Information processing device, information processing method, and program

Abstract

To provide an information processing apparatus, an information processing method, and a program that can prevent a situation in which selection of a machine learning model requires time.SOLUTION: An information processing apparatus comprises: leaning means that generates a plurality of machine leaning models different in learning method or condition for a parameter related to the generation; evaluation means that, based on a result of output from every machine learning model obtained by inputting data for evaluation to each of the machine learning models, of the plurality of machine learning models, in which machine learning is completed after the lapse of a first predetermined time from the start of the machine learning, evaluates the accuracy of each of the machine learning models; and model selection means that selects one machine learning model based on a result of evaluation performed by the evaluation means.SELECTED DRAWING: Figure 6

Description

[Technical field]

[0001] An embodiment of the present invention relates to an information processing device, an information processing method, and a program. [Background technology]

[0002] Conventionally, for example, in stores such as supermarkets, mass retailers, and convenience stores, future sales such as today's sales are predicted based on past sales records. In addition, machine learning is performed using past sales records to create a machine learning model (trained model), and the created machine learning model is used to predict sales, etc.

[0003] Incidentally, there are a number of learning methods for machine learning, and in order to determine which method is more appropriate, a number of machine learning models are actually created and the learning results are checked. Even if the same learning method is used, the learning results may vary depending on the parameters used, so in this case, it is preferable to actually create and check a number of machine learning models. However, depending on the learning method and the parameters, it may take a considerable amount of time to obtain the learning results. In such a case, it may take a considerable amount of time to select one machine learning model from the multiple machine learning models. Summary of the Invention [Problem to be solved by the invention]

[0004] The problem to be solved by the present invention is to provide an information processing device, an information processing method, and a program that are capable of selecting a machine learning model without spending a lot of time. [Means for solving the problem]

[0005] An information processing device according to an embodiment includes a learning means for creating a plurality of machine learning models having different learning methods or parameter conditions related to the creation, an evaluation means for evaluating accuracy of each of the machine learning models based on an output result for each of the machine learning models obtained by inputting evaluation data into each of the machine learning models for which the machine learning has been completed after a first predetermined time has elapsed since the start of machine learning, and a model selection means for selecting one of the machine learning models based on an evaluation result of the evaluation means. A learning termination means for terminating the machine learning for the machine learning model in which the machine learning has not been terminated even after a second predetermined time that is longer than the first predetermined time has elapsed since the start of the machine learning; Equipped with. [Brief description of the drawings]

[0006] [Figure 1] FIG. 1 is a diagram showing an information processing system according to an embodiment. [Diagram 2] FIG. 2 is an explanatory diagram showing the mechanism of machine learning. [Diagram 3] FIG. 3 is a block diagram showing a hardware configuration of the information processing device. [Figure 4] FIG. 4 is a memory map showing the configuration of the sales management section. [Diagram 5] FIG. 5 is a functional block diagram showing a functional configuration of the information processing device. [Figure 6] FIG. 6 is a flowchart showing the flow of a control process related to the selection of a machine learning model in an information processing device. [Figure 7] FIG. 7 is a flowchart showing the flow of a control process related to the output of prediction data in an information processing device. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0007] Hereinafter, an embodiment will be described with reference to the drawings. In the embodiment, a headquarters server (cloud server) installed in a headquarters, for example, which selects a machine learning model to be used for sales forecasting in a store such as a supermarket, mass retailer, or convenience store, will be described as an example of an information processing device. Note that the present invention is not limited to the embodiment.

[0008] Machine learning is a computer algorithm that automatically improves by learning from experience. It is a well-known technology that uses large amounts of learning data to learn patterns and predicts (infers) unknown data using the learning results. For example, it can analyze a person's preferences based on large amounts of data on product purchase history, and provide information that suits that person based on information that is likely to interest or concern them. It can also be used for OCR, which learns from large amounts of character data and infers characters written by humans.

[0009] A machine learning model that performs machine learning is usually used in multiple patterns, and each machine learning model has a different learning method and / or input parameter conditions. This is because the accuracy of each machine learning model differs depending on the learning method and type of parameters used, so it is possible to select a machine learning model with high accuracy for the input data.

[0010] The headquarters server of the embodiment uses the most accurate machine learning model among the machine learning models created by machine learning to predict sales of stores and products. Specifically, past sales data associated with sales performance is input and the most accurate machine learning model is selected. The headquarters server of the embodiment also uses the machine learning model to infer future sales data.

[0011] Fig. 1 is a diagram showing an information processing system according to an embodiment. As shown in Fig. 1, the information processing system 10 includes a headquarters server 1, a store server 3, a POS (Point of Sales) terminal 5, etc. The headquarters server 1 and the store server 3 are communicatively connected via a network N such as the Internet or a dedicated line. One or more store servers 3 are connected to the headquarters server 1 via the network N. The store server 3 and the POS terminal 5 are communicatively connected within a store T via a communication line L such as a LAN (Local Area Network). One or more POS terminals 5 are connected to the store server 3 via the communication line L.

[0012] The POS terminal 5 executes product registration processing and payment processing for products sold at the store T. The product registration processing includes processing for reading a symbol such as a barcode attached to a product, and reading and storing product information of the product (product name, product price, etc.) based on a product code that identifies the product contained in the symbol. The payment processing includes processing for making payment for products that have been registered. Specifically, the processing includes processing for calculating the total price of the registered products, calculating the amount of change if there is any, and issuing a receipt on which payment information such as product information, the total price, and the amount of change are printed.

[0013] In addition, the POS terminal 5 performs a process of transmitting sales information (including product information and payment information) relating to the product for which payment has been made to the store server 3 sequentially or at a predetermined timing.

[0014] The store server 3 receives the sales information transmitted by the POS terminals 5. The store server 3 aggregates the sales information received from each POS terminal 5 and cumulatively stores it in the sales information section 31. In addition, the store server 3 transmits the sales information stored in the sales information section 31 to the head office server 1 periodically (for example, once a day).

[0015] The headquarters server 1 stores the sales information received from each store server 3. The headquarters server 1 cumulatively stores the sales information received from each store server 3. The headquarters server 1 also classifies the received sales information into various units and stores it as information by type. For example, the headquarters server 1 stores the sales information by store T. For example, the headquarters server 1 also stores the sales information by product. For example, the headquarters server 1 also stores the sales information by period (by year, month, week, day, time period, day of the week, season, etc.). The headquarters server 1 also stores the sales information in various other units.

[0016] The headquarters server 1 periodically performs machine learning to predict future sales based on past sales information. FIG. 2 is an explanatory diagram showing the mechanism of machine learning in the headquarters server 1. As shown in FIG. 2, the headquarters server 1 creates (generates) multiple machine learning models using past data. In the embodiment, three types of machine learning models, Model 1, Model 2, and Model 3, are created. Each machine learning model has a different learning method and / or parameter conditions related to the creation of the machine learning model. Specifically, in the creation of each machine learning model, a set of different parameters including information on the weather, temperature, and implemented measures and the sales information of that day is accumulated every day. The headquarters server 1 performs machine learning from the accumulated sets of daily sales information and parameter information to create each machine learning model capable of outputting a predetermined prediction result (sales forecast) showing a trend for each machine learning model. Then, the headquarters server 1 compares prediction results (hereinafter also referred to as learning results) obtained by inputting evaluation data such as sales performance information related to the past sales information to each of the created machine learning models, and selects one machine learning model whose learning result is closest to the sales performance information (i.e., has high accuracy). The evaluation data is a part of the sales information that was not input when the machine learning model was created. Much of the sales information is input when creating a machine learning model, and a portion of the sales information is input into evaluation data created as evaluation data.

[0017] Next, the headquarters server 1 inputs past sales data related to the data for which sales forecast is desired into the selected machine learning model, thereby predicting future sales (for example, today's sales).

[0018] Furthermore, when a first predetermined time has elapsed since the start of machine learning for multiple machine learning models, the headquarters server 1 evaluates the learning results for the machine learning models for which machine learning has been completed at that time, compares them with sales performance information, and selects one machine learning model. Furthermore, after a certain period of time has elapsed, the headquarters server 1 again evaluates the machine learning models for which machine learning has been completed at that time, compares the sales performance information with the accuracy of the one machine learning model previously selected, and selects another one machine learning model with high accuracy.

[0019] In addition, for a machine learning model whose machine learning has not ended even after a second specified time (e.g., 24 hours from the start of machine learning) that is longer than the first specified time has elapsed after the first specified time has elapsed, the headquarters server 1 forcibly terminates machine learning for that machine learning model.

[0020] Such a headquarters server 1 inputs past sales information, for example, when store T closes (a specified time), and performs machine learning on the machine learning model. If there is a machine learning model for which machine learning has been completed at, for example, 5 a.m. the following day, when a first specified time has passed since the specified time, the headquarters server 1 extracts that machine learning model. The headquarters server 1 then evaluates the learning results of the machine learning models for which machine learning has been completed. Then, based on the evaluation of each machine learning model, it selects the machine learning model with the highest evaluation (accuracy).

[0021] Next, the headquarters server 1 inputs sales information for the same day of the week last week, for example, into the selected machine learning model to perform inference and predict today's sales.

[0022] Furthermore, when a certain time has elapsed after the first predetermined time has elapsed, if there is a new machine learning model for which machine learning has been completed, the headquarters server 1 extracts the machine learning model. Then, evaluation data is input to the machine learning model for which machine learning has been completed, and the output result obtained is evaluated. Then, the accuracy is compared with that of the previously selected machine learning model, and one machine learning model with higher accuracy is selected. If the previously selected machine learning model is selected again, inference is not performed again, and if a later selected machine learning model is selected, inference is redone using the machine learning model.

[0023] Furthermore, the headquarters server 1 forcibly ends machine learning for a machine learning model for which machine learning has not ended even after a second predetermined time has elapsed after the first predetermined time has elapsed.

[0024] From here, the hardware of the headquarters server 1 will be described. FIG. 3 is a block diagram showing the hardware configuration of the headquarters server 1. As shown in FIG. 3, the headquarters server 1 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a memory section 14, and the like. The CPU 11 is the main controller. The ROM 12 stores various programs. The RAM 13 expands programs and various data. The memory section 14 stores various programs. The CPU 11, the ROM 12, the RAM 13, and the memory section 14 are connected to one another via a bus 15. The CPU 11, the ROM 12, and the RAM 13 constitute a control section 100. That is, the control section 100 executes the control processing of the headquarters server 1, which will be described later, by the CPU 11 operating in accordance with the control program stored in the ROM 12 and the memory section 14 and expanded in the RAM 13.

[0025] The RAM 33 includes a sales management unit 131, a learning end information unit 132, and an end flag unit 133. The sales management unit 131 cumulatively stores sales information for each store T received from each store server 3. The sales management unit 131 will be described later with reference to FIG. 4. The learning end information unit 132 stores model identification information that identifies (identifies) a machine learning model for which machine learning has ended, and information indicating the learning result of the machine learning model. The end flag unit 133 stores the state of the end flag. If there is a machine learning model for which machine learning has not yet ended, the end flag unit 133 stores the end flag "0". If machine learning has ended for all machine learning models, the end flag unit 133 stores the end flag "1".

[0026] The memory unit 14 is composed of a non-volatile memory such as a hard disk drive (HDD) or a flash memory that retains stored information even when the power is turned off, and includes a control program unit 141 that stores a control program for controlling the headquarters server 1, a model storage unit 142, a parameter unit 143, and a time storage unit 144. The model storage unit 142 stores a machine learning model that periodically performs machine learning. In the embodiment, as an example, three types of machine learning models, model 1, model 2, and model 3, are stored. The parameter unit 143 stores parameters input during machine learning of the machine learning model by model, day, and store. The parameters are, for example, data on daily customer demographics for each store T, information on daily events in the vicinity of each store T, date, day of the week, daily weather, and the like. The time storage unit 144 stores information on the time when sales information is input, a first predetermined time, a second predetermined time, and a fixed time.

[0027] Furthermore, the control unit 100 is connected to an operation unit 17 and a display unit 18 via a bus 15 and a controller 16. The controller 16 receives instructions from the control unit 100 and controls the operation unit 17 and the display unit 18. However, for convenience of explanation, the control performed by the controller 16 will be explained as being performed by the control unit 100. The operation unit 17 is, for example, a keyboard or a touch panel, and is operated by an operator of the headquarters server 1. The display unit 18 is, for example, a liquid crystal display, and displays information to the operator of the headquarters server 1.

[0028] The control unit 100 is also connected to a communication unit 19 via a bus 15. The communication unit 19 is connected to each store server 3 via a network N. The control unit 100 is also connected to a timer 20. The timer 20 keeps track of the current time, and measures a first predetermined time, a second predetermined time, and a fixed time.

[0029] From here, the sales management unit 131 will be described. The sales management unit 131 aggregates and cumulatively stores sales information for each store T received from each store server 3. FIG. 4 is a memory map showing the configuration of the sales management unit 131. As shown in FIG. 4, the sales management unit 131 includes a store sales management unit 1311, a product sales management unit 1312, a total sales management unit 1313, and the like. The store sales management unit 1311 stores, for example, past sales information for each store T, for example, by year. The store sales management unit 1311 similarly stores, for example, sales information for each store T, for example, by month, week, day, and the like. The product sales management unit 1312 stores sales information for each product (or by product classification), for example, by month, week, day, and the like. The total sales management unit 1313 stores sales information received from each store server 3, for example, by month, week, day, and the like. The sales management unit 131 also stores sales information compiled from other perspectives, for example, by month, week, or day.

[0030] From here, we will explain the functional configuration of the headquarters server 1. Fig. 5 is a functional block diagram showing the functional configuration of the headquarters server 1. As shown in Fig. 5, the control unit 100 of the headquarters server 1 functions as a learning means 101, an evaluation means 102, a model selection means 103, a learning termination means 104, and an inference means 105 by following the control programs stored in the ROM 12 or the control program unit 141 of the memory unit 14.

[0031] The learning means 101 has a function of creating multiple machine learning models with different learning methods or parameter conditions for creation. The sales information is a large amount of data, such as one year's worth of data, in which daily sales performance data and parameters related to the day are associated and accumulated.

[0032] The evaluation means 102 has a function of evaluating the accuracy of each of the machine learning models based on the output result of each machine learning model obtained by inputting evaluation data into each of the machine learning models for which machine learning has been completed after a first predetermined time has elapsed from the start of machine learning. Specifically, when model identification information of a machine learning model for which machine learning by the learning means 101 has been completed is stored in the learning completion storage unit 132 at the time when a first predetermined time measured by the timer 20 has elapsed from the start of machine learning, the evaluation means 102 inputs evaluation data into the machine learning model, and evaluates the accuracy of each of the machine learning models based on the obtained learning result (output result) for each machine learning model and performance data of the evaluation data for the learning result. When there are multiple machine learning models for which machine learning has been completed, the evaluation means 102 evaluates the accuracy of each of the machine learning models based on the obtained learning result for each machine learning model and performance data for the learning result.

[0033] The model selection means 103 has a function of selecting one machine learning model that has been most highly evaluated by the evaluation means 102. Specifically, the model selection means 103 selects one machine learning model that has been most highly evaluated by the evaluation means 102. More specifically, the model selection means 103 selects one machine learning model that has been evaluated by the evaluation means 102 to have the highest accuracy. When one machine learning model is evaluated, the model selection means 103 selects that machine learning model. When multiple machine learning models are evaluated, the model selection means 103 selects the machine learning model that has been most highly evaluated from among them.

[0034] The learning termination means 104 has a function of terminating machine learning for a machine learning model for which machine learning has not been completed even after a second predetermined time longer than the first predetermined time has elapsed. Specifically, the learning termination means 104 forcibly terminates machine learning for a machine learning model for which machine learning has not been completed at the time when the timer 20 measures the second predetermined time longer than the first predetermined time.

[0035] The inference means 105 has a function of outputting prediction data such as sales information using one of the machine learning models selected by the model selection means 103. Specifically, the inference means 105 inputs various conditions (parameters) to be predicted into one of the machine learning models selected by the model selection means 103, and acquires the output result of the sales forecast output by the machine learning model as prediction data. For example, the inference means 105 inputs parameters such as the day of the week, time zone, and weather conditions of the day for which sales are to be predicted into the machine learning model, and acquires the output result of the future sales forecast inferred from these parameters as prediction data.

[0036] From here, we will explain the control of the head office server 1. Fig. 6 is a flowchart showing the flow of control processing related to the selection of a machine learning model in the head office server 1. Fig. 7 is a flowchart showing the flow of control processing related to the output of prediction data in the head office server 1.

[0037] 6, the control unit 100 of the head office server 1 judges whether it is a predetermined time based on the timing of the timer 20 (S11). The predetermined time is, for example, a set time after the store T closes, and is stored in the time storage unit 144. If it is judged based on the timing of the timer 20 that it is the predetermined time (Yes in S11), the control unit 100 inputs sales information stored in the sales management unit 131 as past data for outputting forecast data (S12).

[0038] For example, when it is desired to predict today's sales of a certain store T, past sales information for the store T is retrieved and input from the store sales management unit 1311. When it is desired to predict sales of a certain product, sales information for the product is retrieved and input from the product sales management unit 1312. When it is desired to predict the combined sales of all stores T, sales information is retrieved and input from the total sales management unit 1313. In either case, as much sales information as possible (for example, one year's worth) is input, since the more data there is, the more accurate the machine learning model's output will be.

[0039] Next, the learning means 101 starts machine learning for each of the machine learning models (in the embodiment, model 1, model 2, and model 3) (S13). The parameters for model 1, model 2, and model 3 are different from each other. Then, the control unit 100 returns to S11. For the machine learning models for which machine learning has been completed, the control unit 100 stores in the learning end information unit 132 model identification information that identifies the machine learning model and the output of the machine learning learning result by the machine learning model.

[0040] Also, when it is determined that the predetermined time has not arrived (No in S11), the control unit 100 determines whether the first predetermined time has elapsed based on the counting of the predetermined time by the timer 20 and the first predetermined time stored in the time storage unit 144 (S21). The first predetermined time is a time determined in advance from the time when the machine learning model starts machine learning in S13, and is, for example, the time until a predetermined time in the early morning of the store T (for example, 5:00 AM). When it is determined that the first predetermined time has elapsed (Yes in S21), the control unit 100 determines whether there is a machine learning model that has finished machine learning at the time when the first predetermined time has elapsed since the start of machine learning (S22). Specifically, the control unit 100 determines whether there is a machine learning model that has finished machine learning based on whether model identification information is stored in the learning end information unit 132.

[0041] When it is determined that there is a machine learning model that has completed machine learning (Yes in S22), the control unit 100 determines whether all the machine learning models have completed machine learning (S23). In the embodiment, since three types of machine learning models, model 1, model 2, and model 3, are performing machine learning, if model identification information that identifies all (three types) of machine learning models is stored in the learning end information unit 132, the control unit 100 determines that all the machine learning models have completed machine learning. If model identification information that identifies all (three types) of machine learning models is not stored in the learning end information unit 132, the control unit 100 determines that all the machine learning models have not completed machine learning (i.e., there is a machine learning model that is still undergoing machine learning).

[0042] When it is determined that all the machine learning models have completed machine learning (Yes in S23), the control unit 100 stores a flag indicating that all the machine learning models have completed machine learning in the completion flag unit 133 (S24). For example, when it is determined that all the machine learning models have completed machine learning, a flag "1" indicating that all the machine learning models have completed machine learning is stored. Note that when it is determined that all the machine learning models have not completed machine learning, a flag "0" indicating that all the machine learning models have not completed machine learning is stored.

[0043] After the process of S24 or in S23, if it is determined that all machine learning models have not completed machine learning (No in S23), the control unit 100 then inputs evaluation data to the machine learning models for which machine learning has been completed and stored in the learning completion information unit 132 to extract the learning results (S25). Then, the evaluation means 102 compares the extracted learning results for each machine learning model with the performance data for the learning results to evaluate the accuracy of each machine learning model (S26). The performance data is a response (performance data) actually measured for the input past data, and in the case of the embodiment, it is, for example, performance data related to sales information for the past year. The closer the learning results are to the performance data, the higher the accuracy of the machine learning model is evaluated.

[0044] Next, the control unit 100 compares the multiple evaluated machine learning models to determine which machine learning model has a higher evaluation (accuracy) (S27). Then, the model selection means 103 selects the one machine learning model with the highest evaluation (S28).

[0045] Next, the control unit 100 judges whether the end flag stored in the end flag unit 133 is "1" (S29). If the end flag stored in the end flag unit 133 is not "1" (i.e., "0") (No in S29), the control unit 100 then judges whether a certain time stored in the end time unit 144 has elapsed since the control unit 100 selected one machine learning model (S30). The certain time is a time before the second predetermined time is clocked. The control unit 100 waits until the certain time has elapsed (No in S30), and if it is judged that the certain time has elapsed (Yes in S30), it judges whether the second predetermined time stored in the end time unit 144 has elapsed from the clocking of the predetermined time of the timer 20 (S31). The second predetermined time is a time longer than the first predetermined time, for example, 24 hours. For a machine learning model for which machine learning has not ended even after 24 hours have elapsed from the clocking of the predetermined time of the timer 20, there is little point in continuing machine learning any further, since the next day will come. The second predetermined time is thus set to a time at which it becomes less meaningful to continue machine learning from the time the timer 20 reaches the predetermined time. If it is determined that the second predetermined time has not elapsed from the time the timer 20 reaches the predetermined time (No in S31), the control unit 100 returns to S22 and determines whether the machine learning model that is still continuing machine learning has ended machine learning.

[0046] On the other hand, when it is determined that the second predetermined time has elapsed from the time counted by the timer 20 (Yes in S31), the learning termination means 104 forcibly terminates the machine learning in the machine learning model that is continuing the machine learning (S32). Then, the control unit 100 terminates the process. Also, when the termination flag stored in the termination flag unit 133 is "1" (Yes in S29), the control unit 100 terminates the process. Note that, when it is determined in S22 that there is no machine learning model that has completed the machine learning (No in S22), the control unit 100 performs the determinations of S30 and S31. That is, the control unit 100 waits for a certain period of time to elapse again and further determines whether there is a machine learning model that has completed the machine learning. That is, the control unit 100 determines whether there is a machine learning model that has completed the machine learning every time a certain period of time elapses.

[0047] Note that, once one machine learning model is selected in S28, and after determining No in S31, if it is determined again in S22 that there is a new machine learning model for which machine learning has been completed (Yes in S22), the control unit 100 evaluates the machine learning model in S26, and in S27 compares the evaluation of the machine learning model with the evaluation of the machine learning model previously selected in S28. Then, the control unit 100 selects one machine learning model with a high evaluation in S28.

[0048] From here, the control of inference using one machine learning model selected in S28 and outputting predicted data will be described. FIG. 7 is a flowchart showing the flow of control processing related to the output of predicted data in the headquarters server 1. As shown in FIG. 7, the control unit 100 judges whether one machine learning model has been selected in the processing of S28 (S41). The control unit 100 waits until one machine learning model has been selected (No in S41), and when it is judged that one machine learning model has been selected (Yes in S41), the control unit 100 inputs information on parameters, which are past data corresponding to the day for which prediction is desired, to the selected machine learning model (S42). The parameters input in S42 are, for example, parameters for the previous day. Also, for example, parameters for the same day of the week in the previous week. Also, for example, parameters for the same day of the previous month. Also, for example, parameters for the same day of the previous year.

[0049] Next, the inference means 105 causes the machine learning model to make an inference based on the input past data (S43). Then, the control unit 100 outputs future prediction data as the result of the inference by the machine learning model (S44). This prediction data is, for example, today's sales forecast information. Then, the control unit 100 ends the process.

[0050] As described above, the headquarters server 1 of the embodiment includes a learning means 101 that creates multiple machine learning models with different learning methods or parameter conditions related to creation, an evaluation means 102 that evaluates the accuracy of each of the multiple machine learning models based on the output results for each machine learning model obtained by inputting evaluation data into each of the machine learning models for which machine learning has been completed after a first predetermined time has elapsed since the start of machine learning, and a model selection means 103 that selects one machine learning model based on the evaluation result of the evaluation means 102.

[0051] The headquarters server 1 evaluates the outputs of the machine learning models that have completed machine learning after the first predetermined time has elapsed, and selects the single machine learning model with the highest evaluation. This makes it possible to select a single machine learning model without spending much time.

[0052] Although the embodiment of the present invention has been described above, this embodiment is presented as an example and is not intended to limit the scope of the invention. This new embodiment can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the gist of the invention. This embodiment and its modifications are included in the scope and gist of the invention, and are included in the scope of the invention and its equivalents described in the claims.

[0053] For example, in the embodiment, sales information in store T has been described as an example of input past data. However, the input past data is not limited to this, and may be past data of a type corresponding to the output to be obtained.

[0054] Furthermore, in the embodiment, the head office server 1 has been described as a cloud server, but the head office server 1 may be a server other than a cloud server.

[0055] In addition, in the embodiment, the headquarters server 1 has been described as an example of an information processing device, but when implementing a machine learning function in one store, for example, the store server 3 installed in the store T may be an information processing device.

[0056] In addition, the programs executed by the headquarters server 1 in the embodiment are provided in the form of files in an installable or executable format recorded on a computer-readable recording medium such as a CD-ROM, a flexible disk (FD), a CD-R, or a DVD (Digital Versatile Disk).

[0057] The program executed by the headquarters server 1 of the embodiment may be stored on a computer connected to a network such as the Internet and provided by being downloaded via the network. The program executed by the headquarters server 1 of the embodiment may be provided or distributed via a network such as the Internet.

[0058] Moreover, the programs executed by the head office server 1 of the embodiment may be provided by being pre-installed in a ROM or the like. [Explanation of symbols]

[0059] 1 Headquarters Server 3. Store Server 5. POS terminals 10. Information Processing Systems 11 CPU 12 ROM 13 RAM 14 Memory section 17 Control section 18 Display 19 Communications Department 20 Timer 31 Sales Information Department 33 RAM 100 Control section 101 Learning Tools 102 Evaluation methods 103 Model Selection Method 104 Means of Completing Learning 105 Reasoning means 131 Sales Management Department 132 Completion of Study Information Department 133 End Flag Section 142 Model Memory Section 143 Parameter section 144 Time storage section 1311 Store Sales Management Department 1312 Product Sales Management Department 1313 Total Sales Management Department [Prior art documents] [Patent documents]

[0060] [Patent Document 1] JP 2017-228086 A

Claims

1. A learning means for creating a plurality of machine learning models having different learning methods or parameter conditions related to the creation; an evaluation means for evaluating accuracy of each of the machine learning models based on an output result for each of the machine learning models obtained by inputting evaluation data into each of the machine learning models for which the machine learning has been completed after a first predetermined time has elapsed since the start of machine learning; A model selection means for selecting one of the machine learning models based on the evaluation result of the evaluation means; A learning termination means for terminating the machine learning for the machine learning model in which the machine learning has not been terminated even after a second predetermined time that is longer than the first predetermined time has elapsed since the start of the machine learning; An information processing device comprising:

2. the evaluation means evaluates accuracy of each of the machine learning models based on output results for each of the machine learning models obtained by inputting the evaluation data into the machine learning models for which machine learning by the learning means has been completed, and performance data for the output results, within a certain time period before the second predetermined time has elapsed since the machine learning models were evaluated; The model selection means selects the most evaluated machine learning model from the evaluated machine learning models and the selected machine learning model. The information processing device according to claim 1 .

3. an inference means for outputting prediction data inferred based on input past data using the machine learning model selected by the model selection means; The information processing device according to claim 1 , further comprising:

4. A learning process for creating a plurality of machine learning models having different learning methods or parameter conditions related to the creation; an evaluation step of evaluating accuracy of each of the machine learning models based on an output result for each of the machine learning models obtained by inputting evaluation data into each of the machine learning models for which the machine learning has been completed after a first predetermined time has elapsed since the start of machine learning; a model selection step of selecting one of the machine learning models based on the evaluation result of the evaluation step; a learning termination step of terminating the machine learning for the machine learning model for which the machine learning has not been terminated even after a second predetermined time that is longer than the first predetermined time has elapsed since the start of the machine learning; An information processing method comprising:

5. A computer as an information processing device A learning means for creating a plurality of machine learning models having different learning methods or parameter conditions related to the creation; an evaluation means for evaluating accuracy of each of the machine learning models based on an output result for each of the machine learning models obtained by inputting evaluation data into each of the machine learning models for which the machine learning has been completed after a first predetermined time has elapsed since the start of machine learning; A model selection means for selecting one of the machine learning models based on the evaluation result of the evaluation means; A learning termination means for terminating the machine learning for the machine learning model in which the machine learning has not been terminated even after a second predetermined time that is longer than the first predetermined time has elapsed since the start of the machine learning; A program to make it function as such.

Images