Invention title: Analyzer, control method, and program
Technical field
[0001]
　The present invention relates to data analysis.
Background technology
[0002]
　A system for analyzing the business contents of a company has been developed. For example, Patent Document 1 discloses an apparatus for determining the presence or absence of an abnormality in a product by a test. More specifically, the apparatus of Patent Document 1 determines the presence or absence of an abnormality by performing a test comparing a target value with an actual value for a variable representing the final quality characteristic of a product. Then, when there is an abnormality in the final quality characteristic, the apparatus of Patent Document 1 also performs a test comparing the target value and the actual value for other variables related to the final quality characteristic to determine the presence or absence of the abnormality. judge.
Prior art literature
Patent documents
[0003]
Patent Document 1: Japanese Patent Application Laid-Open No. 2019-36061
Outline of the invention
Problems to be solved by the invention
[0004]
　Patent Document 1 is based on the premise that a target value representing an appropriate variable value is known in advance. Therefore, the apparatus of Patent Document 1 cannot be used to specify the target value of the variable.
[0005]
　The present invention has been made in view of the above problems, and one of the objects thereof is to provide a technique for facilitating grasping an appropriate value of a factor affecting the result.
Means to solve problems
[0006]
　The analyzer of the present invention 1) sets a set of target data indicating a pair of factor data which is a value related to a factor and a result data which is a value related to a result, and divided number information indicating the number of divided numerical ranges of the factor data. Regarding the acquisition unit to be acquired and 2) the result data in two samples obtained by dividing the target data included in the set of target data into two by the boundary value for each of the plurality of boundary values ​​that divide the numerical range into numbers. It has a calculation unit that calculates a test statistic representing a difference, and 3) a generation unit that generates test statistic information indicating a plurality of test statistic calculated for each boundary value.
[0007]
　The control method of the present invention is executed by a computer. The control method 1) acquires a set of target data indicating a pair of factor data which is a value related to a factor and result data which is a value related to a result, and information on the number of divisions indicating the number of divisions of a numerical range of the factor data. For each of the acquisition unit and 2) multiple boundary values ​​that divide the numerical range into numbers, the difference regarding the result data between the two samples obtained by dividing the target data included in the target data set into two by the boundary value. It has a calculation unit that calculates the test statistics to be represented, and 3) a generation unit that generates test statistics information indicating a plurality of test statistics calculated for each boundary value.
[0008]
　The program of the present invention causes a computer to execute the control method of the present invention.
Effect of the invention
[0009]
　Techniques are provided that facilitate the understanding of appropriate values ​​for factors that affect results.
A brief description of the drawing
[0010]
[Fig. 1] Fig. 1 is a diagram for explaining an outline of the analyzer of the present embodiment.
[Fig. 2] Fig. 2 is a diagram conceptually exemplifying the process of calculating a test statistic for each sample pair.
FIG. 3 is a diagram illustrating a functional configuration of the analyzer of the first embodiment.
[Fig. 4] Fig. 4 is a diagram illustrating a computer for realizing an analyzer.
[Fig. 5] Fig. 5 is a diagram illustrating a usage environment of an analyzer.
FIG. 6 is a flowchart illustrating a flow of processing executed by the analyzer of the first embodiment.
[Fig. 7] Fig. 7 is a diagram illustrating test statistic information realized as a table.
[Fig. 8] Fig. 8 is a diagram illustrating test statistic information realized as a test statistic graph.
[Fig. 9] Fig. 9 is a diagram illustrating a screen provided to a user.
[Fig. 10] Fig. 10 is a diagram illustrating a screen including a histogram and test statistic information.
FIG. 11 is a diagram illustrating a screen including a time series graph, a histogram, and a test statistic graph.
FIG. 12 is a diagram illustrating a first sample and a second sample in the analyzer 2000 of a modified example.
FIG. 13 illustrates another method of dividing the target data set 10 into two samples using two boundary value pairs.
FIG. 14 is a diagram illustrating information output by the analyzer 2000 of the modified example.
Mode for carrying out the invention
[0011]
　Hereinafter, embodiments of the present invention will be described with reference to the drawings. In all drawings, similar components are designated by the same reference numerals, and description thereof will be omitted as appropriate. Further, in each block diagram, unless otherwise specified, each block represents a functional unit configuration rather than a hardware unit configuration.
[0012]
[Embodiment 1]

　FIG. 1 is a diagram for explaining an outline of the analyzer 2000 of the present embodiment. Note that FIG. 1 is an example for facilitating understanding of the analyzer 2000, and the function of the analyzer 2000 is not limited to that shown in FIG.
[0013]
　The analyzer 2000 acquires a target data set 10 which is a set of a plurality of target data 20. The target data 20 is data in which the factor value (factor data 22) and the result value (result data 24) are associated with each other. For example, in the manufacture of products, the content of certain materials can affect the quality of the product. Therefore, it is conceivable that the factor data 22 indicates the content of a specific material, and the result data 24 handles a flag indicating whether or not the product is a defective product (hereinafter referred to as a defective flag).
[0014]
　The analyzer 2000 divides the target data 20 into two by a plurality of different patterns, and calculates a test statistic representing the difference between the two sets (samples) obtained by dividing each pattern into two. Therefore, the analyzer 2000 further acquires the divided number information 30. The number of division information 30 indicates the number (hereinafter, the number of divisions) indicating how many numerical ranges of the factor data 22 are divided.
[0015]
　The analyzer 2000 calculates the value of each boundary that divides the numerical range of the factor data 22 by the number of divisions indicated by the division number information 30. For example, assume that the numerical range of the factor data 22 is 0 to 100 and the number of divisions is 4. The boundary values ​​that divide this numerical range into four equal parts are 25, 50, and 75. Therefore, these three boundary values ​​are specified by the analyzer 2000.
[0016]
　The analyzer 2000 performs the following processing for each of the specified plurality of boundary values. First, the analyzer 2000 divides the target data set 10 into two sets (hereinafter, first sample and second sample) with the boundary value as a boundary. As a result, a pair of the first sample and the second sample (hereinafter referred to as a sample pair) having a number equal to the number of divided samples can be obtained.
[0017]
　Further, the analyzer 2000 calculates, for each boundary value, a test statistic representing the difference with respect to the result data 24 for the sample pair obtained by the division based on the boundary value. For example, assume that the result data 24 is a bad flag. In this case, for example, the analyzer 2000 calculates a test statistic representing the difference between the defective rate of the product in the first sample (the ratio of defective products to the total number of products) and the defective rate of the product in the second sample.
[0018]
　FIG. 2 is a diagram conceptually exemplifying the process of calculating the test statistic for each sample pair. In the graph of FIG. 2, the horizontal axis shows the value of the factor data 22, and the vertical axis shows the value of the result data 24. The result data 24 shows the defective rate. In this example, the number of divisions is 4. Therefore, three boundary values ​​b1, b2, and b3 are calculated. Note that min and max on the horizontal axis are the upper limit value and the lower limit value of the numerical range of the factor data 22, respectively.
[0019]
　The analyzer 2000 uses the boundary value b1 to use the boundary value b1 as the target data set 10, the first sample including the target data 20 in which the factor data 22 is b1 or less, and the second sample containing the target data 20 in which the factor data 22 is larger than b1. Divide into and. Then, the analyzer 2000 calculates a test statistic representing the difference in the result data 24 between the first sample and the second sample as the test statistic corresponding to the boundary value b1. For example, when the result data 24 indicates whether or not the product is defective, the defective rate can be calculated for each of the first sample and the second sample, and the test statistic representing the difference between the calculated defective rates can be calculated. Conceivable. Similarly, the analyzer 2000 has a test statistic showing the difference in the result data 24 between the first and second samples obtained using the boundary value b2, and the first sample and the second sample obtained using the boundary value b3. A test statistic showing the difference in the defective rate to the difference in the result data 24 in the sample is also calculated.
[0020]
　The analyzer 2000 generates test statistic information indicating a plurality of calculated test statistic. For example, the test statistic information is a chart showing the correspondence between the boundary value and the sample pair obtained by the division based on the boundary value. In FIG. 1, as the test statistic information 40, a line graph is generated in which the horizontal axis shows the boundary value and the vertical axis shows the test statistic calculated for the boundary value.
[0021]

　According to the analyzer 2000 of the present embodiment, "the target data set 10 is divided into two samples based on the value of the factor data 22, and the difference regarding the result data 24 is determined for these two samples. The process of "calculating the test statistic to be represented" is performed in a plurality of patterns using different boundary values. Then, the test statistic information 40 indicating the test statistic obtained for each of the plurality of patterns is output. A user who sees such test statistic information 40 can easily grasp "what is the appropriate value of the factor data 22 in order to obtain a good result". That is, according to the analyzer 2000, it is possible to accurately and easily grasp what kind of value the factor data 22 should have in relation to the result data 24.
[0022]
　Here, advanced statistical knowledge is required to properly perform data analysis. Therefore, if a person other than a specialist with such knowledge tries to analyze the data, it may take a lot of time or an error may occur in the analysis. In this regard, the analyzer 2000 automatically performs processing such as dividing data into a plurality of patterns and calculating test statistics for each pattern, so that a person who does not have advanced statistical knowledge (for example, a company) Even a person in charge of business) can accurately and easily grasp an appropriate value of the factor data 22 that affects the result data 24.
[0023]
　Hereinafter, the present embodiment will be described in more detail.
[0024]

　FIG. 3 is a diagram illustrating the functional configuration of the analyzer 2000 of the first embodiment. The analyzer 2000 has an acquisition unit 2020, a calculation unit 2040, and a generation unit 2060. The acquisition unit 2020 acquires the target data set 10 and the division number information 30. The calculation unit 2040 calculates a plurality of boundary values ​​for dividing the numerical range of the factor data 22 into the number of divisions indicated by the number of division information 30. Further, the calculation unit 2040 calculates a test statistic representing the difference between the first sample and the second sample obtained by dividing the target data set 10 into two at the boundary value for each of the plurality of boundary values. .. The generation unit 2060 generates test statistic information indicating a plurality of calculated test statistic.
[0025]
Each functional component of the analyzer 2000 may be realized by hardware (eg, hard-wired electronic circuit, etc.) that realizes each functional component, or may be hardware. It may be realized by a combination of hardware and software (eg, a combination of an electronic circuit and a program that controls it). Hereinafter, a case where each functional component of the analyzer 2000 is realized by a combination of hardware and software will be further described.
[0026]
　FIG. 4 is a diagram illustrating a computer 1000 for realizing the analyzer 2000. The computer 1000 is an arbitrary computer. For example, the calculator 1000 is a stationary computer such as a PC (Personal Computer) or a server machine. In addition, for example, the calculator 1000 is a portable calculator such as a smartphone or a tablet terminal.
[0027]
　The computer 1000 may be a dedicated computer designed to realize the analyzer 2000, or may be a general-purpose computer. In the latter case, for example, by installing a predetermined application on the computer 1000, each function of the analyzer 2000 is realized on the computer 1000. The above application is composed of a program for realizing the functional component of the analyzer 2000.
[0028]
　The computer 1000 includes a bus 1020, a processor 1040, a memory 1060, a storage device 1080, an input / output interface 1100, and a network interface 1120. The bus 1020 is a data transmission path for the processor 1040, the memory 1060, the storage device 1080, the input / output interface 1100, and the network interface 1120 to transmit and receive data to and from each other. However, the method of connecting the processors 1040 and the like to each other is not limited to the bus connection.
[0029]
　The processor 1040 is various processors such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), and an FPGA (Field-Programmable Gate Array). The memory 1060 is a main storage device realized by using RAM (Random Access Memory) or the like. The storage device 1080 is an auxiliary storage device realized by using a hard disk, an SSD (Solid State Drive), a memory card, a ROM (Read Only Memory), or the like.
[0030]
　The input / output interface 1100 is an interface for connecting the computer 1000 and the input / output device. For example, an input device such as a keyboard and an output device such as a display device are connected to the input / output interface 1100.
[0031]
　The network interface 1120 is an interface for connecting the computer 1000 to the communication network. This communication network is, for example, a LAN (Local Area Network) or a WAN (Wide Area Network).
[0032]
　The storage device 1080 stores a program (a program that realizes the above-mentioned application) that realizes each functional component of the analyzer 2000. The processor 1040 reads this program into the memory 1060 and executes it to realize each functional component of the analyzer 2000.
[0033]

　In order to facilitate understanding of the analyzer 2000, an example of the usage environment of the analyzer 2000 will be described. FIG. 5 is a diagram illustrating a usage environment of the analyzer 2000.
[0034]
　In FIG. 5, the analyzer 2000 is connected to the user terminal 60 via a network. The user operates the user terminal 60 to send a request to the analyzer 2000 to provide test statistic information 40 for the target data set 10. For example, the request includes information representing a condition regarding the target data set 10 and information on the number of divisions 30. The analyzer 2000 acquires the target data set 10 corresponding to the conditions indicated in the request, and generates the test statistic information 40 by using the target data set 10 and the divided number information 30. Then, the analyzer 2000 transmits the generated test statistic information 40 to the user terminal 60.
[0035]
　For example, the analyzer 2000 provides screen data (for example, a Web page) including test statistic information 40 to the user terminal 60. In this case, the user terminal 60 displays her received Web page on the browser. By doing so, the user can browse the test statistic information 40.
[0036]
　The usage environment of the analyzer 2000 is not limited to that shown in FIG. For example, the analyzer 2000 may be operated directly by the user instead of being used via the user terminal 60.
[0037]

　FIG. 6 is a flowchart illustrating a processing flow executed by the analyzer 2000 of the first embodiment. The acquisition unit 2020 acquires the target data set 10 (S102). The acquisition unit 2020 acquires the divided number information 30 (S104). The calculation unit 2040 calculates a plurality of boundary values ​​for dividing the numerical range of the factor data 22 into the number of divisions shown in the number of division information 30 (S106).
[0038]
　S108 to S114 are loop processes A executed for each boundary value. In S108, the calculation unit 2040 determines whether or not the loop process A has already been executed for all the boundary values. When the loop process A has already been executed for all the boundary values, the process of FIG. 6 proceeds to S116. On the other hand, if there is a boundary value that is not yet the target of the loop process A, the calculation unit 2040 selects one of them. Then, the process of FIG. 6 proceeds to S110. The boundary value selected here is called a boundary value i.
[0039]
　The calculation unit 2040 divides the target data set 10 into two at the boundary value i to generate a first sample and a second sample (S110). The calculation unit 2040 calculates the test statistic for the result data 24 for the first sample and the second sample (S112). Since S114 is the end of the loop process A, the process of FIG. 6 proceeds to S108.
[0040]
　After the loop process A, the generation unit 2060 generates test statistic information 40 for the calculated plurality of test statistic (S116). The generation unit 2060 outputs the test statistic information 40 (S118).
[0041]

　The target data set 10 is a set including a plurality of target data 20 which are a pair of factor data 22 and result data 24. Each target data 20 indicates, for example, data for each case in which the same process or work is performed at different time points. For example, as described above, the target data 20 indicates the content of the specific material and whether or not the product manufactured using the material having the content is defective, respectively, in the factor data 22 and the result data 24. Suppose you are. In this case, for example, each target data 20 indicates the content of the material having the above characteristics used in the product manufactured at that time and whether or not the product is defective for the products manufactured at different time points. ing.
[0042]
　Here, as the target data 20, learning data for learning an estimation model that estimates the value of the objective variable from the values ​​of a plurality of explanatory variables may be used. For example, suppose that an estimation model is generated in which "an index related to a product produced under the manufacturing conditions is estimated based on each of the indexes one level above the manufacturing conditions of the product". In this case, each index representing the manufacturing conditions represents an explanatory variable, and the index relating to the manufactured product represents an objective variable. Indicators representing manufacturing conditions are, for example, the environment such as temperature and humidity, and the content of each material. Indicators related to a product include, for example, a flag indicating whether or not the product is defective, an index indicating a specific performance of the product, and the like.
[0043]
　The value of the explanatory variable and the value of the objective variable described above can be regarded as factor data 22 and result data 24, respectively. Therefore, for example, the learning data used to generate such an estimation model can be used as the target data 20.
[0044]
　The acquisition unit 2020 acquires the target data set 10 (S102). There are various methods for the acquisition unit 2020 to acquire the target data set 10. For example, the acquisition unit 2020 acquires the target data set 10 from the storage device in which the target data set 10 is stored. In addition, for example, the acquisition unit 2020 may acquire the target data set 10 by receiving the target data set 10 transmitted from another device.
[0045]
　Here, when a plurality of target data sets 10 are stored in the storage device, the acquisition unit 2020 specifies the target data set 10 to be generated as the test statistic information 40, and selects the specified target data set 10. Obtained from storage device. In this case, for example, the acquisition unit 2020 acquires the condition for the target data set 10 to be acquired, and acquires the target data set 10 that matches the condition.
[0046]
　The conditions for the target data set 10 are, for example, the identification information (name, etc.) of the target data set 10, the period during which the target data set 10 was obtained, and the location related to the target data set 10 (for example, the manufacturing location of the product). And so on. Here, when the target data set 10 is used to generate the estimation model, the identification information of the estimation model generated by using the target data set 10 may be used as the identification information of the target data set 10. Information representing these conditions regarding the target data set 10 is included in, for example, the request transmitted from the user terminal 60 described above.
[0047]
　The target data 20 stored in the storage device may not be divided in units of the target data set 10. In this case, for example, the acquisition unit 2020 acquires the condition of the target data 20 to be included in the target data set 10, and acquires the target data 20 that matches the condition from the storage device. Then, the acquisition unit 2020 treats a set of acquired target data 20 as the target data set 10. The conditions relating to the target data 20 are, for example, a period during which the target data 20 is obtained, a place related to the target data 20 (for example, a place where the product is manufactured), and the like.
[0048]
　In addition, for example, if some processing using the target data set 10 has already been performed before the generation of the test statistic information 40, the analyzer 2000 selects the target data set 10 from such another processing. By taking over, the target data set 10 may be acquired. As another process using the target data set 10, for example, a process of generating an estimation model using the target data set 10 and a process of visualizing the target data set 10 using a graph or the like can be considered.
[0049]
　For example, an interface (button or the like) that requests the generation of test statistic information 40 for the target data set 10 used in the processing is provided on the screen showing the results of these processings. The analyzer 2000 generates and outputs the test statistic information 40 in response to a request using this interface. As a result, the user can easily obtain the information of the test statistic information 40 obtained by analyzing the target data set 10 used for generating the estimation model and the visualized target data set 10 from another viewpoint. be able to.
[0050]
　When the device that has performed the processing such as the generation of the estimation model is a device different from the analyzer 2000, the identification information of the target data set 10 and the target data set 10 itself are included in the request from the device. As a result, the analyzer 2000 can acquire the target data set 10.
[0051]
　When the target data 20 includes data for each of a plurality of types of factors, which factor should be focused on and the target data set 10 is sampled (that is, which factor is used as the factor data 22). It is necessary to specify (whether to handle). The identification method is arbitrary. For example, the acquisition unit 2020 receives from the user an input for designating which factor data is to be handled as the factor data 22. In addition, for example, when the target data set 10 is inherited from another process as described above, the data about the factor of interest in the other process is treated as the factor data 22. For example, when the target data set 10 is inherited from the process of visualizing the data about a certain factor, the data about the visualized factor is treated as the factor data 22.
[0052]
The
　acquisition unit 2020 acquires the divided number information 30 (S104). There are various methods for the acquisition unit 2020 to acquire the divided number information 30. For example, the acquisition unit 2020 acquires the divided number information 30 stored in the storage device. In addition, for example, the acquisition unit 2020 may acquire the division number information 30 by accepting a user operation for inputting the division number. In addition, for example, the acquisition unit 2020 may acquire the division number information 30 by receiving the division number information 30 transmitted from another device. In this case, for example, the divided number information 30 is included in the request transmitted from the user terminal 60 described above.
[0053]
The
　calculation unit 2040 calculates a plurality of boundary values ​​for dividing the numerical range of the factor data 22 by the number of divisions shown in the division number information 30 (S106). Therefore, the calculation unit 2040 specifies the numerical range of the factor data 22. For example, the calculation unit 2040 sets a numerical range in which the smallest factor data 22 included in the target data set 10 is the lower limit value and the largest factor data 22 included in the target data set 10 is the upper limit value. Is treated as the numerical range of the factor data 22.
[0054]
　In addition, for example, the calculation unit 2040 may specify a numerical range of the factor data 22 after excluding a part of the target data 20 included in the target data set 10. In this case, for example, the calculation unit 2040 treats a numerical range in which the minimum value and the maximum value of the factor data 22 included in the excluded target data set 10 are the lower limit value and the upper limit value, respectively, as the numerical range of the factor data 22. .. In the target data 20 excluded from the target data set 10, for example, the factor data 22 indicates an outlier. As a method for specifying outliers from a plurality of numerical values, various existing methods can be used.
[0055]
　In addition, for example, the calculation unit 2040 may acquire information that defines a numerical range of the factor data 22. In this case, the lower limit of the numerical range of the factor data 22 is smaller than the minimum value of the factor data 22 included in the target data set 10, or the upper limit of the numerical range of the factor data 22 is included in the target data set 10. It may be larger than the maximum value of the factor data 22.
[0056]
　The calculation unit 2040 calculates each boundary value that equally divides the numerical range of the specified factor data 22 into the number of divisions indicated by the number of division information 30. Such a boundary value can be calculated by, for example, the following equation (1).
b_i = (max-min) / N * i (1)
　where bi represents the i-th boundary value counting from the smallest. max represents the upper limit of the numerical range of the factor data 22, and min represents the lower limit of the numerical range of the factor data 22. N represents the number of divisions. The number of boundary values ​​is N-1.
[0057]

　For each boundary value, the calculation unit 2040 divides the target data set 10 into two at the boundary value to generate a sample pair (first sample and second sample) (S110). For example, the calculation unit 2040 divides the target data set 10 into two, a set of target data 20 showing factor data 22 below the boundary value and a set of target data 20 showing factor data 22 larger than the boundary value. The former is treated as the first sample and the latter is treated as the second sample. However, the calculation unit 2040 generates a set of target data 20 showing factor data 22 smaller than the boundary value to be the first sample, and generates a set of target data 20 showing factor data 22 equal to or more than the boundary value to generate a second sample. It may be used as a sample.
[0058]

　For each boundary value, the calculation unit 2040 calculates a test statistic representing the difference in the result data 24 between the first sample and the second sample generated by using the boundary value (calculation of test statistic). S112). Here, as a test for comparing two samples in this way, there are a test for a difference in ratio, a test for a difference in mean value, and the like. The calculation unit 2040 calculates test statistics that can be used for these tests. As the type of test statistic, various types such as two-sample t-test statistic can be adopted.
[0059]
　For example, suppose that the difference in ratio is tested for the first sample and the second sample. In this case, the calculation unit 2040 calculates a test statistic representing the difference between the ratio of the predetermined type of the result data 24 included in the first sample and the ratio of the predetermined type of the result data 24 included in the second sample. do. For example, when the result data 24 is a defect flag, it indicates that the factor data 22 is a defective product with respect to the total number of target data 20 included in the sample as a predetermined type of ratio with respect to the result data 24. The ratio of the number of target data 20) can be used.
[0060]
　The type of test and the type of test statistic may be fixed in advance or may be specified by user input or the like.
[0061]
The
　generation unit 2060 generates test statistic information 40 indicating the test statistic calculated for each boundary value (S116). The test statistic information 40 represents the boundary value and the test statistic calculated for the boundary value in association with each other. Various types of information such as charts can be adopted as such information.
[0062]
　For example, the test statistic information 40 includes a table showing the correspondence between the boundary value and the test statistic. FIG. 7 is a diagram illustrating test statistic information 40 realized as a table. The table of FIG. 7 is called a table 50. Table 50 has a boundary value of 52 and a test statistic of 54. In FIG. 7, one record in the table 50 shows the boundary value and the test statistic calculated for two samples obtained by dividing the target data set 10 by the boundary value in association with each other.
[0063]
　The test statistic information 40 may be a graph (hereinafter, test statistic graph) showing the correspondence between the boundary value and the test statistic. FIG. 8 is a diagram illustrating test statistic information 40 realized as a test statistic graph 70. In the test statistic graph 70, the X-axis represents the boundary value and the Y-axis represents the test statistic.
[0064]
　In the test statistic graph 70, each correspondence between the boundary value and the test statistic is plotted, and a line 72 connecting these plots is displayed. The type of graph is not limited to the line graph, and various types of graphs such as bar graphs can be adopted.
[0065]
　The test statistic graph 70 further includes a threshold display 74 representing a threshold for the test statistic. The threshold display 74 represents the threshold of the test statistic obtained based on the significance level. In the test statistic graph 70, the test statistic above the threshold represented by the threshold display 74 is a statistically significant value. That is, when the test statistic exceeds the threshold value, it can be said that there is a statistically significant difference between the two samples divided by the boundary value corresponding to the test statistic. By including the threshold display 74 in the test statistic graph 70 in this way, it is possible to easily grasp whether or not the test statistic corresponding to each boundary value is a statistically significant value.
[0066]
　For example, in FIG. 8, it is assumed that the result data 24 is a defect flag and the factor data 22 represents the content of the component A. Further, it is assumed that the difference in the defective rate is tested between the samples. Further, it is assumed that the test statistic becomes a positive value when the defect rate of the second sample is higher than that of the first sample.
[0067]
　Looking at the test statistic graph 70, in the case where the boundary value is b3 or less, the test statistic is smaller than the threshold value. On the other hand, in the case where the boundary value is b4 or more, the test statistic is larger than the threshold value. From this, it can be seen that there is a significant difference in the defective rate of the products when comparing the products having the content of component A of b4 or more and the products having the content of component A less than b4. Therefore, by looking at the test statistic graph 70 of FIG. 8, the user can easily say, "In order to reduce the defective rate, it is preferable that the content of component A is less than b4." Can be grasped.
[0068]
　In FIG. 8, by adding a color (pattern for convenience of illustration) to the area above the threshold value display 74, it is possible to more easily grasp the part where the statistical test statistic is equal to or more than the threshold value and the other parts. There is. Coloring may be further performed on the area below the threshold display 74. Further, the coloring may not be performed on the area above the threshold value display 74, but may be performed only on the area below the threshold value display 74.
[0069]
　In addition, the generation unit 2060 may highlight the plot representing the test statistic equal to or higher than the threshold value in the test statistic graph 70. As the highlighting method, various methods such as increasing the size of the plot, changing the shape of the plot, and blinking the plot can be adopted.
[0070]
　The threshold of the test statistic can be obtained by converting the significance level into the test statistic. For example, suppose the significance level is 5% and the type of test statistic is t-value. In this case, the t-value threshold can be obtained by converting the p-value = 0.05, which represents the significance level of 5%, to the t-value. As a specific method for converting the value representing the significance level into the test statistic, an existing method according to the type of the test statistic can be used.
[0071]
　The significance level may be fixed in advance or may be specified by the user. For example, the analyzer 2000 provides the user with a screen on which information such as the number of divisions, the type of test, and the significance level can be specified.
[0072]
　FIG. 9 is a diagram illustrating a screen provided to the user. The screen 80 of FIG. 9 includes input areas 82, 84, and 86. These are input areas where the type of test, significance level, and number of divisions can be specified, respectively. Since the number of divisions can be said to be the particle size of the test, it is expressed as "test particle size" in FIG. After inputting the information in these input areas, the user presses the button 88 for executing the test. As a result, the test statistic graph 70 described with reference to FIG. 8 is displayed. The test statistic graph 70 may be displayed on the screen 80, or may be displayed on a screen different from the screen 80. The button 88 is not essential, and when an input is made in each input area, a test statistic graph 70 reflecting the content thereof may be automatically generated and displayed.
[0073]
The
　generation unit 2060 outputs the generated test statistic information 40 (S118). The output destination of the test statistic information 40 is arbitrary. For example, the generation unit 2060 causes the display device accessible from the analyzer 2000 to display the test statistic information 40. In addition, for example, the generation unit 2060 stores the test statistic information 40 in a storage device accessible from the analyzer 2000. In addition, for example, the generation unit 2060 transmits the test statistic information 40 to another device (for example, the user terminal 60) accessible from the analyzer 2000.
[0074]
The
　generation unit 2060 may generate a histogram for the factor data 22 included in the target data set 10, and output this histogram and the test statistic information 40 together. For example, the histogram and the test statistic information 40 are included in the same screen and output. FIG. 10 is a diagram illustrating a screen including the histogram 90 and the test statistic information 40. In FIG. 10, the histogram 90 is included in the test statistic graph 70 described above. As shown in FIG. 11, which will be described later, the test statistic graph 70 and the histogram 90 may be displayed separately.
[0075]
　The histogram 90 is generated, for example, as follows. First, the generation unit 2060 divides the numerical range of the factor data 22 by each boundary value to generate a partial range. The generation unit 2060 counts the number of target data 20 to which the factor data 22 belongs to the subrange for each subrange. Then, the generation unit 2060 generates a histogram 90 showing the correspondence between the partial range and the number of target data 20.
[0076]
　By outputting the histogram 90 in addition to the test statistic information 40 in this way, the user can grasp how the factor data 22 is distributed and set the factor data 22 to what value. It is possible to accurately and easily grasp whether the result can be improved by doing so.
[0077]
The
　generation unit 2060 may generate various other information in addition to the above-mentioned information. FIG. 11 is a diagram showing a more specific example of the information generated by the generation unit 2060. In this example, the screen 130 is generated by the generation unit 2060. The screen 130 includes a display area 131, a display area 132, a display area 133, a display area 134, and a display area 135.
[0078]
　In the display area 131, a time series graph 100 showing the time change of the target data 20 is shown. In the time series graph 100, the line 102 represents the time change of the factor data 22. In this case, for example, the target data set 10 is time series data in which each target data 20 is associated with time. However, it may be regarded as time series data and used by aggregating the data in a specific cycle.
[0079]
　The histogram 90 is displayed in the display area 132. In the display area 133, an input interface for specifying the test method and the test particle size (corresponding to the number of divisions), and a button for instructing the execution of the test are displayed.
[0080]
　In the display area 134, an input interface for designating the significance level and a test statistic graph 70 are displayed. The test statistic graph 70 of FIG. 11 includes bars 72 and explanatory display 76. The bar 72 is an interface for designating a boundary value of interest. The user can change the boundary value to be focused on by moving the bar 72 left and right. The explanatory display 76 shows the boundary value and the test statistic for the boundary value specified by the bar 72.
[0081]
　The display mode of the histogram 90 is different between a rank smaller than the boundary value specified by the bar 72 and a rank larger than the boundary value. Specifically, the data of each rank smaller than the boundary value b3 specified by the bar 72 and the data of each rank larger than the boundary value b3 are given different patterns. The display mode to be different is not limited to the pattern, and may be a color, a border type, or the like.
[0082]
　Further, in the display area 132, a polygonal line 110 indicating the defect rate for each rank is shown by superimposing the histogram 90 on the display area 132. Furthermore, the confidence interval of the defect rate in each rank is shown by the dotted line.
[0083]
　The display area 135 shows a simulation of the degree of improvement of the value with respect to the result data 24 based on the boundary value specified by the bar 72. For example, suppose a test statistic is calculated for the defective rate. Further, it is assumed that the defect rate is lowered by making the value of the factor data 22 smaller. In this case, the display area 135 includes a display area 136 indicating the defect rate of the entire target data 20, and a display area indicating the defect rate when the factor data 22 is limited to a value equal to or less than the boundary value specified by the bar 72. 137 is included.
[0084]
　The information shown in the display area 137 indicates how much improvement can be expected by adjusting the factor (in this example, the content of component A) so that the factor data 22 is equal to or less than the specified boundary value. It can be said that it is. Therefore, according to such a display, the expected effect (improvement of the defective rate in this example) by adjusting the factors can be easily grasped.
[0085]
　Here, as described above, the user can change the designated boundary value by moving the bar 72 left and right. The generation unit 2060 changes the content of the display area 137 according to the movement of the bar 72 (the designation of the boundary value is changed). Therefore, the user can confirm the expected effect by adjusting the factors while moving the bar 72.
[0086]
　In addition, the generation unit 2060 also changes the display mode of the data in the histogram 90 according to the movement of the bar 72. By doing so, the user can easily confirm the sample used for the simulation of improvement.
[0087]
　In the example of FIG. 11, since the defect rate is lowered by making the factor data 22 smaller, "adjustment to b3 or less" is displayed in the display area 137. In this regard, when the defect rate is lowered by making the factor data 22 larger, this display is "adjusted to b3 or more".
[0088]
　Information for specifying whether the factor data 22 should be reduced or increased for improvement is acquired in advance. For example, as described above, it is assumed that the target data 20 uses learning data for learning an estimation model that estimates the value of the objective variable from the values ​​of a plurality of explanatory variables. In this case, depending on whether a certain explanatory variable affects the objective variable in the estimation model, it is specified whether the factor data 22 corresponding to the explanatory variable should be reduced or increased for improvement. be able to.
[0089]
[Modification Example]
　In the examples described so far, the target data 20 associates one factor data 22 with one result data 24. However, the target data 20 may be a combination of two factor data 22 and one result data 24. In this case, when the target data set 10 is divided into two samples, the division is performed by paying attention to two factors. Hereinafter, the analyzer 2000 that performs such processing will be referred to as a modified example analyzer 2000.
[0090]
　In the analysis device 2000 of the modified example, the acquisition unit 2020 obtains the first factor data 22 (hereinafter, factor data 22-1), the second factor data 22 (hereinafter, factor data 22-2), and the result data 24. The target data set 10 including a plurality of associated target data 20 is acquired.
[0091]
　The calculation unit 2040 divides each of the numerical range of the factor data 22-1 and the numerical range of the factor data 22-2 by the number of divisions indicated by the division number information 30. The number of divisions information 30 may indicate a common number of divisions for the factor data 22-1 and the factor data 22-2, or may indicate a different number of divisions for each. For example, assume that the number of divisions of the factor data 22-1 is 4 and the number of divisions of the factor data 22-2 is 5. In this case, the calculation unit 2040 divides the numerical range of the factor data 22-1 into four equal parts and divides the numerical range of the factor data 22-2 into five equal parts. That is, the calculation unit 2040 specifies three boundary values ​​for the factor data 22-1 and four boundary values ​​for the factor data 22-2.
[0092]
　The calculation unit 2040 divides the target data set 10 into two for each of a plurality of pairs of the boundary value for the factor data 22-1 and the boundary value for the factor data 22-2 at the boundary defined by the pair. Generate a first sample and a second sample. Then, the test statistic is calculated for the generated first sample and the second sample as described above.
[0093]
　FIG. 12 is a diagram illustrating a first sample and a second sample in the analyzer 2000 of the modified example. In this example, the first sample and the second sample are generated at the boundary determined by the pair of the boundary value b2 of the factor data 22-1 and the boundary value c2 of the factor data 22-2. Specifically, the calculation unit 2040 uses a set of target data 20 satisfying the condition that "factor data 22-1 is b2 or less and factor data 22-2 is c2 or less" as the first sample 12. A set of target data 20 that satisfies the condition that "factor data 22-1 is larger than b2 or factor data 22-2 is larger than c2" is generated as the second sample 14.
[0094]
　However, as described above, as a method of dividing the target data set 10 into two samples by using two boundary value pairs, a method other than the method illustrated in FIG. 1 (hereinafter, the first method) can be adopted. FIG. 13 is a diagram illustrating another method of dividing the target data set 10 into two samples using two boundary value pairs. As shown in FIG. 13, in addition to the above-mentioned methods, "factor data 22-1 is below the boundary value and factor data 22-2 is above the boundary value" and "factor data 22-1 is larger than the boundary value." Alternatively, the second method of dividing the factor data 22-2 into "smaller than the boundary value", "factor data 22-1 is equal to or more than the boundary value and factor data 22-2 is equal to or less than the boundary value" and "factor data 22-1". Is smaller than the boundary value or factor data 22-2 is larger than the boundary value ", and" factor data 22-1 is greater than or equal to the boundary value and factor data 22-2 is greater than or equal to the boundary value ". And "factor data 22-1 is smaller than the boundary value, or factor data 22-2 is smaller than the boundary value". Which of these methods is used to divide the target data set 10 may be predetermined or may be selected by the user.
[0095]
　FIG. 14 is a diagram illustrating information output by the analyzer 2000 of the modified example. The screen 140 shown in FIG. 14 is an application of the information represented by the screen 130 in FIG. 11 to a case in which two factors are used.
[0096]
　The screen 140 has display areas 141 to 145. The display area 141 includes a time series graph 100 showing a time change for each of the factor data 22-1 and the factor data 22-2. The display area 142 includes a histogram 90-1 for factor data 22-1 and a histogram 90-2 for factor data 22-2.
[0097]
　The display area 143, like the display area 133, includes an input interface for specifying the test method and the test particle size, and a button for instructing the execution of the test. Further, the display area 143 includes an input interface 146 that specifies a method for dividing the first sample and the second sample. The input interface 146 has four areas composed of a rectangle divided into four, and the user can specify any one of the four areas. The designated area is colored.
[0098]
　The input interface 146 mimics a plane consisting of the numerical ranges of the two factor data 22 illustrated in FIGS. 12 and 13. When the user specifies one of the four areas, the target data set 10 is divided into two samples so that the plane consisting of the numerical ranges of the two factor data 22 is divided into the specified area and the other areas. It is divided. For example, in the example of FIG. 14, the upper left area is designated. Therefore, the target data set 10 is divided into two samples: "factor data 22-1 is below the boundary value and factor data 22-2 is below the boundary value" and other samples.
[0099]
　The display area 144 represents the test statistic graph 150. The calculation unit 2040 calculates test statistics for two samples obtained from each of a plurality of pairs of the boundary value of the factor data 22-1 and the boundary value of the factor data 22-2, and the plurality of them. Statistical information 40 (here, test statistic graph 150) representing the test statistic is generated. Test statistic Each cell in graph 150 represents the test statistic obtained for the pair of boundary values ​​corresponding to that cell. For example, in the example of FIG. 14, each cell represents the magnitude of the test statistic obtained for the boundary value pair represented by the lower right corner of the cell.
[0100]
　In the test statistic graph 150, the magnitude of the test statistic is represented by a color, a pattern, or the like. For example, the larger the test statistic, the darker the color or the denser the pattern. In the example of FIG. 14, the larger the test statistic, the larger the dot pattern is attached. Further, a frame 152 is attached to a cell whose test statistic is equal to or higher than the threshold value th.
[0101]
　The user can specify the pair of boundary values ​​that he / she wants to pay attention to. Specifically, the user specifies one cell corresponding to the pair of boundary values ​​to be focused on. In response to this designation, the generation unit 2060 generates information regarding the improvement of the defective rate for the designated cell and the sample pair generated by the designated sample division method.
[0102]
　In the example of FIG. 14, a cell representing a boundary value pair of "boundary value of component A = b2, boundary value of component B = c2" is designated. In addition, a method of dividing the plane of the numerical range into the upper left and the other is specified. Therefore, it is divided into a sample of "value of component A ≤ b2 and value of component B ≤ c2" and a sample of "value of component A> b2 or value of component B> c2". In addition, the display mode of the histograms 90-1 and 90-2 is changed according to this designation. Specifically, as described in the first embodiment, the ranks below the designated boundary value and the other ranks are set to have different colors and patterns.
[0103]
　The generation unit 2060 displays information indicating a simulation of improvement of the defective rate in the display area 145 for the above-mentioned division. In the display area 145, the number of defective products and the number of non-defective products are obtained with respect to the target data 20 satisfying the condition that "component A is set to b2 or less and component B is set to c2 or less" and the defective rate in all the target data 20. A comparison of the number of and the defective rate is shown. Through such a comparison, the user can easily understand how much the defect rate can be improved by manufacturing the product under the manufacturing condition that "component A is set to b2 or less and component B is set to c2 or less". can do.
[0104]
　Although the embodiments of the present invention have been described above with reference to the drawings, these are examples of the present invention, and a combination of the above embodiments or various configurations other than the above can be adopted.
[0105]
　Some or all of the above embodiments may also be described, but not limited to:
1. 1. A set of target data indicating a pair of factor data which is a value related to a factor and a result data which is a value related to a result, an acquisition unit which acquires divided number information indicating the number of divided numerical ranges of the factor data, and the
　numerical value . For each of the plurality of boundary values ​​that divide the range into the number, a test statistic representing the difference regarding the result data in two samples obtained by dividing the target data included in the target data set into two by the boundary value. An analyzer having a calculation unit for calculating the
　above boundary value and a generation unit for generating test statistic information indicating a plurality of test statistic calculated for each boundary value.
2. The result data indicates the presence or absence of defects, and the
　calculation unit calculates the defect rate in each of the two samples for each boundary value, and calculates a test statistic representing the difference between the calculated two error rates. 1. 1. The analyzer described in.
3. 3. The factor data indicates the value of an index indicating the manufacturing conditions of the product, and the
　result data indicates whether or not the product is a defective product. The analyzer described in.
4. The test statistic information is a graph showing a combination of the boundary value and the test statistic. The analyzer according to any one of 3 to 3.
5. The generator includes, in the graph, a display representing the threshold of the test statistic representing the significance level. The analyzer described in.
6. The generation unit outputs a histogram showing the distribution of the factor data together with the graph. Or 5. The analyzer described in.
7. The generation unit
　　accepts an input for designating a boundary value for the graph,
　　and in response to receiving the input, in the histogram, data of ranks corresponding to each boundary value equal to or less than the specified boundary value. 6. The display mode is different from the display mode of the data of the rank corresponding to each boundary value larger than the specified boundary value. The analyzer described in.
8. The target data indicates a set of the first factor data, the second factor data, and the result data, and the
　calculation unit divides the numerical range of the first factor data into the number. For each set of the first boundary value and the plurality of second boundary values ​​that divide the numerical range of the second factor data into the number, the target data included in the target data set is the boundary value. 1. Calculate the test statistics for the two samples obtained by dividing into two based on the set. From 7. The analyzer according to any one.
9. A control method executed by a computer, in which
　a set of target data indicating a pair of factor data, which is a value related to a factor, and result data, which is a value related to a result, and a division indicating the number of divisions of a numerical range of the factor data.
　In two samples obtained by dividing the target data included in the set of target data into two for each of the acquisition unit for acquiring the number information and the plurality of boundary values ​​for dividing the numerical range into the number. , A calculation unit that calculates a test statistic representing the difference in the result data,
　A control method having a generation unit for generating test statistic information indicating a plurality of test statistic calculated for each boundary value.
10. The result data indicates the presence or absence of defects, and in the
　calculation step, the defect rate in each of the two samples is calculated for each boundary value, and the test statistic representing the difference between the calculated two error rates is calculated. 9. The control method described in.
11. The factor data indicates the value of an index indicating the manufacturing conditions of the product, and the
　result data indicates whether or not the product is a defective product. The control method described in.
12. The test statistic information is a graph showing a combination of the boundary value and the test statistic. From 11. The control method according to any one.
13. In the generation step, the graph includes a display representing the threshold of the test statistic representing the significance level. The control method described in.
14. In the generation step, a histogram showing the distribution of the factor data is output together with the graph. Or 13. The control method described in.
15. In the generation step,
　　an input for designating a boundary value is accepted for the graph,
　　and in response to the acceptance of the input, the rank data corresponding to each boundary value equal to or less than the specified boundary value in the histogram. The display mode is different from the display mode of the data of the rank corresponding to each boundary value larger than the specified boundary value. The control method described in.
16. The target data indicates a set of the first factor data, the second factor data, and the result data, and
　in the calculation step, a plurality of numerical ranges of the first factor data are divided into the number. For each set of the first boundary value and the plurality of second boundary values ​​that divide the numerical range of the second factor data into the number, the target data included in the target data set is the boundary value. 9. Calculate the test statistics for the two samples obtained by dividing into two based on the set. To 15. The control method according to any one.
17. 9. From 16. A program that causes a computer to execute the control method described in any one of them.
[0106]
　This application claims priority on the basis of Japanese Application Japanese Patent Application No. 2019-199613 filed on November 1, 2019, and incorporates all of its disclosures herein.
Code description
[0107]
10 Target data set
12 1st sample
14 2nd sample
20 Target data
22 Factor data
24 Result data
30 Divided number information
40 Test statistics information
50 Table
52 Boundary value
54 Test statistics
60 User terminal
70 Test statistics graph
72 Broken line
72 Bar
74 Threshold display
76 Explanation display
80 Screen
82, 84, 86 Input area
88 Button
90 Schematic
100 Time series graph
102 Fold line
110 Fold line
130 Screen
131, 132, 133, 134, 135, 136, 137 Display area
140 Screen
141 , 142 , 143, 144, 145 Display area
146 Input interface
150 Test statistic graph
152 Frame
1000 Computer
1020 Bus
1040 Processor
1060 Memory
1080 Storage device
1100 Input / output interface
1120 Network interface
2000 Analyzer
2020 Acquisition unit
2040 Calculation unit
2060 Generation unit
The scope of the claims
[Claim 1]
　A set of target data indicating a pair of factor data which is a value related to a factor and a result data which is a value related to a result, an acquisition unit which acquires divided number information indicating the number of divided numerical ranges of the factor data, and the
　numerical value . For each of the plurality of boundary values ​​that divide the range into the number, a test statistic representing the difference regarding the result data in two samples obtained by dividing the target data included in the target data set into two by the boundary value. An analyzer having a calculation unit for calculating the
　above boundary value and a generation unit for generating test statistic information indicating a plurality of test statistic calculated for each boundary value.
[Claim 2]
　The result data indicates the presence or absence of defects, and the
　calculation unit calculates the defect rate in each of the two samples for each boundary value, and calculates a test statistic representing the difference between the calculated two error rates. The analyzer according to claim 1.
[Claim 3]
　The analyzer according to claim 2, wherein the factor data indicates a value of an index representing a manufacturing condition of a product, and the
　result data indicates whether or not the product is a defective product.
[Claim 4]
　The analyzer according to any one of claims 1 to 3, wherein the test statistic information is a graph showing a combination of the boundary value and the test statistic.
[Claim 5]
　The analyzer according to claim 4, wherein the generation unit includes a display representing a threshold value of the test statistic representing the significance level in the graph.
[Claim 6]
　The analyzer according to claim 4 or 5, wherein the generation unit outputs a histogram showing the distribution of the factor data together with the graph.
[Claim 7]
　A control method executed by a computer, in which
　a set of target data indicating a pair of factor data, which is a value related to a factor, and result data, which is a value related to a result, and a division indicating the number of divisions of a numerical range of the factor data.
　In two samples obtained by dividing the target data included in the set of target data into two for each of the acquisition unit for acquiring the number information and the plurality of boundary values ​​for dividing the numerical range into the number. A control method including a calculation unit for calculating a test statistic representing a difference in the result data, and
　a generation unit for generating test statistic information indicating a plurality of test statistic calculated for each boundary value.
[Claim 8]
　A program that causes a computer to execute the control method according to claim 7.