Embodiments described herein generally relate to machine learning predictions and, in some embodiments, more specifically to hierarchical tournament-based machine learning predictions.
Machine learning may be used to make predictions based on evaluation of data. A graph network may include a data structure that is hierarchical. Evaluation of data in a hierarchical data structure may completed by evaluating data at differing levels of the hierarchy to establish a prediction. A prediction may provide an indication of a likely outcome based on the state of the data in the data structure.
In the drawings, which are not necessarily drawn to scale, like numerals may describe similar components in different views. Like numerals having different letter suffixes may represent different instances of similar components. The drawings illustrate generally, by way of example, but not by way of limitation, various embodiments discussed in the present document.
Prediction problems are often multidimensional and each dimension often consists of hierarchies, such as stock keeping units (SKUs) belonging to products which in turn can be grouped into categories. It is unknown upfront whether, for a given dataset, the most accurate forecasts may be arrived at if machine learning models are built at a SKU level (e.g., one model for each SKU), at a product level (e.g., one model for each product), or at a category level (e.g., one model for each category). It is also unknown whether the highest forecast accuracy is achieved by executing the model at the SKU level, product level, or category level The same may be simultaneously present in other dimensions such as at a store level, etc.
The combinations of forecast levels, model levels, and possible choices of algorithm are too numerous (e.g., combinatorial) to be able to be arrived by humans directly. To address this issue, hierarchical tournaments are executed over a set of computational nodes known as a cluster. A processor receives input from the configuration interface on the dimensions and hierarchies to be considered for the purposes of determining the best combination of algorithm, model level, and prediction level that generates the best quality of forecast.
Jobs are executed on the cluster that simulates the tournament. A global level model is built and multiple algorithms are used to evaluate each model and prediction accuracy is measured to pick a winning algorithm. Models are created at multiple levels and predicted. In this stage of the tournament, the model level is determined by having multiple model levels compete against each other and the winner is selected via similar prediction accuracy measurement. Multiple forecast levels are evaluated with the winning algorithm and the winning model level preset. Forecasts are generated at multiple hierarchical levels and the data is spread down to the desired level where applicable. The winner is measured by similar prediction accuracy calculation. A conclusion of the best choice of prediction level, model level, and algorithm and generates the prediction with minimum error is derived. The information is encapsulated in a reusable recipe that may be subsequently invoked to generate predictions.
The configuration interface 105 may receive dimensions and hierarchies to be considered for the purposes of determining the best combination of algorithm, model level, and prediction level that generates the best quality of forecast.
The products may be described in a graph cube network including nodes and edges that describe relationships between the products and a supply chain of a logistics system. The nodes may represent stores, products, manufacturers, manufacturing facilities, warehouses, and the like. The edges may define relationships between the nodes, for example, a relationship between a store and a warehouse, a relationship between a manufacturing facility and a component supplier, etc.
The algorithms (e.g., algorithm A 320, algorithm B 325, algorithm C 330, etc.) may include different techniques (e.g., mathematical functions, order of operations, attribute selection, etc.) for performing calculations at a model level and prediction level. The calculations may lead to differing prediction results that may then be compared to determine a tournament winner. The models (e.g., the global model 335, the planning item model 340, the country/brand model 345, etc.) may be machine learning models defined (e.g., trained using training data, configured neural network models, etc.) at hierarchical levels of a hierarchy such as the product hierarchy 205 as described in
The predict levels (e.g., planning items/store/week 350, brand/store/month 355, planning item/store/month 360, etc.) may define the prediction output of the model to be calculated using a selected algorithm and model. For example, the planning items/store/week 350 prediction may request output of a prediction calculation result indicating a number of planning items for a store in a given week indicated by a time period.
Returning to the description of
The storage layer 110 includes data from the graph cube network that describes the activity occurring among the nodes and edges of the graph cube network. For example, the data may include movement of goods throughout a supply chain of the graph cube network, movement of components between manufacturers and suppliers, etc. The storage layer 110 may be a large unstructured data warehouse that relies on big data storage techniques. The storage layer 110 provides the input data for the processor 115.
The processor 115 may be an artificial intelligence or machine learning processor that executes the algorithm selection engine 305, the model level selection engine 310, and the predict level selection engine 315 to select combinations of algorithms, models, and predictions to enter in the tournament. For example, the processor may select the algorithm B 325, the country/brand model 345, and the planning items/store/week 350 prediction as a combination to enter in the tournament. Additional combinations may include the planning items/store/week 350 and variations of the algorithm A 320, the algorithm C 320, the global model 335, the planning item model 340, and the country/brand model 345.
The selections may be based on a prediction by the processor 115 that indicates how likely a combination is to output a result within a threshold of accuracy, precision, or other metric to be achieved by the prediction. In an example, the processor 115 may use a machine learning model that has been trained to identify or classify a set of combinations as likely competitors in the tournament. The models may have been trained or may have learned based on previously conducted tournaments or based on labeled data indicating successful competitors in prior tournaments. The processor 115 generates a cluster 120 of nodes for executing the combinations to calculate their corresponding predictions. The cluster may execute the predictions in parallel to reduce processing time for making calculations or may queue the nodes to reduce resource consumption based on resources available during execution of the combinations.
The processor 115 receives the prediction output of the cluster 120 and evaluates the output to select a winner. Based on the configuration data received by the configuration interface 105, the processor 115 evaluates the predictions calculated by each combination to select the combination that provides the most correct result for the given metric (e.g., accuracy, precision, time to return, etc.) or set of metric provided. For example, the algorithm B 325, the country/brand model 345, and the planning items/store/week 350 prediction combination may have tied with another combination based on accuracy of prediction result, but may be selected as the winner because the combination arrived at the result in a shorter period of time resulting in lower computing resource consumption. The processor 115 may include a baseline prediction result for a given prediction and may evaluate the competitors using the baseline to identify performance of the competitor for a given metric. In an example, the processor 115 may use a classifier to evaluate the results amongst each other to rank the competitors automatically without having a predetermined baseline of the prediction.
The processor 115 may store the winner in the storage layer 110 for sue in calculating a future prediction. The processor 115 may periodically reconduct the tournament to validate that the winning combination remains the best combination for meeting the defined metrics. If the winner does not win a subsequent tournament, the storage layer 110 is updated with the current winner and will be used in subsequent prediction calculations.
The system 400 may include a configuration interface that receives configuration data 410, a storage layer 415 that contains data from a graph cube network 420, a processor 425 that includes an algorithm selection engine 430, a model selection engine 435, and a prediction engine 440. The processor 425 may distribute combinations for execution across a computing node cluster 445. The configuration data may include metrics (e.g., error rate, accuracy, precision, result return time, etc.) to be evaluated in a tournament to select a combination of an algorithm and model at a particular level for calculating a selected prediction. The configuration may be passed by the configuration interface 405 to the processor 425 and used to configure the processor 425 for evaluating the graph cube network data 420 stored in the storage layer 415. The configuration interface 405 may include a graphical user interface for input of configuration data or may include an application programmer interface (API) for programmatic input of configuration data by components of a logistics system.
As noted previously, the graph cube network 420 may include a variety of nodes and edges that represent logistical components within a supply chain. The storage layer 415 may include data that describes current and historical states of the graph cube network and relationships contained therein. A configuration is transmitted to the processor from the configuration interface 405 to the processor 425. The processor 425 may select combinations of algorithms, models, and predictions to enter into a tournament(s) to select a winning combination. The processor may select the entrant combinations based on an evaluation of various combinations to predict a probability that a given combination may calculate a result that will meet a metric provided in the configuration.
When the entrants have been determined, the processor 425 may retrieve data to be evaluated from the storage layer 425 to be passed to the cluster 445 when executing the combinations. The processor 425 may work in conjunction with the algorithm selection engine 430 to select an algorithm to be used in a given combination. The algorithm selection may be based in part on how the algorithm has performed in past evaluations based on feedback received from output derived using the algorithm. The processor 425 may work in conjunction with the model selection engine 435 to select a model for a given combination. The selection of the model may be based in part on how the model has performed in making past predictions. The processor 425 may work in conjunction with the prediction selection engine 440 to select a prediction for a given combination. In an example, the prediction may be based on configuration data included in the configuration received from the configuration interface 405. In another example, several predictions may produce similar outputs and various predictions may be selected in varying combinations to be included as entrants in the tournament. The predictions may be selected based in part on learned performance of the predictions by the processor 425. Algorithms, models, and predictions, by way of example and not limitation, are shown in
The processor 425 may transmit the selected combinations for execution on the computing node cluster 445. The computing cluster 445 may execute the combinations serially or in parallel using the graph cube data retrieved from the storage layer 415 as input. The cluster 445 returns output prediction results to the processor 425 for evaluation. The processor 425 evaluates the output prediction results using the configuration including the designated performance metrics. The processor 425 may evaluate the output prediction results against each other (e.g., using clustering, etc.) to determine a winner or may evaluate the output prediction results against a predetermined baseline to predict a winner. The baseline evaluation method may be useful in validation of an existing selected model to reduce computing resource load. Evaluation of the output prediction results may be useful in establishing a baseline where no baseline exists or where the validity of an existing baseline may be in question.
The winning combination selected by the processor 425 may be stored in the storage layer 415 for use in performing future predictions for similar configurations. The processor 425 may conduct periodic tournaments in the future to validate that the selected combination remains the best available combination for the provided metrics and prediction. Additional tournaments may be triggered when changes (e.g., additions, deletions, modification, etc.) are detected to the available algorithms, models, and predictions. In an example, additional tournaments may be performed on a refresh schedule to periodically validate selected combinations so that the selections do not become stale (e.g., out of date, ineffective, etc.).
At operation 505, a machine learning selection model may be trained (e.g., by the processor 425 as described in
At operation 510, a configuration may be received (e.g., by the processor 425 from the configuration interface 405 as described in
At operation 515, a set of evaluation component combinations may be selected (e.g, by the processor 425 as described in
At operation 520, the set of evaluation component combinations may be transmitted to a cluster of computing nodes. At operation 525, output results may be received for the set of evaluation component combinations.
At operation 530, the output results may be evaluated using the metric to determine a winning evaluation component combination In an example, the output results may be compared to a baseline value for the metric and the winning evaluation component combination may be determined based on an output result of the winning evaluation component combination having the largest positive deviation from the baseline value. In another example, the output results may be evaluated using a machine learning algorithm to determine a fit between the metric and each output result of the output results and the winning evaluation component combination may be determined based on an output result of the winning evaluation component having the closest fit with the metric. At operation 535, the winning evaluation component combination may be stored in storage for use in calculating future predictions for the target prediction.
Examples, as described herein, may include, or may operate by, logic or a number of components, or mechanisms. Circuit sets are a collection of circuits implemented in tangible entities that include hardware (e.g., simple circuits, gates, logic, etc.). Circuit set membership may be flexible over time and underlying hardware variability. Circuit sets include members that may, alone or in combination, perform specified operations when operating. In an example, hardware of the circuit set may be immutably designed to carry out a specific operation (e.g., hardwired). In an example, the hardware of the circuit set may include variably connected physical components (e.g., execution units, transistors, simple circuits, etc.) including a computer readable medium physically modified (eg., magnetically, electrically, moveable placement of invariant massed particles, etc.) to encode instructions of the specific operation. In connecting the physical components, the underlying electrical properties of a hardware constituent are changed, for example, from an insulator to a conductor or vice versa. The instructions enable embedded hardware (e.g., the execution units or a loading mechanism) to create members of the circuit set in hardware via the variable connections to carry out portions of the specific operation when in operation. Accordingly, the computer readable medium is communicatively coupled to the other components of the circuit set member when the device is operating. In an example, any of the physical components may be used in more than one member of more than one circuit set. For example, under operation, execution units may be used in a first circuit of a first circuit set at one point in time and reused by a second circuit in the first circuit set, or by a third circuit in a second circuit set at a different time.
Machine (e.g., computer system) 600 may include a hardware processor 602 (e.g., a central processing unit (CPU), a graphics processing unit (GPU), a hardware processor core, or any combination thereof), a main memory 604 and a static memory 606, some or all of which may communicate with each other via an interlink (e.g., bus) 608. The machine 600 may further include a display unit 610, an alphanumeric input device 612 (e.g., a keyboard), and a user interface (UI) navigation device 614 (e.g., a mouse). In an example, the display unit 610, input device 612 and UI navigation device 614 may be a touch screen display. The machine 600 may additionally include a storage device (e.g., drive unit) 616, a signal generation device 618 (e.g., a speaker), a network interface device 620, and one or more sensors 621, such as a global positioning system (GPS) sensor, compass, accelerometer, or other sensors. The machine 600 may include an output controller 628, such as a serial (e.g, universal serial bus (USB), parallel, or other wired or wireless (e.g., infrared (IR), near field communication (NFC), etc.) connection to communicate or control one or more peripheral devices (e.g., a printer, card reader, etc.).
The storage device 616 may include a machine readable medium 622 on which is stored one or more sets of data structures or instructions 624 (e.g., software) embodying or utilized by any one or more of the techniques or functions described herein. The instructions 624 may also reside, completely or at least partially, within the main memory 604, within static memory 606, or within the hardware processor 602 during execution thereof by the machine 600. In an example, one or any combination of the hardware processor 602, the main memory 604, the static memory 606, or the storage device 616 may constitute machine readable media.
While the machine readable medium 622 is illustrated as a single medium, the term “machine readable medium” may include a single medium or multiple media (e.g., a centralized or distributed database, and/or associated caches and servers) configured to store the one or more instructions 624.
The term “machine readable medium” may include any medium that is capable of storing, encoding, or carrying instructions for execution by the machine 600 and that cause the machine 600 to perform any one or more of the techniques of the present disclosure, or that is capable of storing, encoding or carrying data structures used by or associated with such instructions. Non-limiting machine readable medium examples may include solid-state memories, and optical and magnetic media. In an example, machine readable media may exclude transitory propagating signals (e.g., non-transitory machine-readable storage media). Specific examples of non-transitory machine-readable storage media may include: nonvolatile memory, such as semiconductor memory devices (e.g., Electrically Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM)) and flash memory devices; magnetic disks, such as internal hard disks and removable disks; magneto-optical disks; and CD-ROM and DVD-ROM disks.
The instructions 624 may further be transmitted or received over a communications network 626 using a transmission medium via the network interface device 620 utilizing any one of a number of transfer protocols (eg., frame relay, internet protocol (IP), transmission control protocol (TCP), user datagram protocol (UDP), hypertext transfer protocol (HTTP), etc.). Example communication networks may include a local area network (LAN), a wide area network (WAN), a packet data network (e.g., the Internet), mobile telephone networks (e.g., cellular networks), Plain Old Telephone (POTS) networks, and wireless data networks (e.g., Institute of Electrical and Electronics Engineers (IEEE) 802.11 family of standards known as Wi-Fi®, LoRa®/LoRaWAN® LPWAN standards, etc.), IEEE 802.15.4 family of standards, peer-to-peer (P2P) networks, 3rd Generation Partnership Project (3 GPP) standards for 4 G and 5 G wireless communication including: 3 GPP Long-Term evolution (LTE) family of standards, 3 GPP LTE Advanced family of standards, 3 GPP LTE Advanced Pro family of standards, 3 GPP New Radio (NR) family of standards, among others. In an example, the network interface device 620 may include one or more physical jacks (e.g., Ethernet, coaxial, or phone jacks) or one or more antennas to connect to the communications network 626. In an example, the network interface device 620 may include a plurality of antennas to wirelessly communicate using at least one of single-input multiple-output (SIMO), multiple-input multiple-output (MIMO), or multiple-input single-output (MISO) techniques. The term “transmission medium” shall be taken to include any intangible medium that is capable of storing, encoding or carrying instructions for execution by the machine 600, and includes digital or analog communications signals or other intangible medium to facilitate communication of such software.
The above detailed description includes references to the accompanying drawings, which form a part of the detailed description. The drawings show, by way of illustration, specific embodiments that may be practiced. These embodiments are also referred to herein as “examples.” Such examples may include elements in addition to those shown or described. However, the present inventors also contemplate examples in which only those elements shown or described are provided. Moreover, the present inventors also contemplate examples using any combination or permutation of those elements shown or described (or one or more aspects thereof), either with respect to a particular example (or one or more aspects thereof), or with respect to other examples (or one or more aspects thereof) shown or described herein.
All publications, patents, and patent documents referred to in this document are incorporated by reference herein in their entirety, as though individually incorporated by reference. In the event of inconsistent usages between this document and those documents so incorporated by reference, the usage in the incorporated reference(s) should be considered supplementary to that of this document; for irreconcilable inconsistencies, the usage in this document controls.
In this document, the terms “a” or “an” are used, as is common in patent documents, to include one or more than one, independent of any other instances or usages of “at least one” or “one or more.” In this document, the term “or” is used to refer to a nonexclusive or, such that “A or B” includes “A but not B,” “B but not A,” and “A and B,” unless otherwise indicated. In the appended claims, the terms “including” and “in which” are used as the plain-English equivalents of the respective terms “comprising” and “wherein.” Also, in the following claims, the terms “including” and “comprising” are open-ended, that is, a system, device, article, or process that includes elements in addition to those listed after such a term in a claim are still deemed to fall within the scope of that claim. Moreover, in the following claims, the terms “first,” “second,” and “third,” etc. are used merely as labels, and are not intended to impose numerical requirements on their objects.
The above description is intended to be illustrative, and not restrictive. For example, the above-described examples (or one or more aspects thereof) may be used in combination with each other. Other embodiments may be used, such as by one of ordinary skill in the art upon reviewing the above description. The Abstract is to allow the reader to quickly ascertain the nature of the technical disclosure and is submitted with the understanding that it will not be used to interpret or limit the scope or meaning of the claims. Also, in the above Detailed Description, various features may be grouped together to streamline the disclosure. This should not be interpreted as intending that an unclaimed disclosed feature is essential to any claim. Rather, inventive subject matter may lie in less than all features of a particular disclosed embodiment. Thus, the following claims are hereby incorporated into the Detailed Description, with each claim standing on its own as a separate embodiment. The scope of the embodiments should be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled.
This application is a continuation of U.S. Pat. Application Serial No. 17/449,350, filed Sep. 29, 2021, which is incorporated by reference herein in its entirety.
Number | Date | Country | |
---|---|---|---|
Parent | 17449350 | Sep 2021 | US |
Child | 17979479 | US |