Information processing device, information processing method, and program

Abstract

An information processing device 100 according to the present disclosure comprises: a case generation unit 121 that generates a virtual case so as to satisfy a preset constraint for some data included in a case; a prediction unit 122 that performs a preset prediction for the virtual case; and an evaluation unit 123 that evaluates, on the basis of the prediction for the virtual case, a condition set for data different from data corresponding to the constraint included in the virtual case. Accordingly, it is possible to assist decision making of a user.

Description

Information processing device, information processing method, and program

[0001] The present disclosure relates to an information processing device, an information processing method, and a program.

[0002] Predictions for input data are made using machine learning models in various fields. In this regard, Patent Literature 1 describes evaluation of machine learning models. Specifically, Patent Literature 1 calculates an evaluation value when parameters of the machine learning model are changed.

[0003] Japanese Patent Application Laid-Open No. 2022-068690

[0004] However, while the technology of Patent Document 1 evaluates a machine learning model, it is not possible to evaluate input data for predictions made by the machine learning model. In other words, when making predictions on input data using a machine learning model, if a desired prediction result is to be obtained from the input data, it is not possible to evaluate how to improve the input data. As a result, there is a problem in that the machine learning model cannot be effectively used.

[0005] Therefore, an object of the present disclosure is to solve the above-mentioned problem of being unable to effectively utilize machine learning models that make predictions on input data.

[0006] An information processing device according to an embodiment of the present disclosure includes: a case generation unit that generates virtual cases so as to satisfy preset constraints on some attributes included in cases; a prediction unit that makes a preset prediction for the virtual cases; and an evaluation unit that evaluates, based on the prediction for the virtual cases, conditions set for attributes other than attributes included in the virtual cases that fall under the constraints. An information processing method according to an embodiment of the present disclosure includes: generating virtual cases so as to satisfy preset constraints on some attributes included in cases; making a preset prediction for the virtual cases; and evaluating, based on the prediction for the virtual cases, conditions set for attributes other than attributes included in the virtual cases that fall under the constraints. A program according to an embodiment of the present disclosure includes: generating virtual cases so as to satisfy preset constraints on some attributes included in cases; making a preset prediction for the virtual cases; and evaluating, based on the prediction for the virtual cases, conditions set for attributes other than attributes included in the virtual cases that fall under the constraints.

[0007] With the above-described configuration, the present disclosure can effectively utilize a machine learning model that makes predictions on input data.

[0008] FIG. 1 is a block diagram showing a configuration of a first information processing device according to the present disclosure. FIG. 2 is a diagram showing a processing state by a first information processing device according to the present disclosure. FIG. 3 is a diagram showing a processing state by a first information processing device according to the present disclosure. FIG. 4 is a flowchart showing a processing operation of a first information processing device according to the present disclosure. FIG. 5 is a diagram illustrating a usage example of a second information processing device according to the present disclosure. FIG. 6 is a diagram illustrating a usage example of a second information processing device according to the present disclosure. FIG. 7 is a block diagram showing a hardware configuration of a third information processing device according to the present disclosure. FIG. 8 is a block diagram showing a configuration of a third information processing device according to the present disclosure.

[0009] First Embodiment The present disclosure will be described with reference to the drawings, which may be relevant to any of the embodiments.

[0010] [Configuration] In this embodiment, the information processing device 10 is used to make predictions on input data using a prediction model, which is a machine learning model. In particular, the information processing device 10 is used to evaluate improvements to the input data so that the desired prediction result can be obtained using the prediction model. For example, in this embodiment, a case will be described in which the prediction model predicts the probability of receiving a loan in a loan screening process, such as a mortgage loan, based on the attributes of a subject. The information processing device 10 can then predict the probability of receiving a loan using the prediction model based on the attributes of the subject, and can further output the conditions for obtaining the loan.

[0011] However, the machine learning model used to make predictions in the information processing device 10 is not necessarily limited to use in predicting the loan probability described above, and may be used in any prediction. Therefore, the information processing device 10 can be used to evaluate the improvement of any input data.

[0012] The information processing device 10 is composed of one or more information processing devices each including a calculation device and a storage device. As shown in FIG. 1 , the information processing device 10 includes an input unit 11, a virtual case generation unit 12, a prediction unit 13, a condition evaluation unit 14, and a condition output unit 15. The functions of the input unit 11, the virtual case generation unit 12, the prediction unit 13, the condition evaluation unit 14, and the condition output unit 15 can be realized by the calculation device executing a program for realizing each function stored in the storage device. The information processing device 10 also includes a model storage unit 16 and a case storage unit 17. The model storage unit 16 and the case storage unit 17 are each composed of a storage device. Each component will be described in detail below. FIG. 2 shows the flow of data and processing in the information processing device 10.

[0013] The input unit 11 accepts input of "constraints" D1 related to some data (attributes) from the input data serving as examples. In this embodiment, the input data serving as examples includes multiple attributes of a person undergoing loan screening, and examples of the attributes are the person's "gender, age, occupation, annual income, and debt." In this case, some attributes become "constraints," and in this embodiment, as shown in FIG. 2, "gender, age" are set as "constraints" D1. Specifically, a case where "gender: male, age: 35" is set as a constraint will be described as an example.

[0014] Furthermore, the input unit 11 accepts input of a "target" D2 to be achieved by prediction for input data, which is a case, as described below. For example, in this embodiment, as shown in FIG. 2, a loan probability of 80% (0.8) of the possibility of obtaining a loan is accepted as the target D2 as the target to be achieved for the loan probability. This target D2 will be used in the achievement determination process 15'. The input unit 11 may also accept input of a target value for each case (e.g., a loan probability > 0.70) when making predictions for virtual cases, as described below.

[0015] Based on the "constraint" D1 input as described above, the virtual case generation unit 12 (case generation unit) generates input data, which is a virtual case D3, that satisfies the constraint D1. That is, in this embodiment, a virtual case D3 is generated in which the subject's attributes satisfy the constraints "gender: male, age: 35 years old" and the other attributes "occupation, annual income, and debt" take arbitrary values. In this case, the virtual case generation unit 12 generates multiple virtual cases D3 using a generative model 12' generated using input data that is an actual case and stored in the model storage unit 16, as shown in FIG. 2 , and stores the multiple virtual cases D3 in the case storage unit 17. For example, when there are multiple variables, the virtual case generation unit 12 generates multiple virtual cases by predicting the conditional probability distribution of the remaining variables (occupation, annual income, debt) using a specific variable (constraint: gender, age) as a condition, and sampling values ​​according to this distribution, as shown below: P(occupation, annual income, debt | gender = male, age = 35)

[0016] As a more specific example, as shown in FIG. 3 , the virtual case generation unit 12 masks (leaves blank) a portion of a table representing each attribute of a subject, predicts a conditional probability distribution for the masked portion using the unmasked portion as a condition, and samples values ​​according to this distribution to generate a virtual case supplemented with the value of the masked portion. In this case, by training a machine learning model using a single neural network that outputs a probability distribution for any combination as the generative model 12′, virtual cases can be easily generated using a single model. In other words, the generative model 12′ may be pre-trained to generate values ​​for one or more explanatory variables (constraints) when the values ​​of the other explanatory variables are set. For example, this can be achieved using techniques such as VAEAC (VAE arbitrary conditioning), ACFlow, and ACE.

[0017] By generating the virtual case D3 in this way, it is possible to generate cases that are lacking in the actual cases. In other words, by generating virtual cases that do not correspond to actual cases, the reliability of improving the input data increases, as will be described later.

[0018] The prediction unit 13 performs a preset prediction for the virtual case D3 generated as described above. In this embodiment, the prediction unit 13 outputs a loan probability, which is a prediction D4 for the virtual case D3, using a prediction model 13′ that is generated using input data that is an actual case and the loan results thereof and that is stored in the model storage unit 16, as shown in FIG.

[0019] The condition evaluation unit 14 (evaluation unit) evaluates "conditions" set for data different from the constraints included in the above-mentioned hypothetical case D3 based on a prediction D4 for the hypothetical case D3. That is, as shown in FIG. 2, the condition evaluation unit 14 acquires "conditions" D5 (recourse) set by one or more combinations of data "occupation, annual income, and debt" other than the constraint "gender: male, age: 35" in the hypothetical case D3, and performs a condition evaluation process 14' to evaluate the "conditions" D5 based on the prediction D4. Examples of "conditions" D5 include "occupation = full-time employee AND annual income > 10 million yen," "occupation = unemployed," and "annual income > 6 million yen AND debt 8 million yen," as shown in FIG. 2.

[0020] In this case, the condition evaluation unit 14 may acquire the "condition" D5 in various ways. For example, the condition evaluation unit 14 may use the value of each piece of data randomly selected from training data, which are actual cases, as the condition. Alternatively, the condition evaluation unit 14 may train a random forest using, as (x, y), a combination of the above-mentioned hypothetical case D3 and a flag indicating that a target value for each case (e.g., a loan probability of >0.70) set in advance for the prediction D4 has been achieved, and extract each path of the decision tree included in the random forest as the "condition" D5. Note that the "condition" D5 may be generated by any method or acquired from another device.

[0021] Here, an example of a method for generating the "condition" D5 by the condition evaluation unit 14 will be further described. The condition evaluation unit 14 may randomly generate an atomic condition, which is a simple condition that does not include "AND," such as "feature 1 > X." For example, the condition evaluation unit 14 generates an atomic condition by randomly selecting one of the data items that can be set in the condition D5 as a feature, randomly selecting an inequality sign, and then randomly selecting a threshold X. The condition evaluation unit 14 may also generate a composite condition by combining multiple atomic conditions, for example, by combining them with AND.

[0022] Then, as shown in FIG. 2 , the condition evaluation unit 14 calculates a matching ratio D6 and a matching-time predicted average D7 for each condition D5 in the condition evaluation process 14′. Here, the matching ratio D6 represents the proportion of virtual cases D3 that satisfy the condition D5 among all virtual cases D3. For example, if there are 1,000 virtual cases D3 and 50 virtual cases D3 that satisfy a certain condition D5, the matching ratio D6 is 50 / 1,000 = 0.05. Furthermore, the matching-time predicted average D7 represents a statistic of the loan probability, which is the prediction D4 of the virtual cases D3 that satisfy the condition D5. In this embodiment, the statistic is an average value. For example, the matching-time predicted average D7 is calculated by adding up the loan probability, which is the prediction D4 of each virtual case D3 that satisfies the condition D5, and dividing by the number of applicable virtual cases D3. In addition, the predicted average D7 at the relevant time is not limited to being the average value of the loan probability, which is the prediction D4 of the virtual case D3 that meets the condition D5, but may be any value that is a statistical value calculated from the value of such prediction D4, such as the most frequent value of the prediction D4 of the virtual case D3 that meets the condition D5.

[0023] As shown in FIG. 2 , the condition evaluation unit 14 evaluates each condition D5 based on the relevant proportion D6 and the relevant-time predicted average D7 calculated as described above. At this time, when the “target” D2 to be achieved by the prediction result is input as described above, the condition evaluation unit 14 performs an achievement determination process 15′ to determine whether the relevant-time predicted average D7 satisfies the target D2, assigns a predicted target achievement flag (YES) to the satisfied condition D5, and outputs D9 the flagged condition. For example, if the target D2 is set to a relevant-time predicted average > 0.80, the condition evaluation unit 14 outputs a condition D5 in which the relevant-time predicted average D7 exceeds the target D2. At this time, the condition evaluation unit 14 may further evaluate the relevant condition D5 more highly the higher the value of the relevant-time predicted average D7. For example, if there are multiple conditions D5 that satisfy the target D2, the condition evaluation unit 14 may select and output only the condition D5 with the highest value of the relevant-time predicted average D7. Furthermore, the condition evaluation unit 14 may further evaluate the relevant condition D5 more highly the higher the relevant-time predicted average D7 and the relevant proportion D6. For example, if there are multiple conditions D5 for which the corresponding time predicted average D7 is approximately the same value, the condition D5 with the higher corresponding ratio D6 may be evaluated more highly. Alternatively, weights may be assigned to the corresponding time predicted average D7 and the corresponding ratio D6, and a statistical value of the corresponding time predicted average D7 and the corresponding ratio D6 may be calculated taking these weights into consideration, and the corresponding condition D5 may be evaluated more highly the higher this statistical value is.

[0024] Here, the condition evaluation unit 14 (condition generation unit) may exploratory generate and acquire the "condition" D5 by taking into account the relevant proportion D6 and the relevant-time predicted average D7 described above. In this case, the condition evaluation unit 14 may generate the condition D5 so that either or both of the relevant proportion D6 and the relevant-time predicted average D7 are high. That is, the condition evaluation unit 14 calculates the relevant proportion D6 and the relevant-time predicted average D7 for each of multiple randomly generated composite conditions, as described above, and generates the condition D5 so that either or both are high. As a more specific example, first, the randomly generated composite conditions are added to a list. Then, composite conditions that increase the relevant proportion D6 and the relevant-time predicted average D7 are extracted from the list. Furthermore, new composite conditions are generated by replacing some of the extracted composite conditions with newly randomly generated atomic conditions, and the new composite conditions are added to the list. This process is then repeated until the list exceeds a predetermined threshold, thereby generating conditions. The process of extracting the composite conditions that result in a high relevant ratio D6 and relevant time predicted average D7 can be realized by selecting composite conditions that are not Pareto dominated by other composite conditions.

[0025] The condition output unit 15 outputs the highly evaluated condition D5 as described above. For example, the condition output unit 15 outputs the most highly evaluated condition D5 among the conditions D5 that satisfy the constraint D1 in the attributes of the subject who requested the prediction of the loan probability. An example of output by the condition output unit 15 is shown in FIG. 4. FIG. 4 graphically illustrates the condition D5 for "annual income and debt" that satisfies the constraint D1 (gender: male, age: 35 years old) and the target D2, which is the predicted average at the time of the loan, on a coordinate plane with "annual income and debt" on the vertical and horizontal axes. In this case, if an actual case of a subject "gender: male, age: 35, occupation: temporary staffing, annual income: 3 million, debts: 5 million" is input as input data, and the predicted loan probability using prediction model 13' is lower than target D2 and the subject is rejected, information A of the subject is displayed in the rejected area on the coordinate plane, as well as information B of the most highly rated conditions that satisfy the same constraints as the subject, "gender: male, age: 35, occupation: temporary staffing, annual income: 6 million, debts: 2 million," as shown in Figure 4. In this case, it is even better to display the difference between information A of the subject and information B of conditions, "annual income: 6 million (+300), debts: 3 million (-200)."

[0026] This allows the subject to know how to improve their own attributes. At this time, the output conditions are based on virtual cases that have been judged to have good evaluations, so they are highly reliable. As a result, the prediction model can be used effectively.

[0027] [Operation] Next, a description will be given of the operation of the above-described information processing device 10. Here, as an example, a case will be described in which, as described above, conditions for obtaining a loan are evaluated in a situation in which the probability of obtaining a loan such as a mortgage is predicted based on the attributes of a subject.

[0028] The information processing device 10 accepts input of "constraints" D1 relating to some attributes of the input data serving as examples (step S1 in FIG. 5). For example, in this embodiment, of the target person's attributes "gender, age, occupation, annual income, and debt," as shown in FIG. 2, "gender = male, age = 35 years old" is accepted as the "constraint" D1. The information processing device 10 also accepts input of a "goal" D2 to be achieved by the prediction (step S1 in FIG. 5). For example, in this embodiment, as shown in FIG. 2, a loan probability of 80% (0.8) is accepted as the goal D2 to be achieved for the loan probability.

[0029] Next, the information processing device 10 generates a virtual case D3 that satisfies the constraint D1 (step S2 in FIG. 5). In this embodiment, as shown in FIG. 2, a plurality of virtual cases D3 are generated using a generative model 12' generated using input data that is an actual case. For example, as shown in FIG. 2, a plurality of virtual cases are generated by setting "gender = male, age = 35" as constraints and sampling values ​​of other attributes "occupation, annual income, debt" from a probability distribution that is conditioned by the constraints.

[0030] Next, the information processing device 10 makes a prediction for the generated hypothetical case D3 (step S3 in FIG. 5). In this embodiment, as shown in FIG. 2, the information processing device 10 outputs a loan probability, which is a prediction D4 for the hypothetical case D3, using a prediction model 13′ generated using input data that is an actual case and the loan results thereof.

[0031] Next, the information processing device 10 generates a condition D5 set for an attribute different from the constraints included in the virtual case D3 (step S4 in FIG. 5). In this embodiment, as shown in FIG. 2, the information processing device 10 generates a "condition" D5 (recourse) set by one or a combination of attributes "occupation, annual income, debt" other than the constraint "gender: male, age: 35" in the virtual case D3. Note that the information processing device 10 may acquire a preset condition D5, or may acquire a condition D5 generated by another device, etc.

[0032] Next, the information processing device 10 evaluates the generated conditions D5 based on the predictions D4 (step S5 in FIG. 5). In this embodiment, the information processing device 10 calculates a matching ratio D6 and a matching-time prediction average D7 for each condition D5. Here, the matching ratio D6 represents the ratio of virtual cases that satisfy the condition D5 among all virtual cases D3. Furthermore, the matching-time prediction average D7 represents a statistic of the loan probability, which is the prediction D4 of the virtual cases D3 that match the condition D5. In this embodiment, the statistic is an average value.

[0033] 2, the information processing device 10 evaluates the condition D5 for which the relevant time predicted average D7 satisfies the target loan probability D2 input in advance, and outputs the highly evaluated condition D5 (step S6 in FIG. 5). At this time, the information processing device 10 evaluates the condition more highly as the relevant ratio D6 and the relevant time predicted average D7 are higher, and outputs, for example, the most highly evaluated condition D5.

[0034] Thereafter, when the attributes of the subject requesting a loan probability prediction are input, the information processing device 10 predicts the loan probability based on the subject's attributes. At this time, as shown in FIG. 4 , the information processing device 10 outputs information A of the predicted results of the subject's attributes and information B of the most highly evaluated condition that satisfies the constraints. This allows the subject to know how to improve their attributes. At this time, the output conditions are based on hypothetical cases that have been judged to have good evaluations, and therefore are highly reliable. As a result, the prediction model can be effectively utilized and the subject can be encouraged to make better decisions.

[0035] Second Embodiment Next, a second embodiment of the present disclosure will be described with reference to the drawings. In this embodiment, an example of application of the information processing device described in the above embodiment to the medical and healthcare fields will be described.

[0036] As shown in FIG. 6 , in this embodiment, the subject U inputs his / her own characteristic data (e.g., gender, age) as well as vital data (e.g., weight, walking time) to the information processing device 10 via a network, and the information processing device 10 predicts values ​​representing the subject U's health condition, such as blood pressure. In this case, the information processing device 10 sets the subject U's characteristic data as the above-mentioned constraints and generates virtual cases in which various values ​​are set for the vital data. The information processing device 10 then generates conditions for the virtual cases that are made up of data other than the constraints and evaluates these conditions. In this case, the information processing device 10 highly evaluates conditions that meet the target of a blood pressure value that is not high and is an appropriate value, among the conditions of the virtual cases that satisfy the constraints.

[0037] The information processing device 10 then outputs to the subject U recommended actions consisting of vital data that may provide conditions for improving the current blood pressure value and achieving the goal, as well as predicted future blood pressure values ​​that will be the goal, as shown in Fig. 6. As an example, the information processing device 10 outputs conditions (weight: 80 kg → 75 kg, walking time: 10 minutes → 30 minutes) as recommended actions, and outputs the time-dependent change in predicted blood pressure values, as shown in Fig. 7.

[0038] The above-described example of use of the information processing device 10 is merely an example, and the information processing device 10 may be used for any prediction.

[0039] Third Embodiment Next, a third embodiment of the present disclosure will be described with reference to the drawings. In this embodiment, an outline of the configuration of the information processing device described in the above embodiments is shown.

[0040] 8, the information processing device 100 of this embodiment is configured as a general information processing device, and is equipped with the following hardware configuration, for example: CPU (Central Processing Unit) 101 (arithmetic unit); ROM (Read Only Memory) 102 (storage device); RAM (Random Access Memory) 103 (storage device); programs 104 loaded into RAM 103; a storage device 105 storing the programs 104; a drive device 106 for reading and writing from a storage medium 110 external to the information processing device; a communication interface 107 for connecting to a communication network 111 external to the information processing device; an input / output interface 108 for inputting and outputting data; and a bus 109 for connecting the various components.

[0041] 8 shows an example of the hardware configuration of the information processing device 100, and the hardware configuration of the information processing device is not limited to the above-described case. For example, the information processing device may be configured with only a part of the above-described configuration, such as excluding the drive device 106. Furthermore, the information processing device may use a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor), an MPU (Micro Processing Unit), an FPU (Floating Point Number Processing Unit), a PPU (Physics Processing Unit), a TPU (Tensor Processing Unit), a quantum processor, a microcontroller, or a combination thereof, instead of the above-described CPU.

[0042] The information processing apparatus 100 can be equipped with the case generation unit 121, prediction unit 122, and evaluation unit 123 shown in FIG. 9 by having the CPU 101 acquire and execute the program group 104. The program group 104 is stored in advance in the storage device 105 or the ROM 102, for example, and is loaded into the RAM 103 and executed by the CPU 101 as needed. The program group 104 may be supplied to the CPU 101 via the communication network 111, or may be stored in advance in the storage medium 110, with the drive device 106 reading out the program and supplying it to the CPU 101. However, the case generation unit 121, prediction unit 122, and evaluation unit 123 described above may be constructed using dedicated electronic circuits for realizing such means.

[0043] The case generation unit 121 generates virtual cases so as to satisfy preset constraints on some attributes included in the cases. The prediction unit 122 makes a preset prediction for the virtual cases. The evaluation unit 123 evaluates conditions set for attributes included in the virtual cases that are different from the attributes corresponding to the constraints, based on the prediction for the virtual cases. In this case, the evaluation unit 123 calculates preset statistics of predictions for virtual cases including attributes corresponding to the conditions, and evaluates the conditions based on the statistics. The evaluation unit 123 also evaluates the conditions using a proportion of virtual cases that satisfy the conditions.

[0044] With the above-described configuration, the present disclosure generates virtual cases that satisfy constraints, and evaluates the conditions of the virtual cases according to the predictions of the virtual cases. This improves the reliability of the cases that correspond to the evaluated conditions, and allows for the effective use of prediction models.

[0045] In addition, at least one of the functions of the above-mentioned case generation unit 121, prediction unit 122, and evaluation unit 123 may be executed by an information processing device installed and connected anywhere on the network, that is, they may be executed by so-called cloud computing.

[0046] The above-described program can be stored and supplied to a computer using various types of non-transitory computer-readable media. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (e.g., flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (e.g., magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-RWs, and semiconductor memories (e.g., mask ROMs, PROMs (Programmable ROMs), EPROMs (Erasable PROMs), flash ROMs, and RAMs (Random Access Memory)). The program can also be supplied to a computer by various types of transitory computer-readable media. Examples of transitory computer-readable media include electrical signals, optical signals, and electromagnetic waves. The transitory computer-readable media can be supplied to a computer via wired communication paths such as electric wires and optical fibers, or via wireless communication paths.

[0047] Although the present disclosure has been described above with reference to the above-described embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. Furthermore, each of the above-described embodiments can be combined with other embodiments as appropriate.

[0048] <Supplementary Notes> Some or all of the above embodiments can be described as in the following supplementary notes. Below, an outline of the configurations of an information processing device, an information processing method, and a program according to the present disclosure will be described. However, the present disclosure is not limited to the following configurations. (Supplementary Note 1) An information processing device comprising: a case generation unit that generates virtual cases so as to satisfy preset constraints on some attributes included in cases; a prediction unit that makes a preset prediction for the virtual cases; and an evaluation unit that evaluates, based on the prediction for the virtual cases, a condition set for an attribute other than an attribute included in the virtual cases that satisfies the constraint. (Supplementary Note 2) The information processing device according to Supplementary Note 1, wherein the evaluation unit evaluates the condition based on the prediction for the virtual cases that include an attribute that satisfies the condition. (Supplementary Note 3) The information processing device according to Supplementary Note 2, wherein the evaluation unit calculates preset statistics of the prediction for the virtual cases that include an attribute that satisfies the condition, and evaluates the condition based on the statistics. (Supplementary Note 4) The information processing device according to Supplementary Note 3, wherein the evaluation unit, when the prediction is output with a probability that the desired result will occur, evaluates the condition higher the higher the statistical amount. (Supplementary Note 5) The information processing device according to Supplementary Note 1, wherein the evaluation unit evaluates the condition further using a proportion of the hypothetical cases that satisfy the condition. (Supplementary Note 6) The information processing device according to Supplementary Note 5, wherein the evaluation unit evaluates the condition higher the higher the proportion. (Supplementary Note 7) The information processing device according to Supplementary Note 4, wherein the evaluation unit evaluates the condition higher the proportion of the hypothetical cases that satisfy the condition and the statistical amount are higher. (Supplementary Note 8) The information processing device according to Supplementary Note 7, comprising a condition generation unit that generates a new condition so that at least one of the proportion of the hypothetical cases that satisfy the condition and the statistical amount is higher.(Supplementary Note 9) The information processing device according to Supplementary Note 1, wherein the case generation unit generates the virtual cases using a machine learning model that is pre-trained to generate values ​​of other explanatory variables when values ​​of one or more explanatory variables are set. (Supplementary Note 10) An information processing method that generates virtual cases to satisfy preset constraints on some attributes included in cases, makes preset predictions for the virtual cases, and evaluates conditions set for attributes included in the virtual cases other than the attributes that satisfy the constraints, based on the predictions for the virtual cases. (Supplementary Note 11) The information processing method according to Supplementary Note 10, wherein the condition is evaluated based on the prediction for the virtual cases that include attributes that satisfy the condition. (Supplementary Note 12) The information processing method according to Supplementary Note 11, wherein preset statistics of the predictions for the virtual cases that include attributes that satisfy the condition are calculated, and the condition is evaluated based on the statistics. (Supplementary Note 13) The information processing method according to Supplementary Note 10, wherein the condition is evaluated further using a proportion of the virtual cases that satisfy the condition. (Supplementary Note 14) A computer-readable storage medium storing a program for causing a computer to execute the following processes: generating a virtual case so as to satisfy preset constraints on some attributes included in a case; making a preset prediction for the virtual case; and evaluating, based on the prediction for the virtual case, conditions set for attributes other than the attributes included in the virtual case that fall under the constraints.

[0049] REFERENCE SIGNS LIST 10 Information processing device 11 Input unit 12 Virtual case generation unit 13 Prediction unit 14 Condition evaluation unit 15 Condition output unit 16 Model storage unit 17 Case storage unit 100 Information processing device 101 CPU 102 ROM 103 RAM 104 Program group 105 Storage device 106 Drive device 107 Communication interface 108 Input / output interface 109 Bus 110 Storage medium 111 Communication network 121 Case generation unit 122 Prediction unit 123 Evaluation unit

Claims

1. a case generation unit that generates virtual cases so as to satisfy predetermined constraints on some attributes included in the cases; a prediction unit that performs a preset prediction for the virtual case; an evaluation unit that evaluates a condition set for an attribute that is different from an attribute that corresponds to the constraint and is included in the virtual case based on the prediction for the virtual case; An information processing device comprising:

2. 2. The information processing device according to claim 1, the evaluation unit evaluates the condition based on the prediction for the virtual case including an attribute corresponding to the condition. Information processing device.

3. 3. The information processing device according to claim 2, the evaluation unit calculates a predetermined statistic of the prediction for the virtual case including an attribute corresponding to the condition, and evaluates the condition based on the statistic. Information processing device.

4. 4. The information processing device according to claim 3, When the prediction is output with a probability that the desired result will occur, the evaluation unit evaluates the condition higher as the statistic increases. Information processing device.

5. 2. The information processing device according to claim 1, The evaluation unit further evaluates the condition using a ratio of the virtual cases that satisfy the condition. Information processing device.

6. 6. The information processing device according to claim 5, The evaluation unit evaluates the condition higher as the ratio increases. Information processing device.

7. 5. The information processing device according to claim 4, the evaluation unit evaluates the condition higher as the ratio of the virtual cases that satisfy the condition and the statistical amount increase; Information processing device.

8. 8. The information processing device according to claim 7, a condition generating unit that generates new conditions so that at least one of a rate of virtual cases that satisfy the conditions and the statistics is increased; Information processing device.

9. Generate virtual cases so that some attributes included in the cases satisfy predetermined constraints; making a preset prediction for the hypothetical case; Evaluating a condition set for an attribute that is different from the attribute corresponding to the constraint included in the virtual case based on the prediction for the virtual case; Information processing methods.

10. Generate virtual cases so that some attributes included in the cases satisfy predetermined constraints; making a preset prediction for the hypothetical case; Evaluating a condition set for an attribute that is different from the attribute corresponding to the constraint included in the virtual case based on the prediction for the virtual case; A program that causes a computer to execute a process.