JP7743916B2 - Model analysis device, model analysis method, and program - Google Patents

Description

This disclosure relates to the analysis of machine learning models.

In recent years, predictive models obtained through machine learning have been used in a variety of fields. Patent Document 1 describes a method for visualizing the relationship between the number of training data items used to train a discriminative model and the discrimination accuracy of the model obtained through training, for each category of training data used to train the model.

International Publication No. WO2021/085188

When making predictions using a predictive model, differences in model performance can occur depending on the categories and attributes of the input dataset. In this case, if the dataset used for prediction has a large number of categories or attributes, it becomes difficult to manually check the differences in model performance.

One objective of this disclosure is to evaluate and visualize the performance of a model for each categorical attribute of the dataset used for prediction.

In one aspect of the present disclosure, a model analysis apparatus includes:
a model acquisition means for acquiring a model;
selection means for receiving a selection of attributes;
a dataset acquisition means for acquiring a dataset corresponding to the selected attribute ;
a performance calculation means for calculating the performance of the model for each attribute corresponding to each category of the data set;
an output means for outputting performance information indicating the calculated performance of the model for each attribute;
Equipped with
The performance calculation means calculates the performance of the model for each attribute corresponding to a category other than the category from which the attribute was selected .

In another aspect of the present disclosure, a computer-implemented model analysis method includes:
Get the model,
It takes a selection of attributes,
Retrieve the dataset corresponding to the selected attributes ,
Calculating the performance of the model for each attribute corresponding to each category of the dataset;
Outputting performance information showing the performance of the model for each calculated attribute;
The performance of the model is calculated for each attribute corresponding to a category other than the category for which the attribute was selected .

In yet another aspect of the invention, a program includes:
Get the model,
It takes a selection of attributes,
Retrieve the dataset corresponding to the selected attributes ,
Calculating the performance of the model for each attribute corresponding to each category of the dataset;
causing a computer to execute a process of outputting performance information indicating the calculated performance of the model for each attribute ;
The performance of the model is calculated for each attribute corresponding to a category other than the category for which the attribute was selected .

This disclosure makes it possible to evaluate and visualize the performance of a model for each categorical attribute of the dataset used for prediction.

1 is a block diagram showing the overall configuration of a model generation system according to a first embodiment. FIG. 2 is a block diagram showing a hardware configuration of the model generating device. FIG. 2 is a block diagram showing the functional configuration of the model generation device according to the first embodiment. An example of displaying attribute-specific performance information is shown below. 10 shows another example of displaying attribute-specific performance information. 10 is a flowchart of a model analysis process. FIG. 1 is a block diagram showing a schematic configuration of a model generation system using a server and a terminal device. FIG. 10 is a block diagram showing the functional configuration of a model analysis apparatus according to a second embodiment. 10 is a flowchart of a process performed by a model analysis device according to a second embodiment.

Hereinafter, preferred embodiments of the present disclosure will be described with reference to the drawings.
First Embodiment
[Overall configuration]
1 is a block diagram showing the overall configuration of a model generation system according to a first embodiment. The model generation system 1 includes a model generation device 100, a display device 2, and an input device 3. The model generation device 100 is an application of the model analysis device of the present disclosure, and is configured by a computer such as a personal computer (PC). The display device 2 is, for example, a liquid crystal display device, and displays evaluation information generated by the model generation device 100. The input device 3 is, for example, a mouse, a keyboard, etc., and is used by the user to give instructions and input necessary when modifying a model or viewing evaluation information.

First, we will provide an overview of the operation of the model generation system 1. The model generation device 100 generates a machine learning model (hereinafter simply referred to as a "model") using training data prepared in advance. The model generation device 100 also analyzes and evaluates the generated model. Specifically, the model generation device 100 performs model predictions using evaluation data, etc., and analyzes the predictive performance of the model based on the prediction results. The model generation device 100 calculates a value indicating the performance of the model (hereinafter also referred to as a "performance value") for each attribute of a category of the evaluation data. The model generation device 100 then presents information indicating the performance of the model for each attribute of the category to the user as evaluation information. The user can check the evaluation information and operate the input device 3 to input correction information for correcting the model.

Note that a "machine learning model" is information that represents the relationship between explanatory variables and a response variable. A machine learning model is, for example, a component that estimates the results of an estimation target by calculating a response variable based on explanatory variables. A machine learning model is generated by executing a learning algorithm using training data, for which the response variable value has already been obtained, and arbitrary parameters as input. A machine learning model may be represented, for example, by a function c that maps an input x to a correct answer y. A machine learning model may estimate the numerical value of an estimation target, or may estimate the label of an estimation target. A machine learning model may output a variable that describes the probability distribution of the response variable. A machine learning model may also be referred to as a "learning model," "analysis model," "AI (Artificial Intelligence) model," or "prediction formula." An explanatory variable is a variable used as input in a machine learning model. An explanatory variable may also be referred to as a "feature" or "characteristic."

The learning algorithm for generating the machine learning model is not particularly limited and may be any existing learning algorithm. For example, the learning algorithm may be a piecewise linear model using random forest, support vector machine, naive Bayes, or FAB inference (Factorized Asymptotic Bayesian Inference), or a neural network. Note that a piecewise linear model technique using FAB inference is disclosed, for example, in U.S. Patent Publication US 2014/0222741 A1.

[Hardware configuration]
2 is a block diagram showing the hardware configuration of the model generation device 100. As shown in the figure, the model generation device 100 includes an interface (I/F) 111, a processor 112, a memory 113, a recording medium 114, and a database (DB) 115.

I/F 111 inputs and outputs data to and from external devices. Specifically, training data and evaluation data used to generate the model, as well as instructions and input entered by the user using input device 3, are input to model generation device 100 via I/F 111. In addition, evaluation information for the model generated by model generation device 100 is output to display device 2 via I/F 111.

Processor 112 is a computer such as a CPU (Central Processing Unit) that controls the entire model generation device 100 by executing a pre-prepared program. Processor 112 may also be a GPU (Graphics Processing Unit), TPU (Tensor Processing Unit), quantum processor, or FPGA (Field-Programmable Gate Array). Processor 112 executes the model analysis process described below.

Memory 113 is composed of ROM (Read Only Memory), RAM (Random Access Memory), etc. Memory 113 is also used as working memory while processor 112 is executing various processes.

The recording medium 114 is a non-volatile, non-transitory recording medium such as a disk-shaped recording medium or semiconductor memory, and is configured to be detachable from the model generation device 100. The recording medium 114 records various programs executed by the processor 112. When the model generation device 100 executes various processes, the programs recorded on the recording medium 114 are loaded into the memory 113 and executed by the processor 112.

DB115 stores information about the model generated by model generation device 100 (hereinafter referred to as the "existing model") and the model after correction through retraining (hereinafter referred to as the "corrected model"). DB115 also stores, as necessary, training data, evaluation data, correction information entered by the user, etc., input via I/F111.

(Functional configuration)
3 is a block diagram showing the functional configuration of the model generation device 100 according to the first embodiment. Functionally, the model generation device 100 includes a training data DB 121, a model training unit 122, a model DB 123, an evaluation data DB 124, and an analysis unit 125.

The training data DB 121 stores training data used to generate a model. The training data D1 is input to the model training unit 122. The training data D1 is composed of multiple combinations of input data and correct answers (correct values or correct labels) for the input data.

The model training unit 122 trains a model using the training data D1 and generates a model. The model training unit 122 outputs model data M corresponding to the generated model to the model DB 123 and the analysis unit 125. The model data M includes multiple parameter information constituting the model. The parameter information includes, for example, information on explanatory variables used as input to the model, information on weights for each explanatory variable, and information on weights for each sample constituting the input data.

The model training unit 122 also retrains the existing model to generate a modified model. In this case, the model training unit 122 modifies the parameters that make up the model based on the modification information D3 input by the user using the input device 3, and retrains the model using training data for retraining as necessary. The model training unit 122 stores the model data M of the modified model obtained by retraining in the model DB 123 and outputs it to the analysis unit 125.

The evaluation data DB 124 stores evaluation data used to evaluate the generated model. The evaluation data may be, for example, a portion of the collected data that was not used as training data, newly collected data, or validation data prepared for verification. Training data may also be used as evaluation data. The evaluation data consists of multiple combinations of input data and correct answers (correct values or correct labels) for the input data.

Evaluation data is stored and used in units of datasets. Each piece of data contained in a dataset contains multiple categories. For example, in the case of a predictive model for determining creditworthiness, data on a large number of individuals to be evaluated is prepared as evaluation data, and each piece of data contains categories such as age group, gender, residential area, family structure, and income. Furthermore, each category contains multiple attributes (groups).

The analysis unit 125 uses the evaluation data to analyze the performance of the model for each attribute corresponding to each category in the dataset. For example, suppose a category called "gender" includes the attributes "male" and "female." In this case, the analysis unit 125 calculates a performance value indicating the model's performance for each of the attributes "male" and "female." The analysis unit 125 also calculates model performance values for each attribute for other categories. Note that various indices that can be used to evaluate model performance, such as the model's predictive accuracy and compatibility, can be used as model performance indicators. Other examples of model performance indicators include the F1 score, precision, and recall, and in the case of regression tasks, the coefficient of determination, etc. Furthermore, error-based indices, in which the lower the value, the better, such as mean squared error and cross entropy, can also be used. The analysis unit 125 then generates information indicating the model's performance value for each attribute for multiple categories (hereinafter referred to as "attribute-specific performance information") and outputs this information to the display device 2 as evaluation information D2. The analysis unit 125 is an example of a model acquisition means, a data set acquisition means, a performance calculation means, and an output means.

The display device 2 displays the evaluation information D2 output by the analysis unit 125 on the display device 2. By looking at the displayed attribute-specific performance information, the user can learn the differences in model performance for each attribute in each category.

In addition, the user inputs correction information D3 into the input device 3 as needed to correct the current model. The correction information D3 is information related to the correction, such as information on explanatory variables used as input to the model, information on weights for each explanatory variable, and information on weights for each sample that makes up the input data. The model training unit 122 corrects the model by retraining the model using the input correction information D3.

[Example of attribute-specific performance information display]
Next, an example of displaying attribute-specific performance information will be described.
(First display example)
(1) Initial Display Example Fig. 4(A) shows an initial display example of the first display example of attribute-specific performance information. In the initial display example 40, model performance values are displayed for each attribute for the entire data set used for prediction. In the following example, model accuracy is used as the model performance value.

Specifically, the initial display example 40 includes an overall performance value column 41, a category column 42, and a performance value column 43. The overall performance value column 41 shows the performance value of the model for the entire target dataset. In this example, "overall accuracy 85%" is shown as the performance value of the model for the entire dataset. The category column 42 shows the category for which the model performance value is displayed. In this example, the model performance values are displayed for the four categories "gender," "age," "place of residence," and "annual income."

The performance value column 43 shows the model performance value for each attribute for each category, using a frame corresponding to each attribute (hereinafter referred to as an "attribute frame"). For example, for the category "gender," the performance value for the attribute "male" is 95%, and the performance value for the attribute "female" is 60%. Note that the percentage of each attribute is shown in parentheses next to each attribute name in the performance value column 43. For example, in the example of Figure 4 (A), for the category "gender," the percentage of males in the entire dataset is 60%, and the percentage of females is 40%. Furthermore, the size (horizontal length) of each attribute frame corresponds to the percentage of each attribute. For example, for the category "gender," the length of the attribute frame for "male" and the attribute frame for "female" is 3:2 (60:40).

For the category "age," the performance value for the attribute "children" is 35%, the performance value for the attribute "adults" is 94%, and the performance value for the attribute "elderly" is 95%. Furthermore, within the category "age," the proportion of attributes "children" is 20%, the proportion of attributes "adults" is 60%, and the proportion of attributes "elderly" is 20%.

For the category "Residence," the performance value for the attribute "Kanto" is 84% and the performance value for the attribute "Kansai" is 86%. Furthermore, within the category "Residence," the proportion for the attribute "Kanto" is 55%, and the proportion for the attribute "Kansai" is 45%.

For the category "Annual Income," the performance value for the attribute "Less than 4 million yen" is 83%, and the performance value for the attribute "4 million yen or more" is 87%. Furthermore, within the category "Annual Income," the proportion of those with the attribute "Less than 4 million yen" is 48%, and the proportion of those with the attribute "4 million yen or more" is 52%.

The category column 42 and performance value column 43 are color-coded based on the fairness of the model's performance. Note that in this specification, for ease of illustration, hatching is used instead of color coding. A small difference in model performance values between different attributes in the same category is said to be "fair model performance," and a large difference in model performance values is said to be "unfair model performance." The category column 42 and performance value column 43 are color-coded using different types of hatching depending on whether the model's performance is fair or unfair.

Assume that the threshold for determining model performance as unfair is 5%, and that a difference in accuracy of 5% or greater is considered unfair. For the category "gender," the accuracy of the attribute "male" is 95%, while the accuracy of the attribute "female" is 60%. Since the difference is equal to or greater than the threshold of 5%, the model's performance is considered unfair. Therefore, as shown in the legend in Figure 4(A), the attribute frame for "male," with a high performance value, is indicated by hatching in pattern P1, and the attribute frame for "female," with a low performance value, is indicated by hatching in pattern P2. In actual display, it is preferable to display the attribute frame with the lower performance value more prominently, for example, by color-coding the attribute frame with the lower performance value as "red" and the attribute frame with the higher performance value as "yellow." Furthermore, the category column 42 is indicated by hatching in pattern P2, just like the attribute frame with the lower performance value. This allows the user to easily recognize that the model's performance for that category is unfair simply by looking at the category column 42.

For the category "Age," the performance value of the attribute "Children" is lower than the performance values of the attributes "Adults" and "Elderly" by more than the threshold value, so the model's performance is determined to be unfair. As a result, the category column 42 and the attribute frame for "Children" are indicated by hatching in pattern P2, and the attribute frames for "Adults" and "Elderly" are indicated by hatching in pattern P1.

On the other hand, for the category "Residence," the difference between the performance value for the attribute "Kanto" and the performance value for the attribute "Kansai" is below the threshold, so the model's performance is determined to be fair. Therefore, both the category column 42 and the performance value column 43 are shown with hatching in pattern P3, which indicates that the model's performance is fair.

For the category "Annual Income," the difference between the performance value for the attribute "Less than 4 million yen" and the performance value for the attribute "4 million yen or more" is below the threshold, so the model's performance is determined to be fair. Therefore, both the category column 42 and the performance value column 43 are shown with hatching in pattern P3, which indicates that the model's performance is fair.

As described above, the display of attribute-specific performance information shown in Figure 4(A) displays model performance values for each attribute for each category. This allows the user to know that even for the same model, performance values differ depending on the attributes of the categories included in the dataset used for prediction. Furthermore, for each category, the fairness of the model's performance is determined based on the performance values for each attribute, and categories where the model's performance is fair and categories where it is unfair are displayed separately, allowing the user to see at a glance the fairness of the model for each category.

(2) Display Example When Attribute is Selected Fig. 4(B) shows an example of how attribute-specific performance information is displayed when the user selects an attribute. In the initial display example 40 shown in Fig. 4(A), the user can select any attribute corresponding to each category using, for example, cursor C. In the display example 40x when an attribute is selected shown in Fig. 4(B), the user selects the attribute "male" for the category "gender" and the attribute "adult" for the category "age."

When an attribute is selected, the analysis unit 125 extracts data from the dataset that corresponds to the selected attribute to create a partial dataset, and generates and displays attribute-specific performance information for the partial dataset.

Specifically, in Figure 4 (B), the overall performance value column 41 displays the performance values for the attributes selected by the user, i.e., "Male" and "Adult." The category column 42 displays the "Gender" and "Age" selected by the user in white (no hatching). The attribute boxes for "Gender" and "Age" selected by the user are displayed in black, indicating that they have been selected, while the other attribute boxes are displayed in white, indicating that they have not been selected. Furthermore, each attribute box displays only the percentage of each attribute relative to the total.

On the other hand, for the categories "Place of Residence" and "Annual Income," which are not subject to selection by the user, hatching patterns indicating whether the model's performance for that category is fair or not are displayed in the category column 42 and performance value column 43, similar to the example in Figure 4(A). Specifically, for the category "Place of Residence," the model's performance is fair, so both the category column 42 and performance value column 43 are displayed with hatching pattern P3. On the other hand, for the category "Annual Income," the difference between the performance value for the attribute "Less than 4 million yen" and the performance value for the attribute "4 million yen or more" is equal to or greater than the above-mentioned threshold of 5%, and the model's performance is unfair, so the attribute box for "4 million yen or more" is displayed with hatching pattern P1, and the category column 42 and the attribute box for "Less than 4 million yen" are displayed with hatching pattern P2.

In this embodiment, a user can select any attribute on the attribute-specific performance information display screen to display performance values for each attribute in each category for data corresponding to that attribute. By selecting an attribute and displaying attribute-specific performance information, the user can identify unfairness in model performance that was not apparent in the initial display example. For example, in the example shown in Figure 4(A), which displays attribute-specific performance information for the entire data set, the model's performance is determined to be fair for the "annual income" category. However, as shown in Figure 4(B), by limiting the data to "male" and "adult," the user can notice that the model's performance is unfair in the "annual income" category. Therefore, by selecting any attribute and displaying attribute-specific performance information, the user can confirm whether the model's fairness is guaranteed under various conditions (i.e., combinations of attributes).

(Second display example)
5A shows a second display example of attribute-specific performance information. The first display example above displays model performance for each attribute category for one model and one data set. In contrast, the second display example displays model performance for each attribute category for multiple models and one data set.

As shown, the second display example 50 has a model column 51, an overall performance value column 52, a category column 53, an attribute column 54, and a performance value column 55. The model column 51 shows multiple models. In this example, for five models h1 to h5, the model performance is shown for each attribute in each category. The overall performance value column 52 shows the performance value of the model for the target dataset. In this example, the accuracy of the model for the target dataset is shown for each model.

The category column 53 indicates the category for which the model's performance value is displayed. In this example, model performance values are displayed for three categories: "Gender," "Age," and "Place of Residence." The attribute column 54 indicates the attributes corresponding to each category. The performance value column 55 has an attribute frame for each attribute in each category, and the model's performance value is displayed within the attribute frame. Furthermore, as with the first display example, the attribute frame for each attribute is hatched with one of patterns P1 to P3, which indicates the fairness of the model.

According to the second display example 50, the fairness of model performance for each category can be displayed simultaneously for multiple models. This allows the user to compare and consider the fairness of model performance for multiple models. For example, by targeting multiple models updated over time, the transition in model performance can be evaluated from the perspective of the fairness of model performance.

In the second display example, the user may also select an attribute, which will display attribute-specific performance information for the data set corresponding to the selected attribute. In this case, the user may select the attribute in a window separate from display example 50. Alternatively, the user may be able to select an attribute by clicking on attribute column 54.

(Third display example)
5B shows a third display example of attribute-specific performance information. The first display example above displays model performance for each attribute category for one model and one data set. In contrast, the third display example displays model performance for each attribute category for one model and multiple data sets.

As shown, the third display example 60 has a dataset column 61, an overall performance value column 62, a category column 63, an attribute column 64, and a performance value column 65. The dataset column 61 shows multiple datasets. In this example, the model performance is shown for each attribute in each category for five datasets DS1 to DS5. The overall performance value column 62 shows the performance value of the model for each dataset. In this example, the accuracy of the model for each of the five datasets DS1 to DS5 is shown. The category column 63, attribute column 64, and performance value column 65 are basically the same as those in the second display example 50.

According to the third display example 50, the fairness of model performance for each category can be displayed simultaneously for multiple datasets. This allows the user to compare and consider the fairness of model performance for multiple datasets.

In the third display example, the user may also select an attribute, which will display attribute-specific performance information for the data set corresponding to the selected attribute. In this case, the user may select the attribute in a window separate from display example 60. Alternatively, the user may be able to select an attribute by clicking on attribute column 64.

[Model analysis processing]
Next, the model analysis process performed by the model generating device 100 will be described. FIG. 6 is a flowchart of the model analysis process performed by the model generating device 100. As described above, the model analysis process is a process of calculating the performance of a model for each attribute in each category and displaying attribute-specific performance information. This process is realized by the processor 112 shown in FIG. 2 executing a program prepared in advance and operating mainly as the analysis unit 125 shown in FIG. 3.

First, the analysis unit 125 acquires the target dataset (step S10). Next, the analysis unit 125 divides the dataset by category to create partial datasets, and calculates the model's performance value for each attribute in each category by making predictions using the model (step S11). Then, the analysis unit 125 displays attribute-specific performance information on the display device 2 (step S12), as illustrated in Figures 4 and 5.

Next, the analysis unit 125 determines whether the user has selected an attribute using the input device 3 (step S13). If an attribute has been selected (step S13: Yes), the analysis unit 125 acquires a dataset corresponding to the selected attribute (step S14). The analysis unit 125 then repeats steps S11 to S12 for the dataset corresponding to the selected attribute. In this way, each time the user selects an attribute, attribute-specific performance information is generated and displayed for the dataset corresponding to the selected attribute.

On the other hand, if no attribute has been selected (step S13: No), the analysis unit 125 determines whether the user has input an end instruction (step S15). If an end instruction has not been input (step S15: No), the process returns to step S13. On the other hand, if an end instruction has been input (step S15: Yes), the process ends.

When calculating the model performance value for each attribute in step S11, if the data included in the dataset is a continuous variable such as numerical data, the analysis unit 125 generates attributes by converting the continuous variables into categories. Specifically, the analysis unit 125 converts continuous variables such as age and annual income into ordered categories. For example, the analysis unit 125 divides the range between the minimum and maximum values of the continuous variable equally into two or more attributes (groups) and converts them into categories.

Note that, if a dividing point can be prepared based on business knowledge, the analysis unit 125 may convert a continuous variable into a category according to that dividing point. For example, in the case of age, the analysis unit 125 creates three attributes, such as "child," "adult," and "elderly," based on business knowledge. The analysis unit 125 may also determine a dividing point according to percentiles so that the number of data samples for each attribute is the same, and convert the continuous variable into a category. The analysis unit 125 may also create multiple attributes by clustering the data (variables) included in the dataset. Note that if the variables included in the dataset are originally categorical, such as gender or place of residence, the above processing is not necessary.

[Modification]
(Variation 1)
In the example of FIG. 4(B), when a user selects an attribute in an arbitrary category, the analysis unit 125 generates and displays attribute-specific performance information for a dataset corresponding to the selected attribute. Alternatively, the analysis unit 125 may automatically calculate attribute combinations that result in unfair model performance and present the information to the user. Specifically, as described above, a threshold value for the difference in performance values for determining that model performance is unfair is first determined in advance. The analysis unit 125 calculates attribute-specific performance information for all combinations of attributes corresponding to multiple categories and extracts categories in which model performance is unfair. The analysis unit 125 then outputs and displays information including the extracted categories and the performance values for each attribute corresponding to the categories. This makes it possible to extract and display attribute combinations that result in unfair model performance without the user having to repeatedly select attributes.

(Variation 2)
In the above embodiment, the model generation device 100 is configured as an independent device such as a PC, but instead, the model generation device may be configured by a server and a terminal device. FIG. 7 is a block diagram showing a schematic configuration of a model generation system 1x using a server and a terminal device. In FIG. 7, the server 100x has the configuration of the model generation device 100 shown in FIG. 3. Furthermore, the display device 2x and input device 3x of the terminal device 7 used by the user are used as the display device 2 and input device 3 shown in FIG. 3.

Second Embodiment
8 is a block diagram showing the functional configuration of a model analysis device 70 according to the second embodiment. The model analysis device 70 includes a model acquisition unit 71, a data set acquisition unit 72, a performance calculation unit 73, and an output unit 74.

Figure 9 is a flowchart of processing by the model analysis device 70 of the second embodiment. The model acquisition means 71 acquires a model (step S71). The dataset acquisition means 72 acquires a dataset (step S72). The performance calculation means 73 calculates the performance of the model for each attribute corresponding to each category of the dataset (step S73). The output means 74 outputs performance information indicating the calculated performance of the model for each attribute (step S74).

The model analysis device 70 of the second embodiment makes it possible to evaluate the performance of a model for each attribute of a dataset category.

Some or all of the above embodiments may also be described as, but are not limited to, the following notes:

(Appendix 1)
a model acquisition means for acquiring a model;
a dataset acquisition means for acquiring a dataset;
a performance calculation means for calculating the performance of the model for each attribute corresponding to each category of the data set;
an output means for outputting performance information indicating the calculated performance of the model for each attribute;
A model analysis device comprising:

(Appendix 2)
selection means for receiving a selection of the attribute;
the data set acquisition means acquires a data set corresponding to the selected attribute,
2. The model analysis device according to claim 1, wherein the performance calculation means calculates the performance of the model for each attribute corresponding to a category other than the category from which the attribute was selected.

(Appendix 3)
3. The model analysis device according to claim 1, wherein the performance information distinguishes between categories in which the difference in model performance for each attribute is equal to or greater than a predetermined value and categories in which the difference is less than the predetermined value.

(Appendix 4)
4. The model analysis device according to claim 1, wherein the performance information includes information indicating a ratio of the number of data items for each attribute corresponding to a category.

(Appendix 5)
the model acquisition means acquires a plurality of models;
5. The model analysis apparatus according to claim 1, wherein the performance information includes performance information about the plurality of models.

(Appendix 6)
the data acquisition means acquires a plurality of data sets;
5. The model analysis apparatus according to claim 1, wherein the performance information includes performance information for the plurality of data sets.

(Appendix 7)
7. The model analysis apparatus according to claim 1, wherein the performance calculation means determines attributes corresponding to the categories by clustering data included in the dataset.

(Appendix 8)
7. The model analysis device according to claim 1, wherein the performance calculation means determines attributes corresponding to the categories by percentile division so that the number of data samples included in each attribute is equal.

(Appendix 9)
Get the model,
Get the dataset,
Calculating the performance of the model for each attribute corresponding to each category of the dataset;
A model analysis method that outputs performance information indicating the performance of a model for each calculated attribute.

(Appendix 10)
Get the model,
Get the dataset,
Calculating the performance of the model for each attribute corresponding to each category of the dataset;
A recording medium storing a program for causing a computer to execute a process for outputting performance information indicating the calculated performance of a model for each attribute.

The present disclosure has been described above with reference to embodiments and examples, but the present disclosure is not limited to the above embodiments and examples. Various modifications that would be understood by a person skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure.

1, 1x Model generation system 2, 2x Display device 3, 3x Input device 7 Terminal device 100 Model generation device 112 Processor 121 Training data DB
122 Model Training Department 123 Model DB
124 Evaluation data DB
125 Analysis Department

Claims (9)

a model acquisition means for acquiring a model;
selection means for receiving a selection of attributes;
a dataset acquisition means for acquiring a dataset corresponding to the selected attribute ;
a performance calculation means for calculating the performance of the model for each attribute corresponding to each category of the data set;
an output means for outputting performance information indicating the calculated performance of the model for each attribute;
Equipped with
The performance calculation means is a model analysis device that calculates the performance of the model for each attribute corresponding to a category other than the category for which the attribute was selected . The model analysis device of claim 1, wherein the performance information distinguishes between categories where the difference in model performance for each attribute is equal to or greater than a predetermined value and categories where the difference is less than the predetermined value. The model analysis device according to claim 1 , wherein the performance information includes information indicating a ratio of the number of data items for each attribute corresponding to a category. the model acquisition means acquires a plurality of models;
The model analysis device according to claim 1 , wherein the performance information includes performance information about the plurality of models. the data set acquisition means acquires a plurality of data sets;
The model analysis apparatus according to claim 1 , wherein the performance information includes performance information about the plurality of data sets. The model analysis device according to claim 1 , wherein the performance calculation means determines attributes corresponding to the categories by clustering data included in the data set. The model analysis device according to claim 1 , wherein the performance calculation means determines attributes corresponding to the categories by percentile division so that the number of data samples included in each attribute is equal. 1. A computer-implemented method for model analysis, comprising:
Get the model,
It takes a selection of attributes,
Retrieve the dataset corresponding to the selected attributes ,
Calculating the performance of the model for each attribute corresponding to each category of the dataset;
Outputting performance information showing the performance of the model for each calculated attribute;
A model analysis method in which the performance of the model is calculated for each attribute corresponding to a category other than the category for which the attribute was selected . Get the model,
It takes a selection of attributes,
Retrieve the dataset corresponding to the selected attributes ,
Calculating the performance of the model for each attribute corresponding to each category of the dataset;
causing a computer to execute a process of outputting performance information indicating the calculated performance of the model for each attribute ;
The performance of the model is calculated for each attribute corresponding to a category other than the category for which the attribute was selected .