 2322 is at the top of the list in the factor control, so it is depicted in the graphical representation 2312 as the main axis. In this case, the main axis is displayed as a central horizontal axis. Other techniques could be used to further emphasize a main axis (e.g., a pattern or color assigned to the main axis). Catalyst 2323 is next highest in the plot hierarchy 2332, so it is depicted in the graphical representation 2312 as a secondary axis of the first pattern of the fractal sequence. This can be useful for assessing how the experimental runs are distributed in different configurations of the current design without projecting across less factors or otherwise changing the current design.
In one or more embodiments, the computing device changes a prioritization of factors in the graphical representation by replacing the main axis with a different axis of the plurality of axes. For instance, if catalyst 2323 and feed rate 2322 where switched by the user in the plot hierarchy 2332, the main axis would represent catalyst 2323 and the secondary axes of the first pattern would represent feed rate 2322.
Alternatively or additionally, the computing device changes the prioritization or sets a default prioritization based on the values in the data control 2320. For instance, the computing device could arrange or rearrange the graphical representation to better display coverage holes or gaps. A coverage hole or gap would include a region of the graphical representation correlating with a possible test case that is without plotted data, which would indicate a lack of coverage of this test case in the design of an experiment.
In one or more embodiments, the interactive graphical user interface 2300 displays options for changing the design of a pattern of the fractal sequence. For example, view control 2340 displays two options for a view of the graphical representation. A tree view 2342 is currently selected, but in other embodiments a grid view 2344 is selected.
The method 2400 comprises an operation 2402 that includes generating a plurality of axes for respective ones of the factors used in a fractal sequence. Each of the plurality of axes has endpoints comprising a first endpoint that corresponds to a first level of a respective factor, and a second endpoint that corresponds to a second level of the respective factor. In one or more embodiments, the operation 2402 comprises determining how many factors are to be represented by the graphical representation (e.g., based on user input selecting the factors). Method 2400 is used to show a method for generating a graphical representation of a plurality of factors. One of ordinary skill in the art will appreciate that the graphical representation could be used to represent only a single factor (e.g., in which only a single axis would be generated).
Operation 2402 can be completed prior to, contemporaneous with, or after other operations in the method 2400. For instance, the method 2400 comprises an operation 2404 that includes selecting a main axis of the plurality of axes. Operation 2402 could comprise generating a single axe=is of the plurality of axes for the main axis prior to, after or contemporaneous with selecting a main axis. Other axes of the plurality of axes could be generated prior to, after, or contemporaneous with other method steps. For instance, the method 2400 comprises an operation 2406 that includes joining a secondary axis of the plurality of axes to the main axis forming the first pattern. This operation could be done prior to or contemporaneous with generating an axis corresponding to the secondary axis.
In one or more embodiments, a graphical representation (e.g., a tree fractal sequence or tree view) comprises a plurality of axes (e.g., axes 2512 and 2514). The structure of a first pattern (e.g., tree pattern 2510) comprises a first axis 2512 of the plurality of axes with a secondary axis 2514 of the plurality of axes. The secondary axis 1514 of the plurality of axes is perpendicular to the first axis 2512 of the plurality of axes and at an endpoint of the first axis 2512 of the plurality of axes. In the tree pattern 2510, there is a secondary axis 2514 at each endpoint of first axis 2512 (i.e. secondary axis 2514A and secondary axis 2514B). Each of the second axes can be considered a branch branching out from each endpoint of the main axes 2512, where each endpoint represents different levels of a factor represented by the main axis. A given branch off the main axis branches in opposite directions from the main axes (trunk) 2512 at an exposed endpoint to represent different levels of a given factor represented by the branch.
In one or more embodiments, the tree pattern 2510 is formed by generating two secondary axis (i.e. the branches) 2514. Each of the two secondary axes is perpendicular to the main axis (i.e. trunk) 2512 and comprises a respective middle point between the two endpoints of the respective one of the secondary axes 2514. The respective middle point is attached to each end point of the main axis 2512.
In other embodiments, a different graphical representation is used (e.g., a grid fractal sequence or grid view) comprises a plurality of axes. For instance, grid pattern 2500 also comprises a plurality of axes (e.g., axes 2502 and 2504). In this case the structure of grid pattern 2500 comprises perpendicular axes (i.e., axes 2502 and 2504) and a window 2506 defined by endpoints of the perpendicular axes 2502 and 2504. In this case there is a visual rectangle shown surrounding the perpendicular axes. However, in other embodiments a visual border of the window 2506 is not represented in the grid pattern 2500. Rather, it is merely the endpoints that define a rectangular edge of the window. In one or more embodiments, the window 2506 is a portion or all of a window of a viewer of a graphical user interface (e.g., viewer 2310 of graphical user interface 2300). Generating the window 2506 comprises defining a portion or all of a window of the viewer (e.g., by generating the perpendicular axes 2502 and 2504).
In one or more embodiments, the graphical representation only comprises a single axis or a single pattern. However, in other embodiments, the graphical representation is used to represent more than one or two factors.
Method 2400 comprises a step of determining whether to end the patterns of fractal sequence in an operation 2408. For instance, if there are only one, two, or three factors represented the method would proceed to operation 2412. However if there are more than three factors represented the method would proceed to operation 2410. In an operation 2410, the method comprises joining a pattern with a structure geometrically similar to the first pattern (e.g., grid pattern 2500 and tree pattern 2510) until an end of patterns of a fractal sequence. For instance, the determination of an end of a fractal sequence could be based on how many factors are selected for representation in the graphical representation.
Branch pattern 2518 and tree pattern 2510 form a fractal sequence (e.g., a tree fractal sequence) with a set of a tree pattern (with only one first pattern) and a set of branch patterns (with four second patterns). In one or more embodiments, the fractal sequence is an ordering of a plurality of sets of patterns, where each pattern of a given one of the plurality of sets of patterns is proportionally smaller than a given pattern of a previous set of patterns in the ordering and attached to one of free endpoints of the previous set of patterns in the ordering. For instance, if there was a third set of patterns in the fractal sequence added to branch patterns 2518, there could be up to 16 third patterns attached to the endpoints of the branch patterns 2518. As with any fractal sequence, theoretically there could be an infinite amount or number of patterns in the sequence that are progressively smaller. Practically, the tree view can handle designs with up to 12 factors and still provide good visual clarity to a user of the graphical user interface 2110. Most experiments do not typically involve more than 12 factors, so this visualization technique is useful in the majority of situations involving the design of an experiment
In this fractal sequence for the grid view, the grid fractal sequence comprises an ordering of a plurality of sets of patterns as with the tree view where each pattern of a given one of the plurality of sets of patterns is proportionally smaller than a given pattern of a previous set of patterns in the ordering. With respect to the grid view, each pattern of the given one is within a quadrant formed by respective perpendicular axes in a respective window of a previous set of patterns in the ordering.
In one or more embodiments, the method 2400 optionally comprises an operation 2412 that includes attaching any remaining axis corresponding to a factor of the design of the experiment to the fractal sequence (e.g., in cases where there are an odd number of factors). In one or more embodiments, attaching any remaining axes comprises determining if there are any remaining axes (e.g., by determining if there are an odd or even number of factors for representation in a graphical representation).
One example of a tree view with an odd number of factors is shown in
Another example is a grid view with an odd number of factors as shown in
In one or more embodiments, the method 2400 optionally comprises an operation 2410 that includes determining whether to extend any axes 2414. This is useful in embodiments in which the factors potentially have different numbers of levels assigned to each factor. For instance, a given factor may have more than two levels (or some other minimum number of levels assigned to a factor). In a scenario in which it is determined that an extension is useful to represent more levels, in an operation 2416, there is an extension of a selected axis.
In
In
In
One of ordinary skill in the art will appreciate a computing device may generate graphical representations with an extended self-similar fractal structures with operations completed in different orders than described in the method 2400 (e.g., for computational efficiency). For instance, the graphical representation in
In one or more embodiments, the graphical representation represents data correlated with the design of the experiment. For instance in one or more embodiments, the method 2400 comprises an operation 2418 that includes populating the graphical representation with the plotted data (e.g., data 2140) based on information related to the design of the experiment.
As an example, the information related to the design of the experiment could be unique numbers assigned to test cases within the design of the experiment
One or more embodiments described herein, presents a technical improvement over prior art techniques by providing a two-dimensional representation to a user of a graphical interface of a design having more than three factors and/or more than two levels assigned to the different factors.
One or more embodiments described herein present other technical improvements in specific application that will be described in more detail below. Each of these different examples below are applicable to multi-factor, multi-level applications as described herein. For simplicity examples are described with respect to multiple factor applications, where each factor has two levels.
For instance, one or more embodiments provide an improvement for determining the quality of a design. For instance, a computing device displays in proximity to the graphical representation an indication of a holistic score for the design of the experiment (e.g., a design efficiency).
In the design metrics section 2900, there are three columns: Current Design 2920, Saved Design 2922, and Original Design 2924. The Current Design 2920 column shows the metrics for the design currently being displayed. In one or more embodiments, manipulating a representation of the factor of the experiment changes the holistic score for the design. For example, a value in the Current Design 2920 column will update in response to a change to the displayed design.
Two popular design metrics for assessing experimental designs are Determinant-efficiency (D-efficiency) 2932 and Average-efficiency (A-efficiency) 2934. As the user changes the factor levels in the particular run, efficiency values will change accordingly to reflect the metrics for the current design. For example, as shown in
D-Efficiency is the efficiency of the design to that of an ideal or hypothetical orthogonal design in terms of the D-optimality criterion. A design is D-optimal if it minimizes the volume of the joint confidence region for the vector of regression coefficients according to equation (12):
where X is the model matrix, n is the number of runs in the design and p is the number of terms, including the intercept, in the model.
A-Efficiency is the efficiency of the design to that of an ideal orthogonal design in terms of the A-optimality criterion. A design is A-optimal if it minimizes the sum of the variances of the regression coefficients according to equation (13):
A-efficiency=100p/(nTrace(X′X)−1)) (13)
where X is the model matrix, n is the number of runs in the design and p is the number of terms, including the intercept, in the model.
Other evaluations could be made (e.g., G-efficiency, average variance of prediction, design creation time, etc.). Design creation time is useful in environments in which it may take greater computational time to compute a given experiment outcome based on levels selected for an individual test cases.
Saved Design 2922 column represents visually a storage of the values for a design saved by the user and will update when the user of the graphical user interface 2300 indicates to do so (e.g., by clicking the Save Current Design button 2940) or the user changes the graphical representation (e.g., to project over more or less factors). The values in the Saved Design 2922 column will then change to those shown in the Current Design 2920 column. Original Design 2924 column represents visually a storage of the values for the original design used to generate the initial design and will change only for projection changes. Values in the Saved Design 2922 column and Original Design 2924 column provide the user of the graphical user interface 2300 with a standard for comparing different designs.
As shown in
In one or more embodiments, the user of the graphical user interface 2300 can project down to fewer factors (e.g., unchecking boxes shown in factor control 2330 of
In one or more embodiments, the computing device generates the experiment (e.g., a computer simulation experiment) in response to user input. For instance, the user can click the Make Design button 2942 to indicate to the computing device to generate and run an experiment.
One or more embodiments described herein present technical improvement for applications involving conducted experiments according to a design. For instance, in one or more embodiments, a computing device receives results of the experiment according to the design of the experiment. Results can be provided by user input or from a computing device or system performing the testing (e.g., if the computing device is performing the testing). This is also useful for screening to identify key factors that affect responses in an experiment.
The computing device receiving the results can populate the graphical representation with values. Each of the values corresponds to a result of each of the test cases for the experiment according to the design. An example of a viewer that provides analysis of response data is the analysis fractal by SAS Institute Inc. of Cary, N.C. The computing device displays the interactive graphical user interface of the graphical representation comprising the values.
In one or more embodiments, where the computing device receives information comprising results of the experiment according to the design of the experiment, the computing device can select a prioritization of the plurality of axes in the fractal sequence based on the results of the experiment (e.g., in response to testing result or user input). For instance,
The plot could also be useful for residual checking by investigating a few potentially significant interactions (based on the given model) that were not originally included. If one sees a pattern in the graphical representation, it can suggest the model is missing one or more terms. For example, consider the reactor data if main effects for Catalyst, Temperature, and Concentration, and the interactions Catalyst*Temperature were included in the model, but Temperature*Concentration was not
As shown in
Another way to look for patterns in data is to plot residuals where a residual is the difference between an observed result and a predicted result
One or more embodiments described herein present technical improvement for applications involving designing or otherwise diagnosing a designed experiment with a particular strength for a combinatorial test design as described herein. In particular, one or more embodiments, provide a tool to investigate particular projections of interest such as how well factors are covered and what needs to be done to make a projection of interest have full coverage.
Consider a strength 2 or 3 covering array as described herein, with a graphical representation showing four factors of seven factors (i.e., investigating projections into 4 dimensions).
One or more embodiments include a graphical user interface (e.g., graphical user interface 2300) that allows a user to manually adjust the graphical representation or data plotted by the graphical representation as described herein to fill in gaps for the particular projection. In this case runs 1, 2, 15, and 16 can be adjusted to fill in the gaps. That is, (X1, X2, X5, X7) should be (−1, 1, −1, 1), (−1, 1, 1, −1), (−1, −1, 1, 1) and (1, −1, −1, −1).
One or more embodiments herein are useful in conjunction with manipulating free cells (also known as “do not care” cells). These are cells in a strength t covering array that do not contribute to the t-Coverage of the covering array. This means that they can be changed to any value from the range of values for the factor without affecting the t-Coverage of the covering array.
One or more embodiments enable manipulation of the free cells to improve a holistic score for the design of the experiment that is a score related to a strength of a combinatorial test design.
One or more embodiments are useful for determining an optimal design for an experiment (e.g., an optimal screening design). For instance, one or more embodiments, are particularly useful in scenarios when the number of factors of an experiment or test runs is too numerous to computationally or visually determine an optimal design for an experiment given time or processing constraints, or in scenarios when there are no known preexisting designs for an experiment given a quantity of requested factors and test cases for an experiment
The system 3300 is configured to exchange information relevant to an output for a design for an experiment between devices (e.g., output device 3306) in the system (e.g., via wired and/or wireless transmission) and/or devices in other systems described herein (e.g., system 1300 and system 2100). For example, a network (not shown) can connect one or more devices of system 3300 to one or more other devices of system 3300, system 2100, or system 1300. In one or more embodiments, fewer, different, and/or additional components than shown can be incorporated into the system 3300 (e.g., components of system 2100 or system 1300).
In one or more embodiments, the computing device 3302 obtains information related to outputting a design of an experiment via an input interface 3308. In one or more embodiments, the input interface 3308 comprises one or more features of an input interface described herein or is an input interface described herein (e.g., input interface 1308). In one or more embodiments, the computing device 3302 obtains metric information (e.g., metric 3360). The metric information includes, for example, a metric indicating a quantity of test cases for an output design of an experiment. Quantity is also sometimes referred to as a number or amount (e.g., a number of test cases or amount of test cases). The information may be obtained by the computing device 3302 by receiving the information (e.g., entered from a user of the computing device 3302). A metric could be entered or selected by a user from a qualitative label that represents or indicates a metric N As for example, a user could select a qualitative label of a high number of runs to represent a metric of 300 or more test cases. In one or more embodiments, the system 3300 includes (e.g., input or output devices described herein) for receiving information related to the design of an experiment and providing it to the computing device 3302 (e.g., sending or transmitting over input interface 3308). Alternatively, the information is obtained from one or more default values stored in the computing device 3302 (e.g., in computer readable medium 3312).
In one or more embodiments, the computing device 3302 obtains other information relating to a design of an experiment. For instance, the computing device may receive input 3366 indicating multiple factors for the output design. For instance, the input 3366 could be metric information indicating a number of factors (e.g., factors 3320) and options for those factors (e.g., options 3322). Additionally or alternatively, input 3366 could be labels for the particular factors and labels for the options associated with the experiment. For instance, if the experiment were an experiment for a chemical process, the input 3366 could be labels received from a user for factors relevant to the chemical process (e.g., catalysts used, temperature of the solution, stir rate, etc.). The option labels assigned to each factor could be discrete (e.g., different catalyst types) or continuous (e.g., a range of possible temperatures or stir rates). Alternatively, or additionally, the computing device 3302 could have a default number of factors or options (e.g., a default option of 2 for factors that do not specify particular options).
In one or more embodiments, the computing device 3302 obtains one or more criterion for scoring or otherwise evaluating the output design. For instance, the computing device 3302 obtains a primary criterion 3362 and a different secondary criterion 3364 (e.g., criteria that an optimal design would satisfy). The one or more criterion may be input by the user of the computing device 3302 (e.g., by selecting amongst different options for the design). In one or more embodiments, the computing device 3302 displays a graphical user interface (e.g., on an output device 3306) indicating multiple criterion for the primary criterion and/or the secondary criterion for user selection. The computing device 3302 receives user input indicating a selected one or multiple ones of each of a primary criterion or secondary criterion. Additionally or alternatively, the one or more criteria are default values for determining an optimal design.
In one or more embodiments, the computing device 3302 is useful for determining an optimal screening design for an experiment A screening design is used for identifying one or more active factors of the experiment each of which independently effect an outcome of a given test case of the experiment. A common goal of screening designs is to screen through only those inputs that have the greatest effect on the output. Based on the principle of sparsity of effects, a system is usually dominated by only a few factors or low level interactions between factors. Other potential effects, such as interactions between factors and curvature effects, are typically of less importance at the screening design stage.
In one or more embodiments, the primary criterion is used as a measure of design quality for the effectiveness of a screening design. For instance, the primary criterion is related to scoring an efficiency of a screening design at identifying the one or more active factors of an experiment each of which independently effect an outcome of a given test case of the experiment.
In one or more embodiments, the designer of the experiment has additional goals beyond determining an optimal screening design for an experiment. For instance, there may be multiple optimal screening designs and a user of the computing device 3302 can consider other criteria for the screening design different from the primary criterion.
In one or more embodiments, the computing device 3302 has a computer-readable medium 3312 (e.g., computer-readable medium 2112 or 1312) and a processor 3314 (e.g., processor 2114 and/or processor 1314). For instance, computer-readable medium 3312 comprises one or more features of one or more computer-readable mediums described herein or is one of computer-readable mediums described herein (e.g., computer-readable medium 1312 or computer-readable medium 2112). Additionally, or alternatively, processor 3314 comprises one or more features of one or more processors described herein or is one of processors described herein (e.g., processor 1314 or processor 2114). For instance, in one or more embodiments, computer-readable medium 3312 stores instructions for execution by processor 3314. For example, computer-readable medium 3312 comprises instructions for determining a design of an experiment, evaluating or scoring a design of an experiment, and generating an updated design of an experiment.
In one or more embodiments, the initial screening design application 3340 determines whether there are stored instructions (e.g., stored in a computer-readable medium described herein) for generating an initial screening design for the experiment. The initial screening design is for identifying one or more active factors of the experiment each of which independently affect an outcome of a given test case of the experiment. Another words, in one or more embodiments, the initial screening design application 3340 determines whether there is a preexisting stored design (e.g., a known or preexisting design generated from stored instructions) for an initial screening design for the experiment. The preexisting stored design can preexist or is known before any particular obtained input or metrics for an experiment. For instance, the application may have a search or query function for searching, based on possible quantities of factors and test runs for an experiment, stored design instructions for constructing a known or preexisting design. Alternatively or additionally the stored instructions could be related to fetching stored design data or tables correlated with possible quantities of factors and test runs for an experiment (e.g. simple or commonly selected designs may be stored in an electronic library of complete designs). Responsive to determining that there are stored instructions for generating the initial screening design for the experiment, the initial screening design application 3340 selects, using the stored instructions, the initial screening design for the experiment.
In some cases, the initial screening design application 3340 can use stored instructions for an initial screening design to directly construct an output design. For instance, the initial screening design application 3340 determines whether there are stored instructions for generating a test suite correlated with one or more of a possible metric 3360, input 3366 and factors 3320. In some cases, the computing device determines that there are stored instructions and generates, using the stored instructions, an initial screening design for the experiment. In other case, the computing device determines that there are not stored instructions but determines that one or more of the stored instructions can be used to create one or more candidate designs for the experiment. In this case, the initial screening design application 3340, selects one of the candidate designs as the initial screening design. In yet other cases, the computing device determines that there are not stored instructions for the initial screening design for the experiment (e.g., there are not known designs or candidate designs for the experiment), and no selection is performed for an initial screening design for the experiment
In one or more embodiments, the computer readable medium 3312 comprises stored instructions for generating an initial screening design that is a preexisting design correlated with possible metric 3360, input 3366 and/or factors 3320. In this case, the computing device 3302 can select the initial screening design by executing stored instructions to directly generate, responsive to a given metric indicating an amount of test cases and a given number of factors, one or more candidate initial screening designs. Additionally or alternatively, the computer readable medium 3312 comprises an electronic library of initial screening device correlated with possible metric 3360, input 3366 and/or factors 3320. In this case, the computing device 3302 can select the initial screening design by executing stored instructions for retrieving one or more candidate initial screening designs from memory (e.g., stored in an electronic library).
In one or more embodiments, the initial screening design application 3340 evaluates the initial screening design (e.g., it evaluates a quality of the initial screening design or other criteria for the initial screening design). For instance, the initial screening design application 3340 computes an initial score for a primary criterion for the initial screening design. In one example, the primary criterion is related to a quality of the initial screening design and the initial score indicates an efficiency of the initial screening design at identifying one or more active factors.
In one or more embodiments, the computer readable medium comprises additional applications for refining the initial screening design (e.g., a modify screening design application 3342). The modify screening design application 3342 is shown as a separate application from the initial screening design application 3340 merely as an example. In other examples, functionality described herein for these applications are implemented by a single application or are distributed across different applications. In one or more embodiments, the computing device 3302 comprises a modify screening design application 3342 for determining whether to modify the initial screening design based on modification criteria. For instance, the modification criteria may comprise the secondary criterion in a case where the output design will satisfy or optimize both the first criterion and the secondary criterion. Alternatively or additionally, the modification criteria comprises a quantity of test cases, and indicated factors for the experiment. For instance, the selected initial design may be merely a candidate design as a starting point and modifications are need to ensure an output design has a certain quantity of test cases or factors.
In one or more embodiments, the modify screening design application 3342 responsive to determining, based on the modification criteria, to modify the initial screening design, generates an updated screening design for the initial screening design. For instance, the modify screening design application 3342 generates one or more modified screening designs that each has at least one option for a corresponding test case different from the initial screening design. The modify screening design application 3342 evaluates the one or more modified screening designs by computing a respective score for the primary criterion for each of the one or more modified screening designs. The modify screening design application 3342 determines that each of the one or more modified screening designs satisfies the primary criterion by respectively comparing the respective score for the primary criterion to the initial score for the primary criterion or a threshold score for the primary criterion. The modify screening design application 3342 computes a score for the secondary criterion for a given design of the one or more modified screening designs. The modify screening design application 3342 selects, based on the score for the secondary criterion, a given design of the one or more modified screening designs as the updated screening design.
In one or more embodiments, the modify screening design application 3342 outputs information to output device 3306 via output interface 3310. Output interface 3310 provides an interface for outputting information (e.g., related to a design of an experiment) for review by a user and/or for use by another application or device or multiple applications or devices (e.g., a display, printer, data storage). In one or more embodiments, output device 3306 is a device or is one of devices for displaying information relevant to the experiment (e.g., user selection criteria, design scores, or a design for an experiment).
In one or more embodiments, the computing device 3302 outputs an indication of the updated screening design for the output design of the experiment. For instance, the indication could be the output design itself. As an example, computing device could display, via a graphical user interface, an array for the output design. Each row of the array represents one of the test cases of the experiment and each column of the array represents one of the factors of indicated factors. Alternatively, the updated screening design indication could indicate differences or the presence of differences from a preexisting design (e.g., generated from stored instructions). For instance, the indication could indicate changes in options or a changed evaluation of a design. Output interface 3310 and output device 3306 could be one of or comprise features of output interfaces (e.g., output interface 1310) and output devices (e.g., output device 1306) described herein.
In one or more embodiments, the system 3300 implements a method as described herein (e.g., a method shown in
The method 3400 illustrates an operation 3402 for obtaining a metric N. The metric N indicates a quantity of test cases for an output design of an experiment Test cases is also sometimes referred to in the art as runs in an experiment Each element of a test case of the output design is a test condition for testing one of factors for the experiment A test condition comprises one of different options for a given factor of the experiment. In one or more embodiments, the operation comprises receiving the metric N For example, a designer of an experiment may be limited in the amount of test cases for the design. The designer can input the metric Nor select a default value for metric 1V, which may be predefined and stored.
The method 3400 illustrates an operation 3404 for obtaining input indicating a quantity p of indicated factors for the output design. Obtaining could comprise receiving a user selection of a total number of factors with a default level of options (e.g., two options for each factor of the design). Additionally or alternatively, receiving includes receiving from a user specific labels for different types of indicated factors. For example, if the design is an experiment for the operation of a computer, the different labels could include software factors (e.g., an internet browser used) or hardware factors (e.g., a RAM speed). The number of received labels indicates a quantity p of indicated factors for the output design.
The method 3400 illustrates an operation 3406 for determining whether there are stored instructions for generating an initial screening design for the experiment (e.g., a construction procedure for a preexisting design or instructions for fetching data of a preexisting design). The initial screening design is for identifying one or more active factors of the experiment each of which independently effect an outcome of a given test case of the experiment.
The preexisting designs, are for example, based on stored instructions for generating designs known to be optimal in some sense for screening designs with the same number of factors p and runs N. For instance, orthogonality is a desirable characteristic of a designed experiment. Orthogonal designs provide independent estimates of the active effects, which means that removing an effect from the model will not change the estimates of the other effects. For screening designs with every factor at two-levels, orthogonal designs are G, D, A, and I-optimal. A G-optimal design minimizes the maximum prediction variance for a given model. A D-optimal design minimizes the volume of a confidence ellipsoid about the parameter estimates. An A-optimal design minimizes the average variance of the main effects. An I-optimal design minimizes the average variance of prediction.
In developing screening experiments for two-level factors, regular fractional factorial designs, which are orthogonal and globally D-optimal (i.e. 100% D-efficient), exist if N is a power of two. In addition, non-regular D-optimal orthogonal designs can be generated for almost any N a multiple of four. However, if resource constraints dictate that N is not a multiple of four, an orthogonal design for two-level factors does not exist, although a D-optimal design may still be possible using computer instructions to directly construct an optimal design (e.g., a D-optimal design). For instance, the preexisting design may be one of multiple preexisting designs (e.g., a preexisting design constructed from a Hadamard matrix or a symmetric circulant block matrix). These types of constructions are discussed in more detail with respect to
In one or more embodiments, one or more known techniques are used to create instructions for fetching data and/or fetching instructions associated with a preexisting design.
The method 3400 illustrates an operation 3408 for, responsive to determining that there are stored instructions for generating the initial screening design for the experiment, selecting, using the stored instructions, the initial screening design for the experiment. For instance, the operation 3408 could comprise selecting, using the obtained given number of test case N and indicated factors p, one or more instructions for constructing or fetching a preexisting design. This can be considered a direct construction method. The direct construction method may be employed responsive to determining there is a preexisting design (e.g., generated from stored instructions or from an electronic library) for directly generating the initial screening design for the experiment that has N test cases for an experiment of p factors (e.g., during operation 3406). In this case, the operation 3408 could comprise selecting, using stored instructions, as the initial screening design a given one of any preexisting designs that have N test cases for an experiment of p factors, responsive to determining there are stored instructions for directly generating the initial screening design for the experiment that has N test cases for an experiment of p factors.
Alternatively, it is possible that there is no known or stored instructions for generating a design for the obtained given number of test case N and indicated factors p, but a preexisting design is selected as a starting place for determining an optimal design or a candidate construction method. One or more embodiments, using the candidate construction method improve computation time over techniques such as coordinate-exchange algorithms when no known technique exists for directly constructing a design of a given number of test case N and indicated factors p.
The candidate construction method may be employed responsive to determining there is not a preexisting design for directly generating the initial screening design for the experiment that has N test cases for an experiment of p factors. In this case, the operation 3408 could comprise selecting, using the preexisting design, as the initial screening design a candidate screening design that does not have N test cases or does not have p factors. The selection of the initial screening design may be responsive to determining there is not a preexisting design for directly generating the initial screening design for the experiment that has N test cases for an experiment of p factors.
In one example of a direct construction method, the selecting of a preexisting design comprises retrieving stored instructions for directly generating, responsive to a given metric indicating a quantity of test cases and a given quantity of factors, one or more candidate initial screening designs. The stored instructions may comprises at least one set of instructions, correlated with both a given metric indicating a quantity of test cases and a given quantity of factors, to directly generate a screening design. Alternatively or additionally, preexisting designs could be stored in the form of an electronic library.
In another example of a candidate construction method, the selecting comprises determining that the initial screening design that satisfies the metric N and the obtained input with factors p cannot be directly generated from stored instructions. The method comprises selecting one or more candidate screening designs that can be directly generated from the stored instructions.
The method 3400 illustrates an operation 3410 for obtaining a primary criterion and a secondary criterion for evaluating the output design, the secondary criterion different from the primary criterion. For instance, a primary criterion could be used to determine how optimal a design is for screening purposes (e.g., an efficiency at identifying one or more active factors each of which independently effect an outcome of a given test case of the experiment). As an example in an experiment for a baking process for an optimal cookie, a factor could be the temperature of an oven. In the experiment, regardless of the ingredients in the cookie, how long the cookie is mixed, etc., at certain temperatures the cookie will burn. Thus, the temperature of the oven could be considered an active factor in an experiment for active factors in a baking process. A screening design that efficiently determines the importance of the temperature factor would be a better design for this experiment.
There could be multiple types of primary criterion. For example, in one or more embodiments, the operation 3410 comprises displaying a graphical user interface indicating one or more options for selecting, from multiple primary criterion, the primary criterion. As explained above there are known computations for evaluating the optimality of a screening design (e.g., Alias-efficiency, D-efficiency, the G-efficiency, the A-efficiency, and the I-efficiency). Each of the multiple primary criterion could indicate one of different efficiency computations (e.g., computations for computing an Alias-efficiency, D-efficiency, G-efficiency, A-efficiency, and I-efficiency). The operation 3410 may include receiving a user selection of the primary criterion of the multiple primary criterion.
However, a designer may have other goals for the experiment which could be classified as one or more secondary criteria. Consideration of secondary criteria is particularly useful when there are multiple possible candidate designs that would satisfy the primary criterion. For instance, in a non-deterministic system, experiment outcomes for a test case could differ, resulting in error between the outcome dictated from an estimated model and the experimental outcome. The designer of the experiment may then be interested in an error rate for the test cases (i.e., test cases that had different results given the same selected options for factors in the design of the experiment). The secondary criterion could then be a secondary criterion for determining an error rate for outcomes of the experiment according to the output design.
In another example, while the primary criterion is related to the main effects or active factors that independently affect an outcome, a designer may also be interested in an interaction between multiple factors. In this case, the secondary criterion could be a secondary criterion for evaluating an effect of an interaction between multiple factors on outcomes of the experiment according to the output design. For instance, the secondary criterion is a secondary criterion related to minimizing a bias on estimates of the main effects due to multi-factor interactions (e.g., aliasing or a correlation for multiple factors). With correlating factors, there is some biasing effect on the outcome of an experiment for one factor given an option selected for another factor. The method may comprise computing respective model matrices where each represents modeled predicted effects on the outcomes of testing according to a given screening design for computations related to minimizing this bias as described in the context of specific examples below.
The method 3400 illustrates an operation 3412 for evaluating the initial screening design by determining an initial score for the primary criterion for the initial screening design. The initial score indicates an efficiency of the initial screening design at identifying the one or more active factors. For instance, in one or more embodiments, the initial score indicates an Alias-efficiency, D-efficiency, a G-efficiency, an A-efficiency, or an I-efficiency for the primary criterion compared to a respective efficiency of an orthogonal design (e.g., a score out of 100).
The method 3400 illustrates an operation 3414 for determining whether to modify the initial screening design based on modification criteria. For instance, a modification could be used to provide a design with the requested test cases described by metric N, and indicated factors p. Therefore, the modification criteria could include one or more of the metric N, and indicated factors p. Additionally or alternatively, a design could already satisfy the requested test cases described by metric N, and indicated factors p (e.g., if it was directly constructed or a candidate design modified to arrive at the requested test cases). In this case, the design can be refined by consideration of one or more secondary criteria for the design of the experiment. In this case the modification criteria comprises one or more of the secondary criteria.
The method 3400 illustrates an operation 3416 for, responsive to determining, based on the modification criteria, to modify the initial screening design, generating an updated screening design for the initial screening design. In some cases (e.g., when an initial screening design is selected that has N test cases for experiment with p factors), the modification criteria could comprises only the secondary criterion. In this case, the updated screening design is a refinement of the design to consider the secondary criterion. In other cases (e.g., when an initial screening design is selected that does not have N test cases for experiment or does not have p factors for the experiment), the modification criteria may comprise the secondary criterion, the metric N, and the indicated factors p. For instance, the modification criteria may be used to update the initial screening design to satisfy the obtained metric N and the indicated factors p. An initial design may already satisfy Nor p. In this case, the modification criterion may comprise only one of these factors. In some cases, additionally (e.g., at the same time or separately), modification criteria is used to refine the initial screening design to consider the secondary criterion. One of ordinary skill in the art will appreciate that in embodiments described herein, the secondary criterion could be multiple different secondary criteria different from the primary criterion.
As one example of generating an updated screening design, the method includes modifying an initial screening design by adjusting a given candidate screening design to have an adjusted screening design with N test cases. For instance, the adjusting could comprise generating a test case and adding the test case to the given candidate screening design. This is particular useful when a candidate screening design does not have sufficient test cases. Additionally or alternatively, the adjusting could include selecting a test case from the given candidate screening design and removing the test case from the given candidate screening design. This is particular useful when a candidate screening design has too many test cases. Additionally or alternatively, a modification could include combining multiple candidate screening designs.
As another example of generating an updated screening design, the operation 3416 comprises generating multiple candidate updated screening designs by modifying one or more options of the adjusted screening design. The operation 3416 comprises computing a respective score for the primary criterion for each of the multiple candidate updated screening designs. The operation 3416 comprises selecting one of the candidate updated screening designs as the generated updated screening design based on the respective score. This is particularly useful when there might be multiple designs that satisfy the primary criterion and the secondary criterion.
The method 3400 illustrates an operation 3418 for outputting an indication of the updated screening design for the output design of the experiment. For instance, the method could comprise displaying, via a graphical user interface, an array for the output design. In one type of array, each row of the array represents one of the test cases of the experiment and each column of the array represents one of the factors of the indicated factors. Of course the orientation of the array could be changed to have test cases represented by the columns and factors represented by the rows. Alternatively or additionally, the updated screening design indication could indicate differences or the presence of differences from a preexisting design (e.g., a changed option for a factor or changed score for the design).
One of ordinary skill in the art will appreciate that the method 3400 could be conducted in different orders then shown in method 3400. For instance, the obtained information in operations 3402, 3404, and 3410 could be obtained in any order or simultaneously. Further, multiple operations could be processed in a single operation or a single operation could comprise multiple operations. For instance,
In one or more embodiments, operation 3416 comprises an operation 3420 for generating one or more modified screening designs that each has at least one option for a corresponding test case different from the initial screening design.
For example, in a case where the secondary criterion is a secondary criterion for determining a model-fitting error rate or pure-error for outcomes of the experiment according to the output design, generating one or more modified screening designs that each has at least one option different from the initial screening design could comprise generating a first modified screening design that has more test cases that are assigned the same options for each factor of a respective test case than the initial screening design. In this way there would be increased test cases that have the same options for respective factors, so that the outcome of these test cases can be observed. For example, it may be useful to measure and account for wide variation in responses with runs with the same setting. In another example, where the secondary criterion is a secondary criterion for determining a prediction error rate, it may be useful to instead generate modified screening designs that have greater diversity in test case scenarios to better explore a design space.
In one or more embodiments, operation 3416 comprises one or more operations useful for ensuring that the modified screening designs still satisfies the primary criterion despite changes. For instance, operation 3416 comprises an operation 3422 for evaluating the one or more modified screening designs by computing a respective score for the primary criterion for each of the one or more modified screening designs.
In one or more embodiments, operation 3416 comprises an operation 3424 for determining that each of the one or more modified screening designs satisfies the primary criterion by respectively comparing the respective score for the primary criterion to the initial score for the primary criterion or a threshold score for the primary criterion. This is useful for ensuring that the modified screening designs still satisfy the primary criterion despite changes (either because one or more of the modified screening designs are better than the initial screening design at satisfying the primary criterion or they satisfy, for instance, a predetermined threshold for the primary criterion).
In one or more embodiments, operation 3416 comprises one or more operations for considering a secondary criterion. For instance, operation 3416 comprises an operation 3426 for computing a score for the secondary criterion for a given design of the one or more modified screening designs. In one or more embodiments, operation 3416 comprises an operation 3428 for selecting, based on the score for the secondary criterion, a given design of the one or more modified screening designs as the updated screening design. For instance, a design can be selected that has a maximum score for the secondary criterion. Alternatively, the scores from the primary criterion and secondary criterion could be weighted or averaged to select a design that has a relatively high score for both criterion.
As shown in
As shown in
As shown in
As shown in
The drop-down menu 3550 could have other drop-down menus to specify other constraints relevant to the factor types (e.g. by clicking an arrow 3556 to display more constraints).
As shown in
As shown in
Within a given response window 3506, the user of the graphical user interface 3500 may want to set specific goals for an outcome of the experiment (or response) or the importance of that response. For instance, the user may specify for a given response a goal of maximize, match target, minimize, or none. A goal of maximize indicates that the best value is the largest possible. If there are natural lower or upper bounds, these can be specified using the lower limit or upper limit fields of response window 3506. A goal of minimize indicates that the best value is the smallest possible. If there are natural lower or upper bounds, these can be specified using the lower limit or upper limit fields of response window 3506. A goal of match target indicates that the best value is a specific target value. This can be explicitly indicated or can be assumed to be midway between the lower limit and upper limit fields of response window 3506.
The importance field of response window 3506 allows responses to be weighted (e.g., based on the importance of a goal for that response) when there are multiple responses.
The graphical user interface 3500 may have an indication (e.g., continue button 3554) for indicating that the user has specified the response and factors.
In particular, as shown in number of runs window 3606, graphical user interface 3600 may optionally allow a user to select a minimum level of runs and/or a default level of runs. A minimum level of runs may allow more additional runs (e.g., to satisfy secondary criteria). The values for the minimum level of runs or defaults may be static values or variable based on, for instance, the number of factors of the experiment Additionally or alternatively, the runs window 3606 allows a user to specify a particular number for the runs (e.g., in a run field 3610). Embodiments herein are useful even if the user selects a number of runs in which there are no preexisting designs that could satisfy the requested design (e.g., there are no stored instructions for generating a preexisting design).
Additional constraints for the design may be input. For example in an optional design generation window 3604, a user may select options to group runs, to set center points or replicate runs. For instance, a center point is useful to set a run such that for each continuous factor of the run, the value of a continuous factor of the run is set midway between the high and low settings. In an optional alias term window 3602, the user may enter alias or correlation properties between factors (e.g., a two-way or three-way interaction between specified factors).
In an optional model window 3612 the user can select constraints on the model. In an optional factor constraints window 3614, the user can select factor constraints. For instance, the user may wish to restrict values for individual factors using a disallowed combinations filter.
Once the user is satisfied with their inputs, the user can select a control to generate a design (e.g., by selecting the Make Design button 3608) and start the process of creating the optimal design that maximizes one or more criteria (e.g., a specified efficiency measure) under the constraints provided by the user (in this case, a specified number of runs in run field 3610).
The user can also optionally set a secondary criterion (e.g., a criterion related to interactions between factors). In the example shown in
Selecting the minimize aliasing option 3742 versus selecting minimize correlations option 3744 can allow a designer control over selection between different possible design options. For example, a first design could have a large number of factors uncorrelated with one another at the cost of a few factors being aliased. A second design could have some level of correlation among all the factors. If the minimize correlations option 3744 is selected, the first design may be selected as having the lowest overall correlations. However, this results in a few factors aliased and so not estimable. If instead the minimize aliasing option 3742 is selected, the second design may be selected as having the lowest aliasing. This might be desirable to spread the correlation around so that every factor has some ability to be estimated.
Selecting the None option 3740 in the graphical user interface 3700 as shown in
A=(X1TX1)−1X1TX2.
If the minimize correlations option 3744 is chosen, a subset of columns or designs is chosen to minimize the average of the absolute entries of X1TX1, to minimize the correlations between the main effects model terms.
Another possible criterion not shown in
In the example shown in
In this example, several diagnostics 3904 are provided to the user for evaluating the design. The designation “D Optimal Design” indicates that the design was constructed to maximize the D Efficiency measure (100 is best). Alternative efficiency measures are also provided for convenience (100 is best for each). In addition to the efficiency measures, the average variance of prediction is provided which is a measure of uncertainty in predicted responses based on a model estimated from the design. It is better to have a lower average variance of prediction. The design creation time is also provided. It is better to have a lower design creation time.
In a first operation 4002, the user requests an optimal design for an experiment with test cases of a quantity N and factors for each test case of a quantity p. For instance, the request can be made via a graphical user interface to input or select defaults for N and p and other information relevant to the design. If the user proceeds to make a design, the method determines in an operation 4004 whether there are stored instructions, correlated with N and p, for generating a design of an experiment. In other words, once the input has been provided, the initial inquiry is to determine whether the design can be directly constructed using known methods (e.g., in preexisting designs). If so, then it remains to determine what method to use for the given inputs in an operation 4006.
If the user input does not match any known case where a design can be directly constructed, rather than start from scratch (e.g., using a search algorithm), it may be more beneficial to use a known direct construction method to build a candidate starting design from which to generate the optimal design through additions to or manipulations of that candidate starting design, and the method proceeds to an operation 4012.
For instance, there are known design catalogs, which can be used as preexisting designs for direct construction, especially for designs not generated using one of general construction procedures below. For example, a table of D-optimal design matrices can be found at this website: http://www.indiana.edu/˜maxdet/fullPage.shtml#tableTop.
There are certain known methods, which can be used to create preexisting designs for constructing D-optimal designs for different numbers of runs in some of the following cases:
In the case of N≡0 mod 4, a D-optimal design for two level factors may be constructed by generating a Hadamard matrix. There are multiple different construction methods sufficient for generating all the Hadamard matrices up to and including N=300 for Hadamard matrices, N=p. When N>p, any subset of the columns of the Hadamard matrix of order N excluding the first column for the intercept yields an orthogonal (and globally D-optimal) design. For example, one example construction methods for Hadamard matrices is the Sylvester construction method discussed in more detail below as an example.
For the case of N≡1 mod 4, if a Hadamard matrix of order N−1 exists, only the case where N=p requires additional effort for construction. For N>p construction starts by generating a Hadamard matrix of order N−1 and adding a row of ones for simplicity. Again, any subset of columns can be removed to obtain a design with the desired number of factors.
For the case of N≡2 mod 4, for N−1>p, construction starts by generating a Hadamard matrix of order N−2 and adding two orthogonal rows (or near orthogonal if p is odd). For the case N=p, a general construction procedure can use block circulant matrices. The case of N−1=p can easily be derived from a design with N=p by deleting any row.
The case of N≡3 mod 4, is the most computationally difficult except when p(N+5)/2. For these values of p, construction starts with generating a Hadamard matrix with N+1 rows and columns. Then, removing the first row and choosing an appropriate subset of columns yields the globally D-optimal design. When p>(N+5)/2 constructions are known for special cases.
Since many of the cases are derived from a Hadamard matrix this is discussed in more detail below. Assuming the standard linear model Y=Xβ+ε, where X represents an N×p design matrix consisting of N runs with each run associated with a combination of levels in p−1 factors (p unknown parameters, including an intercept), Y represents the vector of responses that will be collected during the experiment, β represents the vector of unknown parameters to be estimated from the experimental data, and ε represents the vector of errors between the collected responses and the assumed true underlying mean model Xβ. Technically, X is associated with a particular design D and so is often depicted as X(D), but for the sake of simplifying notation, superscripts and subscripts are used instead to specify a specific design matrix.
The focus here is on generating optimal designs, which give layouts of factor levels per run that optimize a specified criterion of interest for a given total number of runs. Often, this criteria, is related to the information matrix of the design, which is computed as M=X′X, where X′ is the transpose of X. Considered as designs with two-level factors and assuming that the two levels have been coded −1 and 1 for each factor.
There are a wide variety of optimal design criteria. D-optimality is given as an example since it is the most commonly used criterion in practice. A design is D-optimal if it maximizes the determinant of M or, equivalently, minimizes the determinant of M−1. Maximizing the determinant of M minimizes the volume of the confidence region around the estimates of β for a fixed confidence level. For example, for a simple linear model (β={β0,β1}′), this would be equivalent to minimizing the 2-dimensional confidence region around {,
}. Thus, D-optimal designs minimize an omnibus measure of the uncertainty about the parameter estimates. Embodiments described herein are applicable for other optimality criteria.
As an example, of generating an initial starting design from known methods, the case where N≡0 mod 4 is discussed. The optimal form of M for this case is
Its determinant is
det(M*0)=Np.
The design matrix X*0 that achieves this form is a member of the Hadamard class of matrices, often simply referred to as Hadamard matrices. Typically, these matrices must be standardized so that the first column is all 1's (corresponding to the intercept term). The name of this class of matrices acknowledges the work of Jacques Hadamard. See e.g., Hadamard, J. (1893). Resolution d'une question relative aux determinants. Bulletin des Sciences Mathematiques 17, 240-246. This work proved that the determinant of X*0 is the largest determinant among all matrices with binary (−1 or 1) entries and so, as a corollary, showed that M*0 has the largest determinant among all M matrices created from binary matrices.
One application of construction of a design using a Hadamard matrices is the Sylvester Construction, with a construction:
The Sylvester construction can generate Hadamard matrices with run sizes that are powers of 2. Another example construction method is the Williamson construction for Hadamard matrices. Let
where A, B, C, and D are symmetric, circulant matrices of order n with entries ±1 that satisfy
XY′=YX′, for X≠Y∈{A,B,C,D}
as well as AA′+BB′+CC′+DD′=4nIn
If these conditions are satisfied, then W is a Hadamard matrix of order 4n.
Circulant matrices are matrices of the form
so they can be specified by their first row. The Williamson construction requires that the circulant matrices have the additional property of being symmetric.
The matrices A, B, C, and D are often referred to as Williamson-type “plug-in” matrices. Notice that when A=B=C=D=[1] then W=H4. As an example, let:
Note that A, B, C, and D are ±1 square matrices of order 3 and since
They also satisfy AA′+BB′+CC′+DD′=12I3. Furthermore, since
it follows that
AB′=BA′; AC′=CA′; AD′=DA′; BC′=CB′; BD′=DB′; CD′=DC′.
Since the required conditions are satisfied, we can use A, B, C, and D to construct H12 as follows:
Other preexisting designs can be defined by other construction methods not specifically described herein (e.g., other preexisting designs constructed from a Hadamard matrix or a circulant block matrix).
Operation 4006 specifies to select stored instructions for generating a design for an experiment. Any which of these construction methods could be used to specify or derive stored instructions for generating a design. However, previously these type of construction methods solely focused on generating a design that maximizes a particular measure of efficiency. That is, once a design is generated, the task is complete; there is no further consideration of additional properties of the design. This often results in undefined or incomplete steps within the construction method (e.g. add an additional run with any combination of factor level inputs). These past approaches did not contemplate refining the design for any sort of secondary criteria. In some cases, there are multiple designs that maximize a particular efficiency, and so these designs are simply catalogued with no other consideration for any further distinctions among them.
Statisticians are interested in other criteria to evaluate a design beyond a measure of efficiency. There are multiple other secondary criteria that can be considered as well. In an operation 4010, these other criteria can be considered by refining the design based on other criteria as needed. For example, if a construction method simply requires adding an additional experimental run to a known design with no specification as to what that run should look, the default might be to make the new run be different from all the other runs before it, resulting in a design that more completely explores the space of possible factor inputs into the process being investigated (e.g., to consider a prediction error rate). A user could optionally select the new run be a repeat of a previous run, which would result in a design that better estimates “pure error” (pure uncertainty within the process untainted by uncertainty introduced in varying factor inputs), which could help yield better estimates of the factor effects. One or more embodiments, improve existing direct construction methods by providing default settings and/or user options representing statistical principles that goes above and beyond the construction methods.
In the case of multiple designs that satisfy an optimal criteria, user options could provide ways to investigate secondary properties of these designs to determine which may be best suited as a default for users. For example, one design might have better predictive properties than another design (reduced uncertainty in predicted responses), but at the cost of making some effects explicitly present in the assumed underlying model completely indistinguishable from effects that, while not explicitly in the assumed model, may actually be present in the true underlying model (this is referred to as “confounding” or “aliasing”). In this case, the user may choose to go with an alternative design based on the default assumption that confounding of effects is much less desirable, while also building in user options to choose the original design should predictive properties be of greater interest. As described with respect to
In the case where there were not stored instructions for generating a design of an experiment to satisfy N and p, the flow diagram 4000 proceeds with an operation 4012 to consider whether there are candidate designs for the experiment. If there are candidate designs, an operation 4016 provides for adjusting the candidate design to achieve the request described (e.g., by methods described herein, such as adding or removing a test case from a candidate design or combining candidate designs). The choice of a starting design may impact the quality of the design further refined using coordinate-exchange. For example, in the case of N≡3 mod 4, where there is less availability in constructions guaranteeing optimality, one can quickly generate a reasonable starting design for p (N+5)/2. Similarly, for the saturated cases where a direct construction is not known, one can start by generating a design having the maximal number of columns for the unsaturated case. Then a highly D-efficient design can be obtained by adding columns using some algorithmic optimization approach.
One such algorithmic approach related to adding one or more additional columns, involves adding one element at a time to a column, and choosing the value that maximizes det(X1TX1), or adding random column(s), and using coordinate exchange on that column(s) only, rather than having to iterate over the whole design. This approach improves computational processing over approaches that use coordinate exchange across a whole design from a random starting point. In an operation 4018, the design can be refined based on other criteria described herein.
If there is not a candidate design, an operation 4014 a search algorithm is used to create the design. A search algorithm is the approach taken in other statistical software to generate a design (e.g., one or more software tools offered by SAS Institute Inc. of Cary, N.C., USA such as JMP®). Once the user provided input, a candidate design would be generated at random and then an algorithm would be used to refine the entries of this design to maximize a user-selected measure of the design's efficiency (how well the design minimizes a measure of uncertainty regarding estimates of the factor effects on the response). The user can specify how many random designs to start with. The more starts, the more likely the best design will be found, but at a cost of more computation time.
In one or more embodiments described herein, a search algorithm is now implemented as a last resort for those cases where a design cannot be directly constructed, or a constructible/known design does not exist
Here, the improvement in design diagnostics brought about by the new flow diagram 4000 are clearly displayed in Tables 2 and 3. Table 2 shows design diagnostics for existing commercially available IMP® version 14. Table 3 shows improvement in design diagnostic using methods described herein.
As shown in Table 3 compared to Table 2, the design time is significantly reduced to nearly instantaneous in some cases. (The true time is not 0; rather, it is simply so small that it has dropped below the display threshold). The time savings are particularly significant in situations such as Example 4 where the number of runs and factors is large (1059 runs and 531 factors). In all the examples, having stored instructions for generating an initial screening design improved processing time. In Example 1 (32 runs), a Hadamard matrix was used as an initial screening design because the number of runs was a multiple of 4. In Example 2 (33 runs), a Hadamard matrix was used as the initial screening design, and the initial screening design was modified to add a run to the Hadamard matrix to achieve a design with 33 runs. In Example 3 (34 runs), a Hadamard matrix was used as the initial screening design, and the initial screening design was modified to add two orthogonal runs to the Hadamard matrix. In Example 4 (1059 runs and 531 factors), a Hadamard matrix with 1060 rows was used as the initial screening design, and the initial screening design was modified to remove a row of the Hadamard matrix and reduce to 531 columns.
Further, the optimization is improved in each of the examples. It should be noted here that the D-Efficiency displayed in Table 3 for Example 3 (34 runs and 31 factors) is the highest efficiency that can be achieved for this number of runs and number of factors. Accordingly, embodiments herein are useful for reducing design time while improving measures of optimization. Further, embodiments herein are useful for optimizing designs to consider secondary criteria beyond optimization criteria.
One or more techniques described herein are useful for applications related to validating data for use in generating a model (e.g., a machine learning algorithm model). One type of model validation technique is a cross-validation technique for assessing how the results of a statistical analysis will generalize to an independent data set. For example, if the model is used to predict responses based on input for various factors, cross-validation can be used to determine how accurate the model is at predicting the responses. Cross-validation can also be referred to as rotation estimation or out-of-sample testing.
Embodiments herein are applicable to different types of validations. For example, there are many types of cross-validation. Cross-validation includes exhaustive and non-exhaustive cross-validation. Exhaustive cross-validation techniques test all possible ways to divide an original sample of data for generating a model. Non-exhaustive cross-validations do not compute all ways of splitting the original data sample. Embodiments are particularly useful in validations in which repeated validations are performed to sample different portions of the original sample. For example, with a K-fold cross validation, the original sample is partitioned into K equal sized subsamples, and a single subsample is retained as the validation data for testing the model and the remaining subsamples are used as training data for generating a test model. The validation can be repeated K times with each of the K samples used as the validation data and the results can be averaged. Practitioners may want to repeat the K-fold process several times over different validation sets, often referred to as repeated K-fold cross validation.
Current approaches to K-fold cross-validation involve selecting the data in the different K-fold validation sets by randomly selecting consecutive K-fold cross validation sets. However, this can result in a fold from one set being highly correlated with a fold from another set. Further, current software approaches may have other limitations. For instance, the software may only provide a single data set for validation or independent data sets such that there is no way to optimize different data sets for different validations.
One or more embodiments described herein control assignment of group identifiers to data elements of a data set. For example, one or more embodiments, selectively assign these group identifiers to test case of a data set by considering the data set as a whole. This can achieve an improved grouping assignment design (e.g., orthogonal or near-orthogonal groups) and an improved grouping assignment design between different validations than current approaches to validation (e.g., with better input variation between groups).
The system 4100 is configured to exchange information relevant to outputting the indication 4130 (e.g., between computing device 4102 and output device 4106) in the system (e.g., via wired and/or wireless transmission) and/or devices in other systems described herein (e.g., system 3300, system 2100, and system 1300). For example, a network (not shown) can connect one or more devices of system 4100 to one or more other devices of system 4100, system 3300, system 2100, or system 1300. In one or more embodiments, fewer, different, and/or additional components than shown can be incorporated into the system 4100 (e.g., components of system 3300, system 2100 or system 1300).
In one or more embodiments, the computing device 4102 obtains information related to outputting the indication 4130. In one or more embodiments, the input interface 4108 comprises one or more features of an input interface described herein or is an input interface described herein (e.g., input interface 1308). In one or more embodiments, the computing device 4102 obtains data 4120 for test cases for testing a system. For example, the system could be a physical environment like a city or geographic region, or a machine like an airplane or a computing system. The system could also be a computer-simulated system. A response of the system could be influenced by selection of various options for factors of the system. For example, in the case of a city, the medium housing price or air pollution in the city could be impacted by factors in the city (e.g., medium age of a home in the city, racial make-up of the city, crime rate, etc.).
In one or more embodiments, the data 4120 includes, for example, inputs 4121 each representing a respective option for each of factors in each of respective test cases. For instance, if one of the factors is crime rate, the different options for each of the test cases could be one of different rate ranges (e.g., <5%, 5-15%, >15%).
Additionally, or alternatively, the data 4120 comprises responses 4122 (e.g., a measure of air pollution or medium housing prices) of the system for each of respective test cases. The response could be a result of one or more options selected for a factor for the system or it could be influenced or determined by other factors not present or captured in the test cases. However, the factors may be selected for the test cases because a practitioner or computing system considers selected factors as useful for generating a model to predict responses. For example, based on data regarding system parameters for an airplane, and failure rates given those parameters, certain system parameters may be used to generate a model to predict future airplane failures.
In one or more embodiments, the computing system 4100 is used to collect, generate, or store the data 4120 (e.g., at computing device 4102). Alternatively, or additionally, the computing device 4102 receives the information via an input interface 4108 (e.g., from an input device described herein).
In one or more embodiments, the computing system 4100 is used to evaluate how useful data associated with selected factors is for generating a model. For example, in one or more embodiments, the computing device 4102 receives a request 4150 requesting an evaluation of the data 4120 for generating a model to predict responses based on selected factors. Additionally, or alternatively, the computing device 4102 receives evaluation information 4140 for evaluating a model (e.g., to evaluate in response to request 4150). The evaluation information 4140 may be obtained by the computing device 4102 by receiving the information (e.g., entered from a user of the computing device 4102). Alternatively, the information is obtained or generated from information stored in the computing device 4102 (e.g., in computer readable medium 4112). For instance, the request 4150 could indicate the evaluation information 4140 or information to obtain or trigger a variation of the evaluation information 4140. The request 4150 could be sent in one or more computer messages.
In one or more embodiments, the evaluation selected for the model uses different groups of test cases, and the evaluation information 4140 is related to assigning the different groups of the evaluation. For example, with cross-validation using a holdout method, data 4120 is assigned to two group's a training set for generating a model and a test set for testing the model. The results of the testing can be used by the computing device 4102 to evaluate the model. As another example, K-fold cross validation divides a data sample into k groups with one of the groups used for testing and the rest of the groups used for training. The process is repeated k times with each of the groups getting used as a test group.
In the same embodiments, or different embodiments, the evaluation selected for the model uses different validation sets and the evaluation information is related to the number of validation sets. For example, a given validation (e.g., K-fold or hold-out) could be repeated with different test cases assigned to different groups.
In one or more embodiments, the computing device 4102 obtains evaluation information 4140 by obtaining different group identifiers 4142 for each of different groups for distributing the plurality of test cases for the system. At least one group of the different groups is a test group for testing a respective test model derived from inputs of one or more training groups of remaining groups of the different groups. For example, the computing device 4102 could receive an indication of a total number of groups or total number of test cases within a group, and generate different group identifiers for the different groups. Alternatively, the computing device 4102 could generate the group identifiers from a default value for a selected or requested evaluation method (e.g., 10 groups for a 10-K-fold cross validation). Additionally, or alternatively the computing device could receive the different group identifiers (e.g., from an input device described herein).
In one or more embodiments, the computing device 4102 obtains evaluation information 4140 by obtaining a validation indication 4144 indicating one or more validations for validating a model derived from the data. For instance, a user could provide a quantity of validations. Alternatively, the computing device 4102 could obtain this information implicitly or using a default value (e.g., using a default value of a single validation set unless a user provides an alternative validation number).
In one or more embodiments, the computing device 4102 generates data sets useful for evaluating the model. In the same or different embodiments, the computing device 4102 evaluates the model (e.g., using the generated data sets).
In one or more embodiments, the computing device 4102 has a computer readable medium 4112 (e.g., computer readable medium 3312, 2112 and/or 1312) and a processor 4114 (e.g., processor 3314, processor 2114 and/or processor 1314). For instance, computer readable medium 4112 comprises one or more features of one or more computer readable mediums described herein or is one of computer readable mediums described herein (e.g., computer readable medium 3312, computer readable medium 2112, or computer readable medium). Additionally, or alternatively, processor 4114 comprises one or more features of one or more processors described herein or is one of processors described herein (e.g., processor, 3314, processor 2114, or processor 1314). For instance, in one or more embodiments, computer readable medium 4112 stores instructions for execution by processor 4114. For example, computer readable medium 4112 comprises instructions for generating data sets useful for evaluating the model, and/or instructions for evaluating the model.
In one or more embodiments, the computer readable medium 4112 comprises a data set generator application 4116 comprising stored instructions for generating a data set comprising a respective data element for each of multiple test cases. For instance, a data set is generated for evaluating data for generating a model to predict responses based on multiple factors. A data set can be generated for each validation (e.g., each validation indicated by the validation indication 4144).
Additionally, or alternatively, the computer readable medium 4112 comprises a group assignment control application 4117 comprising stored instructions for controlling assignment of a group identifier of the different group identifiers to test cases (e.g., by assigning a group identifier to each of the respective data elements).
Additionally, or alternatively, the computer readable medium 4112 comprises a model evaluator application 4118 for evaluating a model (e.g., according to a data set generated by the data set generator application 4116 and/or a group assignment according to the group assignment control application 4117).
Data set generator application 4116, group assignment control application 4117 and model evaluator application 4118 are shown for explanation as separate applications. However, computer instructions or operations associated with the applications could be configured differently and/or work together (e.g., in the same application). For example, one or more claimed embodiments are packaged as an add-in where a user can create orthogonal, or near-orthogonal, K-fold validation columns for an existing software tool (e.g., one or more SAS software tools described herein).
In one or more embodiments, the computing device 4102 outputs an indication 4130 of a generated data set for each of one or more validations. For instance, the indication 4130 could be the generated data set itself.
Additionally, or alternatively the indication 4130 could indicate a model assessment or option 4138 itself. For example, the computing device 4102 could display on output device 4106 a score for one or more validations. For example, the model evaluation indicates how much variance a model based on the data accounts for in observed responses of the system according to test cases of the plurality of test cases. For instance, a perfect model trained on training data would be able to predict correctly all the observed responses of the test data with no variance. However, a model could still be useful or valid even if there is marginal variance in the predicted responses compared to the observed responses in the test data. The model evaluation could indicate how valuable that model is by indicating a variance.
Alternatively, or additionally, the computing device 4102 could display on output device 4106 an option for improving a model. For example, if the model is a machine learning algorithm trained on one or more observed inputs and one or more observed responses, the indication 4130 could indicate an option of whether to discard or add test cases to the plurality of test cases to improve the machine learning model.
For example, in one or more embodiments, the computing device 4102 could associate a first group identifier of the different group identifiers with a test group that comprises all the test cases assigned the first group identifier. The computing device 4102 could associate at least a second group identifier of the different group identifiers with a training group that comprises all the test cases assigned the second group identifier. The computing device could then generate a first test model (e.g., generate a machine learning algorithm) based on the training group. The computing device could then generate a first test evaluation of the first test model at predicting the responses for each of respective test cases of the test group. The computing device could then generate, based on all test evaluations generated for each of the one or more validations, a model evaluation of the data for generating the model to predict responses based on the plurality of factors. The computing device could then output an assessment of the model evaluation.
In an example where the validation requires more than one training group (e.g., a k fold cross-validation, where k is more than 2), the computing device could be configured to receive user input (e.g., via input interface 4108) indicating multiple training groups; and for each of the one or more validations, generate a first test model based on the multiple training groups.
Output interface 4110 and output device 4106 could be one of or comprise features of output interfaces (e.g., output interface 3310 or 1310) and output devices (e.g., output device 3306 or 1306) described herein.
In one or more embodiments, the system 4100 implements a method as described herein (e.g., a method shown in
An operation 4201 of method 4200 comprises receiving data for a plurality of test cases for testing a system. The data comprises inputs and the inputs represent a respective option for each of a plurality of factors in each of respective test cases of the plurality of test cases. For example, the data could comprise inputs for a continuous factor (e.g., a continuous factor described herein). In this case, the inputs include values within a range of continuous values for the continuous factor. Additionally, or alternatively, the data comprises inputs for a categorical factor. The inputs could then include values representing one of discrete options for the categorical factor within a range of options (e.g., a categorical factor described herein). The data comprises a response of the system for each of respective test cases of the plurality of test cases (e.g., according to the inputs of the test cases).
An operation 4202 of method 4200 comprises receiving a request requesting an evaluation of the data for generating a model to predict responses based on the plurality of factors. An operation 4203 of method 4200 comprises obtaining different group identifiers for each of different groups for distributing the plurality of test cases for the system. For example, a computing device may receive a request by receiving a user request requesting an evaluation of a model using K-fold cross-validation and obtain different group identifiers by generating k group identifiers for each of K-folds of the K-fold cross-validation. In this embodiment or different embodiments, at least one group of the different groups is a test group for testing a respective test model derived from inputs of one or more training groups of remaining groups of the different groups.
An operation 4204 of method 4200 comprises obtaining a validation indication (e.g., indication 4130) indicating one or more validations for validating a model derived from the data. For example, it may be beneficial to have multiple validations for validating a model (e.g., with different data distributions for each of different k folds).
An operation 4205 of method 4200 comprises for each of the one or more validations: generating a data set comprising a respective data element for each of the test cases of the plurality of test cases and controlling assignment of a group identifier of the different group identifiers to each of the respective data elements. For example, the data set 4132 in
In some embodiments, there are multiple validations. In such cases, it may be useful to control assignment of different group identifiers to each test case of the plurality of test cases (e.g., test case information 4134) such that all test cases identified by a same group identifier of a given validation are distributed among as many of different groups of other validations as possible given one or more criterion for distributing the test cases of a validation according to an orthogonal design.
An operation 4206 of method 4200 comprises outputting an indication of one or more generated data sets for the one or more validations (e.g., indication 4130 as shown in
In the same or different embodiment as an embodiment for executing method 4200, the design of a given validation could be determined by one or more iterative operations.
An operation 4251 of method 4250 comprises determining an initial distribution of the test cases in the different groups by assigning each of the different group identifiers to each of the respective data elements. In an example, initial distribution a first group identifier of the different group identifiers is assigned to a data element for one or more test cases of the initial distribution. Determining an initial distribution could comprise, for instance, randomly distributing the group identifiers to each of the test cases or could be a distribution provided by a user. As another example, the method 4250 could be performed over several iterations and the initial distribution could be generated from a previous iteration of the method 4250.
An operation 4252 of method 4250 comprises evaluating the initial distribution by determining an initial score. For instance, the initial score could indicate the initial distribution compared to an orthogonal design or a previous design considered optimal in some way for distributing the test cases in the different groups.
An operation 4253 of method 4250 comprises determining an updated distribution of the test cases in the different groups by changing one or more assignments. For example, an assignment of a first group identifier to a first data element in an initial distribution could be changed to an assignment of a second group identifier of the different group identifiers to the first data element.
In one or more embodiments, the groups do not need to have the same number of test cases, therefore an assignment of a group identifier to one test case could be changed without changing an assignment of a group identifier to any other test cases. For example, a coordinate exchange approach can be used to select a value when changing the assignment. In other cases, assignment could involve swapping identifiers and/or changing multiple identifiers.
For example, it may be beneficial to change a group identifier to switch identifiers between different test cases. For instance, the initial distribution could comprises an assignment of the second group identifier of the different group identifiers to a second data element for a second test case of the initial distribution; and determining the updated distribution could comprise changing the assignment of the first group identifier to an assignment of the second group identifier by switching the assignments of the first data element and the second data element.
Operation 4254-4256 comprise operations to evaluate an updated distribution. An operation 4254 of method 4250 comprises determining an updated score. For example, the updated score could indicate a score for the updated distribution compared to the orthogonal design for distributing the test cases in the different groups. An operation 4255 of method 4250 comprises determining a first evaluation by comparing the updated score to the initial score. An operation 4256 of method 4250 comprises selecting, based on the first evaluation, the updated distribution or the initial distribution for assignment of the different group identifiers to each of the respective data elements.
In one example, the determining the initial score comprises computing |X′X| according to the initial distribution where X is a model matrix for modeling the data according to a given distribution of a validation of the one or more validations. The determining the updated score comprises computing |X′X| according to the updated distribution; and the selecting, based on the first evaluation, the updated distribution comprises selecting based on an indication of a D-efficiency for the updated distribution.
The evaluations could be based on other optimality criterion. For example, a computing device (e.g., computing device 4102) could receive a selection of a criterion for evaluating the updated distribution. The selection could then comprise one of the user's selection, such as an Alias-efficiency, a G-efficiency, an A-efficiency, a D-efficiency or an I-efficiency.
The method 4250 could be performed for one or more iterations. For example, an operation 4257 could be used to determine whether to end the method 4250. For example, the decision could be based on a criterion for the score of the updated distribution or an improvement in distribution. Alternatively, or additionally, the decision could be based on user input or computer system input indicating a default level of iterations. If the method is not ended, an operation 4258 can be used to determine whether to update the initial distribution. For example, the initial distribution could be updated in response to a selection of the updated distribution or other control of the initial distribution (e.g., user control).
In one example, of an iterative process in pseudo code, if there are N K-folds sets for a K-fold cross-validation there could be NK runs or iterations.
For idx=1 to N
For runidx=1 to NK
Repeat from runidx a set number of times.
Repeat from the beginning with a different set of random columns.
Choose the best design that maximizes the information matrix.
In one or more embodiments, comprising a method, operations could be performed in different orders, simultaneously or in conjunction with other operations described herein. For instance, operations 4201-4204 could be performed in different orders or simultaneously. Operations 4205 could be executed simultaneously with operations of method 4250.
In one or more embodiments, a computing device (e.g., computing device 4102) displays on a display device (e.g., output device 4106) a graphical user interface for user control of an assessment of a model.
An evaluation type 4304 is selected and validation data sets can be generated (e.g., validation 4306).
In this example, the user wants to generate a set of K-fold cross-validation sets (columns) for use in machine learning algorithms that use K-Fold cross-validation. Each one of these K-fold sets/columns will partition the data into K (usually equal sized) parts. Repeated K-folds means multiple of these K-Fold sets are needed.
In the example, the select columns dialog box 4320 is used to specify instructions for generating a data set in a table format with different columns, but other data format schemes could be used (e.g., selecting rows).
In one or more embodiments, there is a Validations Controls option 4312 for specify user controls on the validation. For instance, the user has a text box 4314 for specifying the number of K-fold columns and a text box 4316 for specifying the number of folds within each column. Other entry methods could be used (e.g., by inserting the identifiers of the K-fold columns into a dialog box or using a slide bar to indicate an amount or number).
The validations controls 4312 could be used to specify other controls or criterion that are used in creating the validations (e.g., in determining a distribution of group identifiers). In the example, shown the user can select a criterion for evaluating the updated distribution in the criterion text box 4318. In this case, the computing device is configured to receive a selection of a criterion for evaluating the updated distribution, and the user has selected a D for D-efficiency. However other selection options could include one or more of Alias-efficiency, a G-efficiency, an A-efficiency, or an I-efficiency.
Additionally, or alternatively, a criterion could relate to a number of changes of group identifiers. For example, a computing device (e.g., computing device 4102) could display, on a display device (e.g., output device 4106), a graphical user interface for user entry indicating a number of changes of group identifiers. The computing device can then receive, from a user of the graphical user interface, via one or more input devices, user input indicating the number of changes of group identifiers. The computing device can determine, based on the user input, a set of updated distributions comprising the updated distribution (e.g., a set of updated distributions generated using method 4250). The computing device can evaluate the updated distribution by evaluating each of the set of updated distributions and select the updated distribution out of the set of updated distributions.
A Cast Selected Columns into Roles dialog box 4310 is used to give a user the ability to specify one or more columns to use as a covariate in creating the K-fold columns or to stratify data to achieve a certain proportion of data observations associated with a particular factor. One or more embodiments can be used to create K-fold validation columns that are near-orthogonal to those covariates. For instance, if the user enters a response as the covariate, the output validations will have a better spread of response data points within each of the folds compared to random assignment with, for instance, all low values of a response within a single fold. One or more inputs or responses can be specified in the Cast Selected Columns into Roles dialog box 4310.
In one or more embodiments, a control is given to generate data sets for the validations or otherwise generate an indication of the validations. In this case by clicking the “Go” button invokes a platform (e.g., JMP's “Custom Design” platform), where the number of factors used by the platform is the value specified by the “Number of K-fold Columns to Create” control text box 4314, and the levels for each factor set to the value specified by the “Number of Folds (K) Within Each Column” control text box 4316. The columns selected by the “Stratify By” control dialog box 4310 are used as covariates within the “Custom Design” platform, which ensures that the values of those columns are used in constructing the design.
Traditional approaches to creating K-folds was to randomly distribute test cases in the different folds. However, claimed embodiments improve on this approach and create multiple K-fold sets.
Using one or more embodiments, described herein, K-fold columns are generated by making a number of K-fold sets with K-fold columns that are near-orthogonal under an orthogonal coding. In other words, for any level of a K-folds column, there is an even spread for the remaining levels of the K-folds columns. For example, column 4401 shows a portion of test cases assigned to a first group with the group identifier “1”. In other folds shown in columns 4402 and 4403, these same test groups are distributed amongst one of different groups with identifiers 1-5.
In one or more embodiments, generated K-fold sets are data sets that can be passed off to some machine learning algorithm for validation. For example, they can be added to a data table to be used by XGBoost. XGBoost is an optimized distributed gradient boosting library that implements machine learning algorithms under the Gradient Boosting framework. XGBoost can be run in distributed environments (e.g., a parallel processing database provided by Hadoop®). The validation data sets are shown as part of a single data set as shown in
For K-fold validation, different groups can be used as a test group and training group, where the training group is used to construct a model that is tested by data of the test group. For instance, a first test model can be constructed or generated with test cases assigned to a first group identifier associated with a training group in a fold (e.g., Fold B of column 4402). A first test evaluation can be generated by testing the first test model with data of test cases assigned a second group identifier associated with a test group. Then, a second test model can be constructed or generated with test cases assigned to the second group identifier associated with a training group in a fold (e.g., Fold B of column 4402). A second test evaluation can be generated by testing the second test model with data of test cases assigned a first group identifier associated with a test group. A model evaluation of all the data for generating the model to predict responses based on the factors can be generated based on multiple test evaluations (e.g., the first test evaluation and the second test evaluation).
There are times where it is helpful to have one or more inputs or responses from data be accounted for in the K-fold sets (e.g., columns indicated in dialog box 4310 in
For example, the portion of the data set 4400 shown has a large distribution of tax rate ratios in column 4404 (ranging from 233 to 469). Segments in this range could comprise one or more segments (e.g., below 350 and 350 or greater). The amount of inputs in a range segment do not need to be the same. As shown, there are inputs in both segments distributed in group 1 of Fold A.
In the same or different embodiment, a given validation set can be used to control assignment in other validation sets. For instance, a computing device (e.g., computing device 4102) can receive a user indication indicating control of assignment of the different group identifiers to each test case of the plurality of test cases is to be based on a group of a first validation of different validations of the one or more validations. For example, a fold (e.g., Fold A of column 4401) can be selected indicating each of its groups (e.g., by selecting a fold column in a select columns dialog 4320). For a second validation of the different validations, a computing device (e.g., computing device 4102) can control assignment of the different group identifiers to each test case of the plurality of test cases such that all test cases of the group of the first validation are distributed among all or most of different groups of other validations of the different validations.
In the same or different embodiment, a computing device (e.g., computing device 4102) can receive a user indication indicating control of assignment of the different group identifiers to each test case of the plurality of test cases is to be based on responses of the test cases. Data can be received comprising a value, associated with a range of values, for the response in each of the respective test cases. For each of multiple validations of the one or more validations, a computing device controls assignment of the different group identifiers to each test case of the plurality of test cases such that: all test cases identified by a same group identifier of a given validation of the multiple validations are distributed among all of different groups of other validations of the multiple validations, and all test cases with inputs in a particular segment of the range of values are distributed among all of the different groups of the other validations of the multiple validations.
In the example shown in
In one or more embodiments, multiple factors, response, or validation data sets can be considered in controlling assignment. For example, a computing device (e.g., computing device 4102) may receive a user indication indicating control of assignment of the different group identifiers to each test case of the plurality of test cases is to be based on responses of the test cases as well as a set of the plurality of factors, where the set comprises more than one factor. The computing device receives data comprising a response value associated with a response range of values for the response in each of the respective test cases. The computing device also receives respective values, each associated with a respective range of values for each of the set of the plurality of factors. For each of multiple validations, the computing device controls assignment of the different group identifiers to each test case of multiple test cases such that all test cases identified by a same group identifier of a given validation of the multiple validations are distributed among all of different groups of other validations of the multiple validations. Further, all test cases with inputs in a particular segment of a respective range of values are distributed among all of the different groups of the other validations of the multiple validations and all test cases with inputs in a particular segment of the response range of values are distributed among all of the different groups of the other validations of the multiple validations.
In one or more embodiments, it may be desirable to keep certain test cases grouped within a given validation data set. For example, information may be known about various test cases which shows a relationship or correlation between those test cases. In that case it may be useful to group the test cases in a single group. As one example, if the system is a wild cat ecosystem, and the response is a population of wild cats in an area over time, factors influencing this population may relate or be linked to individuals of the population. For example, a test case that had a factor related to size of a recorded footprint and where it was spotted (e.g., in a particular area of a preserve) may be data from the same animal (e.g., an animal with the same identifier). It may be useful to group data likely to be attributed to a same individual wild cat within a same group to ensure a good distribution of data from different wild cats in different groups.
In this case, assignment may be controlled by using a so-called restricted factor to restrict levels or ranges of that factor with the same group.
In this example, the factor “chas” could stand for Charles River with one level assigned to whether the property was on the river (1) and one level if it was not (0). It would be expected that homes on a river in a land-locked area would have a higher than a comparable home not on a river location. Thus, to discount the artificial increase in medium home prices from this individual factor, the chas factor was restricted such that all levels of 0 were in one or more groups and all levels of 1 were in one or more other groups. In the example, shown in each of the validation sets in columns 4602, 4603 and 4604, chas of 0 are all shown in a group together (although not necessarily the same group in each of the validations).
In one or more embodiments, a computing device (e.g., 4102) receives a restriction request indicating to restrict a factor in some way. For instance, the inputs for a given factor in a set of factors indicates a value associated with a respective range of values for the given factor in each of respective test cases. The respective range of values comprises a respective plurality of segments of a contiguous range of values. The restriction request may indicate to restrict a given segment of the respective plurality of segments to a same group of different groups. For each of multiple validations, a computing device can control assignment of the different group identifiers to each test case of the plurality of test cases such that all test cases identified by a first group identifier of a given indicated validation of the indicated validations comprise the test cases of the given segment
In one example, of pseudo code for a process to control assignment for N K-folds sets:
In the case of a restricted factor, rather than simply changing a single identifier, all the group assignments for data rows of the same value or value range for the restricted factor may be changed as well in a considered distribution.
Computing device 4802 includes an input interface 4808, an output interface 4810, a computer readable medium 4812, and a processor 4814. In other embodiments, fewer, different, and additional components can be incorporated into computing device 4802.
The computing device 4802 can receive information or requests from one or more input devices 4804 via input interface 4808. For instance, as shown in
In one or more embodiments, the computing device 4802 also receives one or more inputs or parameters for indicating a type of design for the experiment (e.g., a supersaturated design or group-orthogonal supersaturated design). These parameters may be specifications for the type of design for the experiment. For example, one or more parameters of multiple parameters for the design of the experiment could include or indicate a total number of the plurality of test cases for the design of the experiment, a total number of the different factors for the design of the experiment and/or a total number of different groups for grouping factors of the different factors.
For example, the computing device 4802 can receive parameter information 4852 from the same or different input devices 4804 as it received request 4850. For instance, the computing device 4802 may receive a value for a parameter of multiple parameters for the design of the experiment (e.g., a value indicating 8 total test cases). The parameter information 4852 may be received with or separately from the request 4850. In one or more embodiments, the computing device 4802 stores default values for parameters (e.g., stored in computer readable medium 4812) and the computing device 4802 receives a value for a parameter by receiving a user indication to select or use the default value. More than one value for a parameter may be received and/or respective values for more than one parameter of the multiple parameters may be received.
In one or more embodiments, input devices 4804 comprise one or more devices for user entry (e.g., entry of value for a parameter) into the system 4800. For instance, the input device 4804 could include one or more of a mouse 4820 or a keyboard 4822. Alternatively, or additionally, the input device 4804 includes a display, a track ball, a keypad, one or more buttons, a sensor, a phone, etc. Input interface 4808 in the same or different embodiments further provides an interface for receiving information from another device or machine such as a computing system 4824.
The computing device 4802 outputs information to output device 4806 via output interface 4810. Output interface 4810 provides an interface for outputting information (e.g., information representing a supersaturated design indication 4854) for review by a user and/or for use by another application or device or multiple applications or devices. For example, output interface 4810 interfaces with various output technologies including, but not limited to, display device 4826 and a printer 4828. In the same or different embodiments, output interface 4810 interfaces with a data storage device 4830 for data storage of output information.
In an alternative embodiment, the same interface supports both input interface 4808 and output interface 4810. For example, a touch screen provides a mechanism for user input and for presentation of output to the user. Alternatively, the input interface 4808 has more than one input interface that uses the same or different interface technology. Alternatively, or additionally, the output interface 4810 has more than one output interface that uses the same or different interface technology.
In one or more embodiments, the system 4800 is useful for outputting to output device 4806 an indication 4854 of a supersaturated design. In some embodiments, the output device 4806 may be one or more of a display device 4826, a printer 4828, a data storage device 4830 or another device not shown. For example, the indication 4854 of the supersaturated design may be the design itself displayed on a display device (e.g., a generated test condition for each element of a test case of the design of the experiment). Additionally, or alternatively, the indication 4854 may be an indication of parameters of the design (e.g., for selection of the supersaturated design or generation of the design). Additionally, or alternatively, the indication 4854 may describe one or more properties or features of the design, factors, or test cases of the design (e.g., correlation information between factors according to the supersaturated design).
Computer readable medium 4812 is an electronic holding place or storage for information so the information can be accessed by processor. Computer readable medium 4812 can include, but is not limited to, any type of random access memory (RAM), any type of read only memory (ROM), any type of flash memory, etc. such as magnetic storage devices (e.g., hard disk, floppy disk, magnetic strips), optical disks (e.g., compact disc (CD), digital versatile disc (DVD)), smart cards, flash memory devices, etc.
Processor 4814 executes instructions (e.g., stored at the computer readable medium 4812 for generating the indication). The instructions can be carried out by a special purpose computer, logic circuits, or hardware circuits. In one or more embodiments, processor 4814 is implemented in hardware and/or firmware. Processor 4814 executes an instruction, meaning it performs or controls the operations called for by that instruction. The term “execution” is the process of running an application or the carrying out of the operation called for by an instruction. The instructions can be written using one or more programming language, scripting language, assembly language, etc. Processor 4814 operably couples with input interface 4808, with output interface 4810 and with computer readable medium 4812 to receive, to send, and to process information. Processor 4814 in one or more embodiments can retrieve a set of instructions from a permanent memory device and copy the instructions in an executable form to a temporary memory device that is generally some form of RAM.
In one or more embodiments, computer readable medium 4812 stores instructions for execution by processor 4814. For example, computer readable medium 4812 could comprise instructions for a value(s) application 4860 for generating respective value(s) for each of one or more parameters for the design of an experiment Additionally, or alternatively, the computer readable medium comprises instructions for a supersaturated design application 4862 for generating a supersaturated design for the experiment
For example, in one or more embodiments, the computing device 4802 generates a respective value for each of one or more other parameters of the multiple parameters such that the design of the experiment according to the multiple parameters is a supersaturated design. In the supersaturated design the total number factors (e.g., F1 . . . Fx) is greater than the total number of test cases in the design (e.g., TC1 . . . TCy). The generated respective values can be generated for parameters different than parameters in which values were received. In other words, the computing device 4802 can determine missing values needed to generate a supersaturated design.
In the same or different embodiment, the computing device can generate a supersaturated design (e.g., based on values received by computing device 4802 for parameters and/or based on generated values for parameters). In one or more embodiments, the generated supersaturated design for the design of the experiment is a design that distributes each of the different factors into one of different groups. For example, the supersaturated design may be a group-orthogonal supersaturated design in which the groups are designed to be orthogonal to one another or optimal in some way. Factors (e.g., represented in columns) in the design may be partitioned into groups such that the factors/columns within a group are correlated to the others within the same group, but are orthogonal to any factor/column in any other group.
For example, the supersaturated design application 4862 may have different stored combinations of values (e.g., of runs or factors for the experiment) that would yield a group-orthogonal supersaturated design or GO-SSD. The computing device 4802 can be used to let the user specify, for example, either the number of runs of the number of factors. These combinations can be used to suggest values for unspecified parameters (e.g., to present to a user). Selected combinations can then be used to compute or construct a group-orthogonal supersaturated (e.g., without using a constructed design catalog).
In one or more embodiments, one or more applications stored on computer readable medium 4812 are implemented in software (e.g., computer readable and/or computer-executable instructions) stored in computer readable medium 4812 and accessible by processor 4814 for execution of the instructions. The applications can be written using one or more programming languages, assembly languages, scripting languages, etc. The one or more application can be integrated with other analytic tools. In one example, a value(s) application 4860 and a supersaturated design application 4862 are integrated with data analytics software application and/or software architecture such as that offered by SAS Institute Inc. of Cary, N.C., USA. Merely for illustration, the applications are implemented using or integrated with one or more SAS software tools such as IMP®, Base SAS, SAS® Enterprise Miner™, SAS/STAT®, SAS® High Performance Analytics Server, SAS® Visual Data Mining and Machine Learning SAS® LASR™ SAS® In-Database Products, SAS® Scalable Performance Data Engine, SAS® Cloud Analytic Services, SAS/OR®, SAS/ETS®, SAS® Inventory Optimization, SAS® Inventory Optimization Workbench, SAS® Visual Analytics, SAS® Viya™, SAS In-Memory Statistics for Hadoop®, SAS® Forecast Server, and SAS/IML® all of which are developed and provided by SAS Institute Inc. of Cary, N.C., USA.
In one or more embodiments, the system 4800 implements a method as described herein (e.g., a method shown in
An operation 4901 of method 4900 comprises receiving a request for a design of an experiment. For instance, the design of the experiment comprises a plurality of test cases. Each test case of the plurality of test cases comprises multiple test conditions for testing one of different factors for the experiment A test condition of the multiple test conditions comprises one of different options for a given factor in the design of the experiment.
An operation 4902 of method 4900 comprises receiving a first value for a first parameter of multiple parameters for the design of the experiment. The multiple parameters indicate a total number of the plurality of test cases for the design of the experiment, a total number of the different factors for the design of the experiment and a total number of different groups for grouping factors of the different factors. For example, the operation 4902 could comprise receiving the first value by receiving input indicating the total number of the plurality of test cases for the design of the experiment
Operations 4901 and 4902 could be performed in any order or simultaneously.
An operation 4903 of method 4900 comprises generating a respective value for each of one or more other parameters of the multiple parameters such that the design of the experiment according to the multiple parameters is a supersaturated design where the total number of the different factors is greater than the total number of the plurality of test cases. The one or more other parameters of the multiple parameters are different than the first parameter. For example, options could be presented for parameters such that each of different groups could comprise a same number of factors, and the total number of (unspecified) factors is a power of 2.
In one or more embodiments, the operation 4903 comprises generating multiple options indicating different possible values for each of the one or more other parameters of the multiple parameters. These options may be ones that would allow generation of a complete group-orthogonal supersaturated design. For example, in a case where a computing device has received a selection of a number of test cases, a computing device could generate options indicating different total numbers of the different factors for the design of the experiment and the total number of different groups for the different factors. The multiple options could be displayed (e.g., on an output device 4806 for displaying a graphical user interface shown in Figure SOB or
An operation 4904 of method 4900 comprises generating, based on the first value for the first parameter and the respective value for each of the one or more other parameters, the supersaturated design for the design of the experiment that is a design that distributes each of the different factors into one of the different groups. For example, generating the supersaturated design could be based on a received value for a parameter and based on selected generated parameters.
In one or more embodiments, a supersaturated design is constructed based on a Kronecker product of generated matrices. For example, in one or more embodiments, a computing device selects a respective order for each of one or more Hadamard matrices based on received values and generated values parameters of multiple parameters; and computes, based on the respective order, the one or more Hadamard matrices.
An operation 4905 of method 4900 comprises optionally outputting, to an output device, an indication of the supersaturated design.
For example, after the specifications of the parameters are set to the user's liking, the Make Design button 5051 can be used to trigger generation and display of a design (e.g. a design shown in
In one or more embodiments, the graphical user interface 5050 is part of the graphical user interface 5000 or is a separate interface. In the example shown in
In one or more embodiments, a computing device (e.g., computing device 4802) determines an optimal assignment of factors. For example, the computing device may assign factors to different groups for the different factors according to a first design. The computing device computes one or more measures of orthogonality between the different groups according to the first design. For example, a dot product is taken between options for the factors of one group and options for factors of another group. The computing device assigns factors to different groups for the different factors according to a second design, different from the first design, and computes one or more measures of orthogonality between the different groups according to the second design. The computing device selects a selected one of the first design or the second design based on comparing the one or more measures of orthogonality between the different groups according to the first design to the one or more measures of orthogonality between the different groups according to the second design. The computing device can then generate the supersaturated design for the design of the experiment according to the selected one of the first design or the second design.
However, the factors are assigned, in one or more embodiments, a user can update these assignments. For example, a user may believe that different factors may be important to observe. The user may want to separate these important factors into different groups. As shown in
In one or more embodiments, a computing device (e.g., computing device 4802) assigns design options to each of the reserved factors. The computing device generates, based on the reserved factors with assigned design options, received values/and or generated values, the supersaturated design for the design of the experiment.
The options in this example are assigned by the computing device for each of the different types of factors and the options are represented as “1” and “−1”. If the factor is a categorical factor, a “1” is assigned to represent one option and a “−1” is assigned to represent the other. Where the factor is continuous, a 1 can be assigned to represent a high value in the continuous range and the −1 can be assigned to represent a low value in the continuous range. Different values or selection for options can be made. In other examples, a user can assign the option types or change an assigned option type. For example, a user may have a belief as to the direction of an effect (i.e. a positive or negative slope) and a 1 can be assigned to a level expected to increase the response of a model. This may improve the ability of the design to detect active effects.
In one or more embodiments, a computing device displays in a graphical user interface (e.g., graphical user interface 5600) a recommended value for a parameter of the multiple parameters based on one or more user goals for the supersaturated design. For example, the options displayed via the graphical user interface 5600 in a structure options box 5710 provide design parameters that would create a group-orthogonal supersaturated design with 32 factors. In one or more embodiments, the computing device receives, from a user of the graphical user interface, a selection of a selected one of the multiple options or a recommended value; and generates the supersaturated design based on the received value and the selected one of the multiple options.
In one or more embodiments, the user may provide values for multiple parameters. For example, as shown in the drop-down box 5604, the user has restricted the factors to two groups by entering a “2” into the text box 5605.
This can restrict the number of suggested structure options in a structure options box 5610 (e.g., by graying out options that would not reflect the user's goals for the design or by not displaying those designs at all).
In this example, the user has explicitly indicated a value for the number of factors and the number of groups. However, in other embodiments a value for a parameter could be implicitly indicated. For example, the group size is implicitly indicated here based on the restrictions placed on the number of factors and number of groups. In other examples, other values could be indicated implicitly or explicitly. For example, the total number of the different groups of the different factors could be implicitly indicated by indicating the number of factors and the total number of factors in a group for the design of the experiment (or group size).
As in other embodiments, a user can generate a design by selecting the make a design button 5620. Additionally, or alternatively, the user sets features about the design as in other embodiments (e.g., the user can now change the design role from continuous to categorical, change the factor names, and change the factor levels).
In an operation 5704 the method 5700 comprises user customization (e.g., user customization shown in
In an operation 5705 the method 5700 comprises generating a design.
In one example of generating a design, let H be a Hadamard matrix of order m, m=0(mod 4). Let T be a matrix with more rows w then columns q, and let
X=H⊗T (1)
where “⊗” is the standard Kronecker product so that X is an mw×mq matrix.
For example, T could be related to an optimal design (e.g., a transpose of a D-optimal design for q runs and w columns) or T could be related to a covering array. In one example let T correspond to the first w rows of a Hadamard matrix of order q. Assuming the first column of both H and T is an n x 1 vector of ones which has first column comprised of all 1's. This will be the design (or model) matrix of a main-effects design for the p=mq−1 two-level factors having number of observations n=mw. The first column represents the intercept, while the other columns give the levels of the factors. For convenience of notation, an intercept is a “factor” with one level only, and the factors are labelled according to the columns of X.
Define m groups of factor indeces:
G
j={(j−1)q+1, . . . , qj}, j=1, . . . , m.
where q is the number of columns in T, and let Gj, j=1 . . . , m denote the submatrix of X comprised of the columns in the jth group. With this setup, it is easy to see that any factor from group Gj is uncorrelated with any factor from another group Gj, for j≠j′. To see this, observe that
X′X=H′H⊗T′T=mI
m
⊗T′T
Thus X′X is a block diagonal matrix, which means that columns corresponding to distinct groups are orthogonal. In particular, X′X has block-diagonal structure comprised of m q×q blocks as follows:
where O denotes matrix of all zeros. In addition, all of the matrices Gj, for j>1, are balanced. A design constructed using equation (1) is a group-orthogonal supersaturated designs (GO-SSDs).
As an example, assume that based on received values and generated values, a supersaturated design was constructed for 12 test cases and 16 columns (comprising factors, intercept and pseduo-factors). Let H be a 4×4 Hadamard matrix so that m=w=4 and let T be any 4×4 Hadamard matrix with a row removed, so that q=3. The number of rows is mq=12 and the number of columns is mw=16 as required, where:
The resulting design 5750 D=H⊗T is shown in
In addition to creating the GO-SSDs, one or more embodiments include analyzing output of an experiment taking advantage of the generated design structure.
The method 5780 comprises an operation 5781 of providing a design with factors and a response. For example, the design may be a supersaturated design generated using method 5700. The response may come from implementing the experiment according to the design. Alternatively, the design may not even be a supersaturated design at all.
In analyzing the output, one issue is how to handle validating that the design has a group-orthogonal structure. Method 5780 comprises an operation 5782 to verify the structure of the design. For example, a computing device can validate whether the supersaturated design has a group-orthogonal structure. The computing device can selectively, based on whether the supersaturated design has a group-orthogonal structure, generate an analysis of the supersaturated design; and output, to an output device, the indication of the supersaturated design by indicating the analysis of the supersaturated design.
In one or more embodiments, a computing device studies the structure of a design and checks if the structure is group orthogonal by, for instance, partitioning inputs based on their orthogonality to the other factors. Alternatively, or additionally, if the design was created in IMP®, metadata is stored reflecting the number of runs, factors, groups, and group sizes. This metadata can be checked to confirm the structure of the design.
In one or more embodiments, if the structure is not group-orthogonal (i.e. has an invalid structure), an operation 5784 could be used to generate an error message. Additionally, or alternatively, an operation 5783 can be used to suggest alternative modeling. For example, a generalized regression platform (or stepwise regression) could be pre-populated with the main effects model since the GO-SSD analysis cannot be done. The order of operations 5783 and 5784 is merely an example. Operations 5783 and operations 5784 can be performed in any order or performed simultaneously.
In the same or different embodiments, if the structure is group-orthogonal (i.e. has a valid structure), an operation 5785 can be used to perform analysis for a group-orthogonal supersaturated design. As one example, the analysis of the design comprises identifying active factors. An active factor could be indicated as a factor that significantly contributes to an outcome of the experiment. The active groups comprising these active factors could be displayed as part of operation 5785. In one or more embodiments, the operation 5786 could indicate groups that require further study. For example, it could indicate further consideration of whether the groups are at full rank. For instance, if the number of active factors exceeds the rank of a Hadamard matrix used to generate the design, this could indicate that it is difficult to identify the active factors in those groups.
In one or more embodiments, the indication of the supersaturated design comprises an indication of a contribution of the active factor to an outcome of the experiment or other information related to an active factor (e.g., to display active groups).
In one or more embodiments, a computing device (e.g., a computing device 4802) generates a model for determining predicted responses of a computer-simulation of the experiment according to the supersaturated design for the design of the experiment.
In the example shown in
y=3.99X2+3.52X3+3.56X6+3.76X7+3.35X6+3.5X7
The model may not need to consider all the factors, or the response assigned may be 0. Other features of the model can be selected, such as normal, binomial or poisson distributions and a value for error in the model. In this case a normal distribution is selected. In one or more embodiments, a computing device generates an error variance (e.g., one that is independent of the model). In this case an error variance (σ) of 1 is selected in graphical user interface 5800. The computing device can generate a simulated set of responses for each of the test cases according to the model, the error variance, and the supersaturated design. The computing device can output, to the output device, an indication of the supersaturated design by outputting the simulated set of responses (e.g., by selecting an apply button 5801).
One or more embodiments show a standard error or variance of the factor. A t ratio is also shown, which is the estimate divided by the standard error. An absolute value of the t ratio is used to determine a probability of the significance of the factor.
Graphical user interface 5950 is another example of an indication of a contribution of an active factor to an outcome of an experiment according to the supersaturated design.
In this example, the active factors are combined into a combined model parameter estimates 5951.
Other information about a model according to a supersaturated design can be given. For example, a degree of freedom (DF) 5953 for the error can be computed. A degree of freedom indicates the number of independently variable factors effecting an outcome. This is a measure of the freedom in the model (e.g. what extra information or assumptions about the relationship between factors may be limiting the model). A higher value means there should be more confidence in the model because there is a greater independence of factors. It can also be used as a test (F-test) or indicator of active factors. A root mean square estimate (RMSE) 5952 can be computed. The RMSE is a measure of the difference between values predicted by the model and values observed. A user can use this information in deciding to make a model 5954 or run a model 5955.
Accordingly, embodiments herein present techniques useful for generating a design for an experiment (e.g., a supersaturated design) and analyzing a model of the experiment according to the design.
Accordingly, embodiments herein present techniques useful for generating a design for an experiment (e.g., a supersaturated design) and analyzing a model of the experiment according to the design.
In one or more embodiments, herein a computing system is used to generate a group-orthogonal supersaturated design. In the same or different embodiments, a computing system (e.g., system 4800 and/or a computing device 4802) can generate a supersaturated design that is based on a D-optimal design. D-optimal designs are historically unsuited for supersaturated designs. Bayesian D-optimal designs, for instance, often would have pairs of fully confounded factors (i.e., the effects of factors in different test cases are indistinguishable). However, even with improvements herein with group-orthogonal designs to have flexibility in run size, there are still limitations on the number of runs and number of groups possible. In the absence of information on how to group the factors, a practitioner may end up with all active factors in one group, and an experiment analysis will not be able to identify effects of those factors on the outcome of an experiment. If the effects of factors of test cases in a test suite are indistinguishable, an entire test suite would likely be thrown out as useless for an experiment
Embodiments herein enable use of test cases in supersaturated designs by adding greater flexibility in producing optimal supersaturated design (e.g., designs that alleviate confounding of factors in traditional approaches for D-optimal designs and designs that improve criteria applicable to supersaturated design).
An operation 6001 of method 6000 comprises receiving a request for a design of an experiment. The design of the experiment comprises a plurality of test cases. Each test case of the plurality of test cases comprises multiple test conditions for testing one of different factors for the experiment. A test condition of the multiple test conditions comprises one of different options for a given factor in the design of the experiment. For instance, with a two-level design, each test condition is one of two levels (e.g., 1, and −1), where the levels represent two options for a factor (e.g., a high or low temperature in the experiment).
An operation 6002 of method 6000 comprises obtaining for the design of the experiment a run indication of a total number of the plurality of test cases for the design of the experiment and a factor indication of a total number of the different factors for the design of the experiment (e.g., receiving from a user or computing system parameter information, such as parameter information 4852 in
An operation 6003 of method 6000 comprises generating a data representation of a matrix Y. The matrix Y defines a supersaturated design for the design of the experiment. The generating the data representation of the matrix Y comprises selecting, based on the run indication and factor indication, dimensions for the matrix Y. For example, the dimensions of the matrix may comprise N number of rows and n number of columns. The generating the data representation of the matrix Y comprises generating a data representation of a matrix X according to an obtained design (e.g., a design obtained by approaches described herein). For example, generating the matrix X according to an obtained design may comprise generating a D-optimal design from adding rows to a predefined or constructed matrix (e.g., a Hadamard matrix or a circulant block matrix).
The matrix X has n+1 number of rows and M number of columns. M is greater than or equal to N. The generating the data representation of the matrix Y comprises computing an indication of correlation between effects of factors of a matrix Y. The matrix Y is a transposition of the matrix X or a matrix X*. Matrix X* is a first subset of the matrix X such that the transposition of matrix X* represents a same number of factors as a transposition of matrix X. The generating the data representation of the matrix Y comprises generating, based on the indication of correlation, the data representation of the matrix Y that is a transposition of the matrix X or the matrix X*. A transposition of the matrix X or matrix X* may have some advantages over other supersaturated designs (e.g., because of its D-optimality). The indication of the correlation can be used to select between possible matrices (e.g., matrix X or X*). For example, correlation of rows of X may correspond to columns of Y.
An operation 6004 of method 6000 comprises outputting, to an output device, a setting for each test condition for test cases of the supersaturated design for the experiment according to the matrix Y.
In one or more embodiments, the different factors for the experiment comprise one or more discrete factors. The test conditions comprise at least two levels representing respective discrete options for each of the one or more discrete factors. A computing system described herein outputs a setting for each test condition of a design by outputting a respective selection between two levels for respective ones of the one or more discrete factors in each test case of the plurality of test cases.
Additionally, or alternatively, the different factors for the experiment comprise one or more continuous factors with one or more respective ranges of options. The test conditions comprise at least two levels representing respective options for each of the one or more continuous factors within the one or more respective ranges of options. A computing system described herein outputs the setting for each test condition of a design by outputting a respective selection between two levels for respective ones of the one or more continuous factors in each test case of the plurality of test cases. For instance, the experiment could correspond to an experiment for a machine learning algorithm with different hyperparameters. The hyperparameters could be continuous factors with range(s) of options. The setting could indicate a value for a hyperparameter of the machine learning algorithm.
In one or more embodiments, the user can select a role for the factor of type continuous or discrete and values corresponding to one of levels for options for the factor.
In an operation 6201, the user requests a design (e.g., a supersaturated design with N runs and n factors). In an operation, 6202, a computing system generates a D-optimal design, X, with (n+1) runs and M columns (where M>=N). The design can be a chemical weighing or two-level design in that a given value in the matrix will take on one of two values. The extra row can be added to provide for an intercept row as described herein.
In one or more embodiments, multiple D-optimal designs are possible responsive to the given dimensions of the matrix X (e.g., based on a run indication and factor indication). A computing system can select between multiple D-optimal designs. For example, a design can be selected with the highest D-efficiency score or highest other efficiency score described herein (e.g., A-efficiency). A design can be selected based on one or more secondary criteria related to balancing test conditions for a factor as described herein. For instance, the one or more secondary criteria may be related to an evaluation of a E(s2) criterion, a UE(s2) criterion, or both, wherein:
where n is the number of columns of a matrix Y for a supersaturated design; and s is the elements of a matrix S where S=Y′Y. The E(s2) criterion may assume all factors are balanced (equal levels of + and −1's) such that s1j=0 for all j=2, . . . n, whereas that requirement can be relaxed with a UE(s2) criterion. A design or set of designs may be selected that minimizes a value for the criterion or meets a threshold value for the criterion. Additionally, or alternatively, other second criteria can be considered such as amount of maximization of the number of level-balanced factors.
In the example, in
If the matrix X requires additional rows, a computing system can selectively add rows favorable to a supersaturated design application (e.g., rows that are not identical to rows already in the matrix X, rows that are not mirror-image rows to rows already in the matrix X, rows that are themselves a balanced subset, and/or add rows that further balance the matrix X. A mirror row has the same value for each element of the row as another row multiplied by −1.
An intercept row for the design can be selected. A computing system can set the intercept row to a same option for each test condition. For example, if the intercept row will be the first row of a final design, a computing system multiplies the columns of X by −1 for any column in which the first row is −1. This has no effect on the optimality of X but ensures the first row of a final design is set to 1 for all columns.
In an operation 6203 it is determined whether M is equal N. In other words, does the matrix X have the same number of columns already as the requested design?
If M=N, an operation 6204 comprises computing an indication of correlation between effects of factors of a matrix Y that is a transposition of matrix X. In this example, the method 6200 comprises detecting favorable test conditions for correlation. For example, if it is determined that there is an acceptable indication of correlation (e.g., no rows in X are the same or mirrors), in an operation 6204, then a D-optimal supersaturated design is given by X′ (transpose of X) in an operation 6205.
Otherwise, if a computing system determines that M is not equal to N; subsets X* of the matrix X can be considered. An operation 6206 of method 6200 comprises generating an X* which is a subset of N columns from X. For example, the subset can be a random selection of N columns. As shown, in
In some embodiments, multiple candidate matrices that are each a different subset of matrix X and each represent the same number of factors as matrix X can be selected (e.g., for a set number of times). The method 6200 can be used to compute a respective indication of a measure of balance of test conditions for respective factors for each of the multiple candidate matrices; and select, based on a given respective indication for each of the multiple candidate matrices, a matrix X* for outputting a supersaturated design. For example, the operation 6208 comprises determining whether a current matrix has the best measure of balance. A measure of balance could indicate a number of level-balanced (or nearly level-balanced) factors and/or number of orthogonal (or nearly orthogonal) pairs of factors for a design resulting from the matrix.
In some embodiments, if the measure of balance of test conditions is better, further analysis of the design can be conducted. For example, an operation 6209 comprises computing an indication of correlation between effects of factors of a matrix Y that is the transposition of the matrix X*. It is then determined whether the indication of correlation is acceptable. For instance, the computing may comprise detecting favorable test conditions for a supersaturated design such as a set of test conditions indicating that that there is no row in matrix X* with a same or mirror test conditions as another row. In an operation 6210, the indication I is updated and the X* is updated based on detecting the set of test conditions. Alternatively, if the indication of correlation is not acceptable, in one or more embodiments, a computing system refrains from selecting a matrix for transposition (e.g., a matrix with two or more rows with same test conditions or two or more rows with mirror test conditions).
If the measure of balance of test conditions is not better or the indication of correlation is not acceptable, the process may generate another subset if it is determined that there is still processing time left in an operation 6211. For example, in one or more embodiments, a computing system obtains a process indication indicating an amount of allowed time, or number of computations, for processing the measure of balance of test conditions. Alternatively, or additionally, the processing time may be set based on the number of possible candidate matrices (e.g., a ratio of the number of possible candidate matrices).
In an operation 6212, the best design is returned by transposing the best X*.
In one or more embodiments, (e.g., according to the method shown in
In this example, a respective indication of a measure of balance of test conditions of subset 6304 (e.g., an operation 6207 of the method 6200 in
where i is one or more indexes for rows of a respective candidate matrix corresponding to subset (e.g., subset 6304); j is one or more indexes for columns of the respective candidate matrix; A is a total number of columns of the respective candidate matrix; B is a total number of rows of the respective candidate matrix; and w is a value assigned to a location of the respective candidate matrix indexed by row i and column j. A minimum SR2 value is preferred as a proxy for more evenly balanced columns. The SR2 data value 6334 for the design in
Using method 6200 in
In one or more embodiments, a candidate matrix may be further vetted by computing an indication of correlation (e.g., in operation 6209 of
In this example, the matrix 6362 in
where pij is correlation between columns i and j; and
m is number of columns in the matrix Y.
A design may be selected that minimizes a maximum of the ψj, that minimizes an average of the φj values, or that minimizes the maximum absolute correlation (i.e., minimize the maximum of |ρij|).
Different design diagnostics can be performed on, and different options set for, a supersaturated design (e.g., diagnostics and options shown in
In one or more embodiments, a computing system computes a respective indication of a measure of balance of test conditions of a candidate matrix by evaluating one or more of: a measure of orthogonality between factors according to a supersaturated design according to matrix based on the transposition of the candidate matrix; error in fitting a model according to the supersaturated design; and closeness to a Bayesian D-optimal design.
For example,
In contrast,
In one or more embodiments, a computing system outputs the setting for each test condition for the test cases of the supersaturated design for the experiment according to the matrix Y by displaying in a graphical user interface the representation of the measure of correlation or orthogonality corresponding to these settings.
In one or more embodiments, a computing system described herein generates a representation of efficiency of estimating a factor in modeling the supersaturated design for the experiment
The fractional increase in confidence interval length compares the length of a parameter's confidence interval to an ideal design (e.g., an orthogonal D-optimal design would have a fractional increase of zero for each of the parameters).
The relative standard (Std) error of estimate gives a ratio of the standard deviation of a parameter's estimate to the error standard deviation. The relative standard error can be defined as the diagonal entries of a posterior distribution covariance matrix equal to:
(X′X+K2)−1
where K is a diagonal matrix with 1's on the diagonal and zero's elsewhere. The element is zero for necessary terms and 1 for possible main effects. X is a matrix of the response of the system according to the supersaturated design. Smaller values on the diagonal entries indicate a better design. Bayesian D-optimal designs maximize a determinant of the inverse of the posterior covariance matrix (i.e., det(X′X+K2)).
As shown in
In one or more embodiments, a computing system described herein outputs the setting for each test condition for the test cases of the supersaturated design for the experiment according to the matrix Y by displaying in a graphical user interface the representation of an efficiency corresponding to those settings.
Overall, the settings corresponding to
This application claims the benefit of and priority based on 35 U.S.C. § 119 to U.S. Provisional Application No. 63/039,280, filed Jun. 15, 2020, and is a continuation-in-part of U.S. application Ser. No. 16/719,200, filed Dec. 18, 2019 and U.S. application Ser. No. 16/692,172, filed Nov. 22, 2019. U.S. application Ser. No. 16/719,200 claims the benefit of and priority based on 35 U.S.C. § 119 to U.S. Provisional Application No. 62/861,563, filed Jun. 14, 2019. U.S. application Ser. No. 16/692,172 claims the benefit of and priority based on 35 U.S.C. § 119 to U.S. Provisional Application No. 62/886,189, filed Aug. 13, 2019 and is a continuation-in-part of U.S. application Ser. No. 16/507,769, filed Jul. 10, 2019, which issued as U.S. Pat. No. 10,535,422 on Jan. 14, 2020, which claims the benefit of and priority based on 35 U.S.C. § 119 to U.S. Provisional Application No. 62/796,214, filed Jan. 24, 2019, U.S. Provisional Application No. 62/807,286 filed Feb. 19, 2019, and U.S. Provisional Application No. 62/816,150 filed Mar. 10, 2019, and is a continuation-in-part of U.S. application Ser. No. 16/240,182, filed Jan. 4, 2019, which issued as U.S. Pat. No. 10,386,271 on Aug. 20, 2019, which claims the benefit of and priority based on 35 U.S.C. § 119 to U.S. Provisional Application No. 62/728,361 filed Sep. 7, 2018, and which is a continuation-in-part of U.S. application Ser. No. 16/154,290, filed Oct. 8, 2018, which issued as U.S. Pat. No. 10,338,993 on Jul. 2, 2019, which claims the benefit of and priority based on 35 U.S.C. § 119 to U.S. Provisional Application No. 62/702,247 filed Jul. 23, 2018, and U.S. Provisional Application No. 62/661,057, filed Apr. 22, 2018. The disclosures of each of these are incorporated herein by reference in their entirety.
| Number | Date | Country | |
|---|---|---|---|
| 63039280 | Jun 2020 | US | |
| 62861563 | Jun 2019 | US | |
| 62886189 | Aug 2019 | US | |
| 62796214 | Jan 2019 | US | |
| 62807286 | Feb 2019 | US | |
| 62816150 | Mar 2019 | US | |
| 62728361 | Sep 2018 | US | |
| 62702247 | Jul 2018 | US | |
| 62661057 | Apr 2018 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 16719200 | Dec 2019 | US |
| Child | 16985512 | US | |
| Parent | 16692172 | Nov 2019 | US |
| Child | 16719200 | US | |
| Parent | 16507769 | Jul 2019 | US |
| Child | 16692172 | US | |
| Parent | 10535422 | Dec 2005 | US |
| Child | 16507769 | US | |
| Parent | 16240182 | Jan 2019 | US |
| Child | 10535422 | US | |
| Parent | 16154290 | Oct 2018 | US |
| Child | 16240182 | US |
