Information Processing Apparatus, Information Processing Method, and Program

Abstract

To verify a performance of a machine learning model.SOLUTION: An information processing device comprises a first generation section, a reception section, a second generation section, an estimation section, a first learning section and an evaluation section. The first generation section generates first data representing observation data at a time when a measure is not executed, to be inputted to a model for inferring an effect of the measure. The reception section receives a first parameter which is used in estimation processing for estimating observation data at a time when the measure is executed, and exerts influences upon the effect. The second generation section generates second data representing the observation data, which are estimated by the estimation processing, by executing the estimation processing using the first parameter. The estimation section uses the first data and the second data to estimate the effect. The first learning section uses learning data including the first data and the second data to train the model. The evaluation section evaluates a performance of the model by comparing the effect estimated by the estimation section with an effect estimated by the trained model.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] Embodiments of the present invention relate to an information processing apparatus, an information processing method, and a program.

Background Art

[0002] Techniques for applying a machine learning model to a causal effect estimation problem have been proposed. For example, using observational data corresponding to the execution result of a certain measure, the effect of the measure is estimated by a machine learning model.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Patent Document 3

Non-Patent Documents

[0004]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, in the prior art, it has sometimes been difficult to verify the performance of a machine learning model.

Means for Solving the Problems

[0006] The information processing apparatus according to the embodiment includes a first generation unit, a reception unit, a second generation unit, an estimation unit, a first learning unit, and an evaluation unit. The first generation unit generates first data representing observation data when a policy is not executed, which is input to a model for inferring the effect of the policy. The reception unit receives a first parameter that affects the effect and is used in an estimation process for estimating the observation data when the policy is executed. The second generation unit executes an estimation process using the first parameter and generates second data representing the observation data estimated by the estimation process. The estimation unit estimates the effect using the first data and the second data. The first learning unit learns the model using learning data including the first data and the second data. The evaluation unit evaluates the performance of the model by comparing the effect estimated by the estimation unit with the effect estimated by the learned model.

Brief Description of the Drawings

[0007]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Embodiments for Carrying Out the Invention

[0008] Hereinafter, with reference to the accompanying drawings, a preferred embodiment of the information processing apparatus according to the present invention will be described in detail.

[0009] The following mainly describes an example of a traffic control system that provides services such as traffic congestion prediction using traffic flow data (an example of observation data) and a machine learning model. The applicable systems are not limited to traffic control systems. Any system that inputs observation data into a machine learning model and infers the effects when implementing measures can be applied to any system. Observation data is not limited to traffic flow data as long as it is data that varies depending on whether or not measures have been implemented.

[0010] Many transportation operators and local governments are progressing with the collection of traffic big data, and the provision of congestion prediction services based on the data, dynamic pricing that varies tolls according to traffic (congestion) conditions, and the consideration of traffic measures such as traffic regulations in accordance with the concept of EBPM (Evidence Based Policy Making) (hereinafter referred to as "measures") have become full-fledged. One of the major purposes of providing congestion prediction services and dynamic pricing is to suppress congestion by dispersing traffic flow. When actually implementing measures such as providing congestion prediction services and dynamic pricing, it is expected that the driver of the vehicle in motion will change their behavior such as taking a detour, thereby achieving a congestion suppression effect. However, in order to accurately estimate this congestion suppression effect from traffic flow data, it is necessary to randomly implement measures and calculate the difference in traffic volume between when the measures are implemented and when they are not. However, comparative tests that randomly implement measures cannot always be conducted.

[0011] When measures cannot be randomly implemented, the necessity of implementing measures depends on factors such as the traffic conditions and weather around the target section, and these factors also affect the traffic volume of the target section. The problem of estimating the congestion suppression effect of measures by adjusting the influence of the traffic factors behind both the feasibility of implementing measures and the traffic volume of the target section can be regarded as a problem of causal effect estimation.

[0012] As described above, although techniques for applying a machine learning model to the problem of estimating causal effects have been proposed, it has sometimes been difficult to verify the performance (validity) of such a machine learning model. For example, a method for estimating a causal effect using a machine learning model has been shown to be valid under theoretical assumptions, but it is not guaranteed that actual data satisfies the assumptions. Therefore, it is difficult to confirm whether the effect estimated using the machine learning model is appropriate.

[0013] In the present embodiment, the performance of the machine learning model can be verified by comparing the effect of a measure estimated by an estimation process different from the machine learning model with the effect of the measure estimated by the machine learning model.

[0014] In the estimation process, parameters that are assumed to change by executing a measure are set. The parameters are, for example, the selection probabilities of each of a plurality of routes (roads). The estimation process is executed according to the set parameters (selection probabilities), and traffic flow data when the measure is executed is estimated. Further, the traffic flow data when the measure is executed is compared with the traffic flow data when the measure is not executed, and the effect of the measure is estimated. Then, the effect of the measure estimated by the estimation process is compared with the effect of the measure estimated by the machine learning model.

[0015] The machine learning model is a model that is learned to infer the effect when a measure is executed by inputting traffic flow data that can vary depending on whether the measure has been executed or not. For example, it can be verified that a machine learning model that obtains an effect with a small difference with respect to the effect estimated by the estimation process is a model that can estimate the effect of the measure at least to the same extent as the estimation process.

[0016] FIG. 1 is a block diagram showing an example of the configuration of the traffic control system 10 according to the present embodiment. As shown in FIG. 1, the traffic control system 10 includes a management system 100, a vehicle sensor 300, a display device 401, an information providing terminal 402, and a toll system 403. Each component is connected via a network such as the Internet, for example. The network may be any of a wireless network, a wired network, and a network in which wireless and wired are mixed.

[0017] The vehicle sensor 300 senses vehicles passing on the road and outputs traffic flow data based on the sensing result to the management system 100. The traffic flow data is data representing the traffic flow on the road and includes, for example, the average speed of vehicles in the target section and the number of vehicles (number of vehicle units) passing through the target section per unit time. The number of vehicle units per unit time may be referred to as traffic volume. Although one vehicle sensor 300 is shown in FIG. 1, the traffic control system 10 may include a plurality of vehicle sensors 300. For example, a plurality of vehicle sensors 300 may be installed on the roadside of the road.

[0018] The display device 401 is a device that displays display information transmitted from the management system 100. The display device 401 is, for example, a large display provided in the traffic control room where the management system 100 is installed, a large display provided in a service area and a parking area, and an electric signboard installed above the road. The display information is information (traffic information) representing traffic conditions such as the time required to reach a specific point and traffic jam information, for example.

[0019] The information providing terminal 402 is a terminal device to which information transmitted from the management system 100 is provided. The information providing terminal 402 is, for example, a terminal for providing information installed in a service area and a parking area, a mobile terminal (smartphone, mobile phone, etc.) owned by a road user, and a car navigation device.

[0020] The toll system 403 is a system for managing road tolls. The toll system 403 performs dynamic pricing to change the toll for each road according to traffic information transmitted from, for example, the management system 100. Dynamic pricing increases the toll for a road with traffic congestion and decreases the toll for a road without traffic congestion, for example. This can guide vehicles to roads without traffic congestion and suppress the occurrence of traffic congestion.

[0021] In FIG. 1, one display device 401, information providing terminal 402, and toll system 403 are shown respectively, but the traffic control system 10 may include a plurality of display devices 401, a plurality of information providing terminals 402, and a plurality of toll systems 403.

[0022] The management system 100 is a system that controls various processes related to traffic control. The management system 100 may be referred to as a center processing device, a central processing unit, etc. The management system 100 includes a storage unit 121, an information generation unit 101, an output control unit 102, and an estimation system 200.

[0023] The management system 100 or the estimation system 200 corresponds to an information processing device. The management system 100 is realized, for example, as a server device provided in a traffic control room. Part or all of the functions of the management system 100 may be constructed on a cloud environment.

[0024] The storage unit 121 stores various data used in the management system 100. For example, the storage unit 121 stores traffic flow data transmitted from the vehicle sensor 300 and parameters of the machine learning model used in the estimation system 200. The estimation system 200 may include a storage unit that stores parameters of the machine learning model and the like.

[0025] The storage unit 121 can be configured by any commonly used storage medium such as a flash memory, a memory card, a RAM (Random Access Memory), an HDD (Hard Disk Drive), and an optical disk.

[0026] The information generation unit 101 generates display information to be displayed on the display device 401 and traffic information to be transmitted to the information providing terminal 402, the fee system 403, etc. For example, the information generation unit 101 uses the traffic flow data obtained from the vehicle sensor 300 to determine whether congestion has occurred in the target section, and generates traffic information representing the determination result. Any method can be used to determine whether congestion has occurred. For example, a method of determining congestion when the number of vehicles is greater than a threshold value or when the average speed is less than a threshold value can be used.

[0027] The output control unit 102 controls the output of information to devices such as the display device 401, the information providing terminal 402, and the fee system 403. For example, the output control unit 102 outputs the display information generated by the information generation unit 101 to the display device 401. Also, the output control unit 102 outputs the traffic information generated by the information generation unit 101 to the information providing terminal 402 and the fee system 403.

[0028] In the present embodiment, the output control unit 102 controls the output of information corresponding to the measures whose effects have been confirmed by the estimation system 200. The measures include, for example, changes in the display information for the display device 401, changes in the traffic information output to the information providing terminal 402, changes in the medium for providing information, and changes in the fee system in the fee system 403. The change in the medium for providing information means, for example, changing to which of the output destinations of one or more display devices 401 and one or more information providing terminals 402 to provide the information.

[0029] The estimation system 200 is a system that estimates the effects when measures for controlling the traffic flow are executed. FIG. 2 is a block diagram showing an example of the configuration of the estimation system 200. As shown in FIG. 2, the estimation system 200 includes an acquisition unit 201, a no-measure data generation unit 202 (first generation unit), a determination unit 203, a reception unit 204, a with-measure data generation unit 205 (second generation unit), an estimation unit 206, a learning data generation unit 207, a learning unit 208 (first learning unit), an evaluation unit 209, a learning unit 210 (second learning unit), and an inference unit 211.

[0030] The acquisition unit 201 acquires various data used in various processes by the estimation system 200. For example, the acquisition unit 201 reads out traffic flow data transmitted from the vehicle sensor 300 and stored in the storage unit 121. The traffic flow data transmitted from the vehicle sensor 300 includes traffic flow data collected when the operation of the measure has not been started (hereinafter also referred to as traffic flow data before measure operation), and traffic flow data collected when the operation of the measure has been started (hereinafter also referred to as traffic flow data during measure operation). The acquisition unit 201 outputs these traffic flow data to each unit that requires them.

[0031] Note that the start of the operation of the measure means starting the process of executing the measure confirmed to be effective by the estimation system 200 according to the actually collected traffic flow data. Even after the operation of the measure has started, for example, in a traffic situation where congestion does not occur, the measure may not be executed.

[0032] The no-measure data generation unit 202 generates traffic flow data (first data) when the measure is not executed using the traffic flow data before measure operation. Hereinafter, the traffic flow data when the measure is not executed is referred to as traffic flow data DA. The traffic flow data DA is used in the determination process of whether to execute the measure by the determination unit 203. The traffic flow data DA is also used as data input to the machine learning model.

[0033] As described above, when the implementation of the measure has not started, the acquisition unit 201 acquires traffic flow data before the implementation of the measure from the storage unit 121. The traffic flow data before the implementation of the measure is data representing the traffic flow data at a certain time, and is not necessarily data of the day and time zone that exactly matches the traffic situation to be estimated for the effect of the measure. That is, the traffic flow data before the implementation of the measure acquired by the acquisition unit 201 has variations.

[0034] Therefore, the data generation unit 202 without measure generates traffic flow data DA by executing an estimation process using the traffic flow data before the implementation of the measure. For example, the data generation unit 202 without measure generates traffic flow data DA for a specified number in advance based on the traffic flow data before the implementation of the measure. As a method for generating the traffic flow data DA, for example, a method using a traffic flow simulator and a method generated by Monte Carlo simulation based on a statistical model can be applied. The specified number in advance is, for example, the number of days or the number of time zones.

[0035] One or more parameters (second parameters) used in the estimation process by the data generation unit 202 without measure may be specified by the user or the like and configured to be received by the reception unit 204 described later. Also, if variations in the traffic flow data before the implementation of the measure acquired by the acquisition unit 201 are allowed, etc., the data generation unit 202 without measure may generate the acquired traffic flow data before the implementation of the measure as the traffic flow data DA as it is.

[0036] The determination unit 203 determines whether or not to execute the measure using the traffic flow data DA. For example, the determination unit 203 determines whether or not congestion has occurred in the target section using the traffic flow data DA, and determines that the implementation of the measure is necessary when congestion has occurred.

[0037] When it is determined that the execution of a measure is necessary, the reception unit 204 receives the input of one or more parameters (first parameters) used in the estimation process by the with-measure data generation unit 205. The estimation process by the with-measure data generation unit 205 is a process of estimating traffic flow data when the measure is executed. The parameters are parameters that are assumed to change when the measure is executed, in other words, parameters that affect the effect of the measure.

[0038] For example, when the estimation process by the with-measure data generation unit 205 is realized by a traffic flow simulator, the selection probability of each road, the selection probability of the departure time zone, and the setting of the toll system, etc. can be parameters. Hereinafter, an example in which the selection probability of the road is set as a parameter will be mainly described. FIG. 3 is a diagram showing an example of the selection probability of the road to be set. In FIG. 3, a setting example of the selection probability before the execution of the measure and the selection probability after the execution of the measure for two roads R1 and R2 is shown.

[0039] As described above, the reception unit 204 may be configured to receive the input of the parameters used in the estimation process by the without-measure data generation unit 202.

[0040] The with-measure data generation unit 205 executes an estimation process using the parameters (for example, the selection probability of the road) received by the reception unit 204, and generates traffic flow data (second data) when the measure is executed. Hereinafter, the traffic flow data when the measure is executed is referred to as the traffic flow data DB. As a method for generating the traffic flow data DB, for example, a method using a traffic flow simulator and a method of generating by Monte Carlo simulation based on a statistical model can be applied.

[0041] When executing the estimation process using the traffic flow simulator, the data generation unit 205 with measures changes the selection probability of the road among the parameters required by the traffic flow simulator to the received selection probability, and executes the estimation process without changing the others. When executing the estimation process using the Monte Carlo simulation, the data generation unit 205 with measures changes the selection probability of the road among the parameters required by the Monte Carlo simulation to the received selection probability, and executes the estimation process without changing the others.

[0042] The estimation unit 206 compares the traffic flow data DA generated by the data generation unit 202 without measures and the traffic flow data DB generated by the data generation unit 205 with measures, and estimates the effect of the measures. For example, the estimation unit 206 estimates the difference between the average vehicle speed included in the traffic flow data DA and the average vehicle speed included in the traffic flow data DA, or the difference between the number of vehicles included in the traffic flow data DA and the number of vehicles included in the traffic flow data DA, as the effect of the measures.

[0043] The learning data generation unit 207 generates learning data used for learning the machine learning model. For example, the learning data generation unit 207 determines which of the traffic flow data DA and the traffic flow data DB will be observed during the implementation of the measures based on the determination result of the necessity of implementing the measures by the determination unit 203, and generates learning data such that either the traffic flow data DB or the traffic flow data DA is adopted according to the necessity of implementation. For example, when it is determined that the implementation of the measures is not necessary, the traffic flow data DA is adopted, and when it is determined that the implementation of the measures is necessary, the traffic flow data DB is adopted. The generated learning data is data that simulates the traffic flow data actually observed through the vehicle detector 300 during the implementation of the measures.

[0044] The learning unit 208 learns the machine learning model using the generated learning data. Note that the number of machine learning models to be learned may be one or more. Hereinafter, an example will be described in which a plurality of different machine learning models are each learned, and the machine learning model with better performance than other machine learning models is selected and used for inference.

[0045] For example, the learning unit 208 learns a plurality of machine learning models for estimating the effect of a measure using learning data. The machine learning model may be any model, and for example, a model such as Causal Forest can be applied. The plurality of machine learning models may be a plurality of Causal Forests with different structures (e.g., the number of branches), or may include one or more Causal Forests and one or more machine learning models based on a method different from Causal Forest.

[0046] The evaluation unit 209 evaluates the performance of the machine learning model by comparing the estimation result E1 of the effect of the measure by the learned machine learning model with the estimation result E2 of the effect of the measure by the estimation unit 206. For example, the evaluation unit 209 calculates, for each of the plurality of machine learning models, an index representing the difference between the estimation result E1 and the estimation result E2 as the performance of the machine learning model. The index is, for example, the RMSE (Root Mean Squared Error) between the estimation result E1 and the estimation result E2, but may be calculated by any other method. Note that the estimation results E1 and E2 are estimated using, for example, traffic flow data different from the learning data.

[0047] The evaluation unit 209 selects, as the machine learning model to be used for the inference by the inference unit 211, a machine learning model with a performance greater than that of other machine learning models (for example, the machine learning model corresponding to the index with the smallest difference). In this way, the evaluation unit 209 can select a machine learning model that can obtain an estimation result closest to the effect of the measure estimated by an estimation process different from the machine learning model (the effect estimated by the estimation unit 206). In other words, it becomes possible to verify the performance of the machine learning model using the effect obtained by an estimation process different from the machine learning model as the correct data.

[0048] In the case of a configuration in which one machine learning model is used, the evaluation unit 209 may at least have a function of evaluating the performance of one machine learning model. That is, the evaluation unit 209 evaluates the performance of this machine learning model by comparing the effect estimated by the estimation unit 206 with the effect estimated by the learned machine learning model.

[0049] The learning unit 210 learns the model selected by the evaluation unit 209 using the learning data. The learning data used at this time is traffic flow data (traffic flow data during policy operation) transmitted from the vehicle sensor 300 during policy operation and stored in the storage unit 121.

[0050] The inference unit 211 infers the effect of the policy for which the operation has been started by inputting the traffic flow data during policy operation into the machine learning model learned by the learning unit 210.

[0051] Note that each part (information generation unit 101, output control unit 102) in the management system 100 shown in FIG. 1 and each part (acquisition unit 201, data generation unit 202 without policy, determination unit 203, reception unit 204, data generation unit 205 with policy, estimation unit 206, learning data generation unit 207, learning unit 208, evaluation unit 209, learning unit 210, inference unit 211) in the estimation system 200 shown in FIG. 2 are realized by, for example, one or a plurality of processors. For example, each of the above parts may be realized by causing a processor such as a CPU (Central Processing Unit) to execute a program, that is, by software. Each of the above parts may be realized by a processor such as a dedicated IC (Integrated Circuit), that is, by hardware. Each of the above parts may be realized by using a combination of software and hardware. When using a plurality of processors, each processor may realize one of each part or two or more of each part.

[0052] Next, the estimation process by the estimation system 200 according to the present embodiment will be described. FIG. 4 is a flowchart showing an example of the estimation process in the present embodiment. The estimation process in FIG. 4 is a process of estimating the effect of a measure by comparing traffic flow data DA and traffic flow data DB.

[0053] The no-measure data generation unit 202 generates traffic flow data DA when no measure is executed, using the traffic flow data before the execution of the measure acquired by the acquisition unit 201 (step S101). The determination unit 203 determines whether the measure needs to be executed, using the generated traffic flow data DA (step S102). When it is determined that the execution of the measure is necessary, the with-measure data generation unit 205 sets the selection probability of the road received by the reception unit 204 as, for example, parameters of a traffic flow simulator, and generates traffic flow data DB when the measure is executed (step S103). The estimation unit 206 estimates the effect of the measure by comparing traffic flow data DA and traffic flow data DB (step S104).

[0054] Next, the model evaluation process by the estimation system 200 will be described. FIG. 5 is a flowchart showing an example of the model evaluation process in the present embodiment.

[0055] The learning data generation unit 207 generates learning data using the determination result of whether the measure needs to be executed by the determination unit 203 (step S201). For example, the learning data generation unit 207 generates learning data including traffic flow data DA when it is determined that the execution of the measure is not necessary and traffic flow data DB when it is determined that the execution of the measure is necessary.

[0056] The learning unit 208 learns a plurality of machine learning models respectively using the generated learning data (step S202). The evaluation unit 209 compares the estimation result of the effect of the measure by the plurality of learned machine learning models with the estimation result of the effect of the measure by the estimation unit 206, and evaluates the performance of each machine learning model (step S203). The evaluation unit 209 selects a machine learning model with better performance than other machine learning models (step S204).

[0057] Next, the inference process by the estimation system 200 will be described. The inference process is a process of estimating the effect of a measure on traffic flow data during measure operation using the machine learning model whose performance has been evaluated. FIG. 6 is a flowchart showing an example of the inference process in the present embodiment.

[0058] The learning unit 210 first uses the traffic flow data during measure operation as learning data to train the selected machine learning model (step S301). In this way, using the traffic flow data obtained after actually starting the operation of the measure, the machine learning model whose performance has been evaluated is further trained. The inference unit 211 uses the trained machine learning model to estimate the effect of the measure on the traffic flow data during measure operation obtained after training (step S302).

[0059] As described above, according to the present embodiment, the performance of the machine learning model is verified in advance using the traffic flow data DB and the traffic flow data DA, and for the machine learning model whose performance has been confirmed, the effect of the measure during measure operation is learned using the traffic flow data during measure operation. Thereby, it is possible to ensure that the effect of the measure during measure operation is appropriately estimated.

[0060] The estimated measure effect may be displayed, for example, on a large display (display device 401) provided in the traffic control room by the output control unit 102. As described above, according to the present embodiment, it is possible to display, on the display device 401 provided in the traffic control room or the like, in addition to the current traffic situation (such as traffic jam information), the traffic situation during measure operation or the traffic situation before measure operation.

[0061] The output control unit 102 may output information necessary for actually executing the measures for which the effects have been confirmed. For example, the output control unit 102 gives an execution instruction and outputs necessary information in order to execute measures such as changing the display information for the display device 401, changing the traffic information output to the information providing terminal 402, changing the medium for providing information, and changing the fee system in the fee system 403.

[0062] Next, the hardware configuration of the information processing apparatus (management system 100, estimation system 200) according to the embodiment will be described with reference to FIG. 7. FIG. 7 is an explanatory diagram showing a hardware configuration example of the information processing apparatus according to the embodiment.

[0063] The information processing apparatus according to the embodiment includes a control device such as a CPU 51, a storage device such as a ROM (Read Only Memory) 52 and a RAM 53, a communication I / F 54 that connects to a network and performs communication, and a bus 61 that connects each part.

[0064] The program executed by the information processing apparatus according to the embodiment is provided by being pre-installed in the ROM 52 or the like.

[0065] The program executed by the information processing apparatus according to the embodiment may be configured to be recorded on a computer-readable recording medium such as a CD-ROM (Compact Disk Read Only Memory), a flexible disk (FD), a CD-R (Compact Disk Recordable), or a DVD (Digital Versatile Disk) in an installable format or an executable format file and provided as a computer program product.

[0066] Furthermore, the program executed by the information processing apparatus according to the embodiment may be stored on a computer connected to a network such as the Internet and provided by being downloaded via the network. Also, the program executed by the information processing apparatus according to the embodiment may be configured to be provided or distributed via a network such as the Internet.

[0067] The program executed by the information processing apparatus according to the embodiment can cause the computer to function as each part of the information processing apparatus described above. This computer can execute the program read from a computer-readable storage medium onto the main storage device by the CPU 51.

[0068] Although some embodiments of the present invention have been described, these embodiments are presented by way of example and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the gist of the invention. These embodiments and their modifications are included in the scope and gist of the invention, and are also included in the invention described in the claims and the equivalent scope thereof.

Explanation of Reference Numerals

[0069] 10 Traffic control system 100 Management system 101 Information generation unit 102 Output control unit 121 Storage unit 200 Estimation system 201 Acquisition unit 202 Data generation unit without measures 203 Determination unit 204 Reception unit 205 Data generation unit with measures 206 Estimation unit 207 Learning data generation unit 208 Learning unit 209 Evaluation unit 210 Learning unit 211 Inference unit 300 Vehicle Sensor 401 Display Device 402 Information Providing Terminal 403 Fare System

Claims

1. Data to be input into a model for inferring the effect when the measure is executed, by inputting observation data that varies depending on whether the measure has been executed or not, the first generation unit generates first data representing the observation data when the measure is not executed; A parameter used in the first estimation process for estimating the observation data when the measure is executed, the reception unit receives an input of one or more first parameters that affect the effect; The second generation unit executes the first estimation process using the received first parameter and generates second data representing the observation data estimated by the first estimation process; An estimation unit that estimates the effect by a second estimation process that is different from the process of inferring the effect by the model and that compares the first data and the second data; A first learning unit that learns the model using learning data including the first data and the second data; An evaluation unit that evaluates the performance of the model by comparing a first effect that is the effect inferred by the learned model and a second effect that is the effect estimated by the second estimation process; An information processing apparatus comprising the above.

2. The first learning unit learns a plurality of the models using the learning data; The evaluation unit evaluates the performance of the plurality of models by comparing the second effect with a plurality of first effects estimated by each of the plurality of models, selects the model having a greater performance than other models, and The inference unit further comprises an inference unit that infers the effect by inputting the observation data obtained when the measure is executed into the selected model. The information processing apparatus according to claim 1.

3. The information processing apparatus according to claim 2, further comprising a second learning unit that learns the selected model using, as learning data, the observation data obtained when the measure is executed. The information processing apparatus according to claim 2.

4. The reception unit receives an input of one or more second parameters used in the process of estimating the observation data when the measure is not executed; The first generation unit executes a process of estimating the observation data when the measure is not executed using the received second parameter and generates the first data representing the estimated observation data. The information processing apparatus according to any one of claims 1 to 3.

5. Further provided is a determination unit that determines whether the measure needs to be executed based on the first data generated by the first generation unit. When the determination unit determines that the execution of the measure is necessary, the second generation unit generates the second data. The information processing apparatus according to any one of claims 1 to 4.

6. The observation data is traffic flow data representing the traffic flow of a plurality of roads through which the vehicle passes. The first parameter includes the probability that a plurality of the roads are selected. The information processing apparatus according to any one of claims 1 to 5.

7. An information processing method executed by an information processing apparatus, A first generation step of generating first data representing the observation data when the measure is not executed, which is data input to a model for inferring the effect when the measure is executed by inputting observation data that varies depending on whether the measure has been executed; A reception step of receiving an input of one or more first parameters that affect the effect, which are parameters used in a first estimation process for estimating the observation data when the measure is executed; A second generation step of executing the first estimation process using the received first parameter and generating second data representing the observation data estimated by the first estimation process; An estimation step of estimating the effect by a second estimation process that compares the first data and the second data, which is a process different from the process of inferring the effect by the model; A first learning step of learning the model using learning data including the first data and the second data; An evaluation step of evaluating the performance of the model by comparing a first effect that is the effect inferred by the learned model and a second effect that is the effect estimated by the second estimation process; An information processing method including the above steps.

8. In a computer, A first generation step of generating first data representing the observation data when the measure is not executed, which is data input to a model for inferring the effect when the measure is executed by inputting observation data that varies depending on whether the measure has been executed; A reception step of receiving an input of one or more first parameters that affect the effect, which are parameters used in a first estimation process for estimating the observation data when the measure is executed; A second generation step of executing the first estimation process using the received first parameter and generating second data representing the observation data estimated by the first estimation process; An estimation step of estimating the effect by a second estimation process that is different from the process of inferring the effect by the model and compares the first data and the second data; A first learning step of learning the model using learning data including the first data and the second data; An evaluation step of evaluating the performance of the model by comparing a first effect that is the effect inferred by the learned model and a second effect that is the effect estimated by the second estimation process; A program for causing the above to be executed.

Images