Information processing device, information processing method, and information processing program

Abstract

To provide an information processing device capable of evaluating the quality of learning data for generating a learning model.SOLUTION: An information processing device provided herein comprises: an acquisition unit for acquiring learning data for generating a learning model, a first computation unit configured to compute a first index indicative of a degree of the amount of the learning data based on the learning data and a first parameter corresponding to the type of the learning model, and a second computation unit configured to compute a second index indicative of distribution uniformity of the learning data on the basis of the learning data.SELECTED DRAWING: Figure 7

Description

[Technical field]

[0001] The present invention relates to an information processing device, an information processing method, and an information processing program. [Background technology]

[0002] 2. Description of the Related Art A method is known in which a learning model is generated for performing abnormality diagnosis of equipment, such as a factory equipment, using outputs from sensors provided in the equipment.

[0003] For example, Patent Document 1 describes a method in which a diagnostic model is generated by applying multivariate statistical process control (MSPC) to normal vibration data output from a vibration sensor as learning data. Then, the generated diagnostic model is used to execute an abnormality diagnosis on the vibration data of the diagnosis target. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2004-279056 A Summary of the Invention [Problem to be solved by the invention]

[0005] Incidentally, the reliability of a learning model generally depends on the quality of the learning data, so in order to generate a highly reliable learning model, it is necessary to use learning data having high quality.

[0006] However, in the method described in Patent Document 1, evaluation of the quality of the learning data used when generating a learning model is not taken into consideration.

[0007] An object of the present invention is to provide an information processing device capable of evaluating the quality of learning data for generating a learning model. [Means for solving the problem]

[0008] One invention for achieving the above object is an information processing device including: an acquisition unit that acquires learning data for generating a learning model; a first calculation unit that calculates a first index indicating the amount of the learning data based on the learning data and a first parameter corresponding to the type of the learning model; and a second calculation unit that calculates a second index indicating the uniformity of distribution of the learning data based on the learning data. Other features of the present invention will become clear from the description of this specification. Effect of the Invention

[0009] According to the present invention, it is possible to provide an information processing device capable of evaluating the quality of learning data for generating a learning model. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is a diagram illustrating an information processing system. [Diagram 2] FIG. 1 illustrates an example of a hardware configuration of an information processing device. [Diagram 3] FIG. 2 is a diagram illustrating an example of functional blocks realized in the information processing device. [Figure 4] FIG. 13 is a diagram illustrating an example of learning data. [Diagram 5] FIG. 13 is a diagram for explaining an index I1. [Figure 6] FIG. 13 is a diagram for explaining index I3. [Figure 7] 11 is a flowchart illustrating the process flow from generating a learning model to evaluating the quality of learning data. [Figure 8] FIG. 13 illustrates an example of a setting screen. [Figure 9] FIG. 13 is a diagram showing an example of an output screen. [Figure 10] FIG. 13 illustrates an example of a setting screen. [Figure 11] FIG. 13 is a diagram showing an example of an output screen. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] == Implementation form == <<Information Processing System>> FIG. 1 is a diagram showing an overview of an information processing system 1 according to this embodiment. The information processing system 1 is a system for generating a learning model used for abnormality diagnosis of a device to be diagnosed. The information processing system 1 is also a device for evaluating the quality of learning data used in generating the learning model when generating the learning model. The information processing system 1 includes a plurality of sensors 2 and an information processing device 3. Each of them will be described below.

[0012] <Sensor 2> The sensors 2 are attached to various devices provided in a plant P that is the diagnosis target. In this embodiment, six sensors 21 to 26 are attached to various devices (not shown) provided in the plant P.

[0013] <Information processing device> The information processing device 3 is a device for generating a learning model using learning data based on outputs from the sensors 21 to 26, and for evaluating the quality of the learning data. Below, the hardware configuration of the information processing device 3 and the realized functional blocks will be described in that order.

[0014] Hardware Configuration 2 is a diagram showing an example of a hardware configuration of an information processing device 3. The illustrated information processing device 3 includes a processor 300, a main memory device 301, an auxiliary memory device 302, an input device 303, an output device 304, and a communication device 305. Note that the illustrated information processing device 3 may be realized, in whole or in part, by using virtual information processing resources provided by using a virtualization technique, a process space separation technique, or the like, such as a virtual server provided by a cloud system. In addition, all or in part of the functions provided by the information processing device 3 may be realized by a service provided by a cloud system via an API (Application Programming Interface), for example.

[0015] The processor 300 is configured using, for example, a CPU (Central Processing Unit) and the like.

[0016] The main memory device 301 is a device that stores programs and data, and is, for example, a read only memory (ROM) or a random access memory (RAM).

[0017] The auxiliary storage device 302 is, for example, a hard disk drive, a storage area of ​​a cloud server, etc. Programs and data can be read into the auxiliary storage device 302 via a recording medium reading device or a communication device 305. The programs and data stored (memorized) in the auxiliary storage device 302 are read into the main storage device 301 as needed.

[0018] The input device 303 is an interface that accepts input from the outside, and is, for example, a keyboard, a mouse, a touch panel, or the like.

[0019] The output device 304 is an interface that outputs various information such as the process progress, the process results, etc. The output device 304 is, for example, a display device (such as a liquid crystal monitor) that visualizes the various information described above, a device that converts the various information described above into voice (a voice output device (such as a speaker)), or a device that converts the various information described above into text (such as a printer).

[0020] The input device 303 and the output device 304 constitute a user interface that receives information from the user and presents information.

[0021] The communication device 305 is a device that realizes communication with other devices. The communication device 305 is a wired or wireless communication interface that realizes communication with other devices via a communication network NW such as the Internet.

[0022] The information processing device 3 may have installed therein, for example, an operating system, a file system, and the like.

[0023] Each function of the information processing system 1 is realized by a processor 300 of the information processing device 3 reading and executing a program stored in a main memory device 301, or by hardware constituting the information processing system 1. The information processing system 1 stores various types of information (data), for example, as a table of a database or a file managed by a file system.

[0024] Functional blocks 3 is a diagram showing an example of functional blocks realized in the information processing device 3. When the processor 300 of the information processing device 3 executes a predetermined program, an acquisition unit 310, a reception unit 311, a generation unit 312, a first calculation unit 313, a second calculation unit 314, a third calculation unit 315, a fourth calculation unit 316, an alarm unit 317, and a display unit 318 are realized in the information processing device 3. Each of these will be described below.

[0025] [Acquisition unit 310] The acquiring unit 310 acquires learning data for generating a learning model, which is a target of quality evaluation. The learning data is the output of a plurality of sensors 2 provided in the device to be diagnosed.

[0026] 4 is a diagram showing learning data D, which is an example of learning data acquired by the acquiring unit 310. The learning data D in this embodiment includes 30 pieces of data d (d1 to d30). Each of the 30 pieces of data d has six variables v1 to v6 corresponding to the six sensors 21 to . Each of the six variables is an output value from the six sensors 21 to at the same time.

[0027] The learning data D includes data classified into two classes. Specifically, the learning data D includes data (corresponding to the first data) classified into an abnormal class (corresponding to the first class) and data (corresponding to the second data) classified into a normal class (corresponding to the second class).

[0028] In this embodiment, the number of data items classified into the abnormal class is two, ie, data items d1 and d2, and the number of data items classified into the normal class is 28, ie, data items d3 to d30.

[0029] [Reception Department 311] The receiving unit 311 receives from a user of the information processing system (hereinafter simply referred to as "user") a selection of information on a learning model desired by the user, and input of parameters P1 to P3 used when evaluating the quality of learning data.

[0030] In this embodiment, the "information on the learning model" refers to whether the learning model is based on supervised learning or unsupervised learning, and the type of the learning model. The parameters P1 to P3 will be described later.

[0031] [Generation unit 312] The generation unit 312 uses the learning data D to generate a learning model the selection of which is accepted by the acceptance unit 311 (hereinafter, simply referred to as a “learning model” unless otherwise specified).

[0032] [1st calculation section 313] The first calculation unit 313 calculates an index I1 (corresponding to a first index). The index I1 is an index indicating the amount of learning data D (hereinafter, may be referred to as "completeness of learning data"). The amount of learning data D is the number of data d included in the learning data D.

[0033] The first calculation unit 313 calculates the index I1 based on the learning data D and a parameter P1 (corresponding to the first parameter) according to the type of learning model. The parameter P1, the index I1, and the relationship between them will be described in detail below.

[0034] The parameter P1 is a parameter for specifying the number L of data d required when generating a learning model, and is a parameter that is set in advance by the user.

[0035] The number L of data d required may differ depending on the type of learning model. For example, when the learning model is a model generated based on MSPC, the number L increases in proportion to the number n of variables that the data d has. In particular, the number L is preferably about 10 times or more the number n of variables. In this embodiment, the number of variables is 6, so the number L of data d required is 60.

[0036] As an example of the parameter P1, the parameter P1 may directly indicate the number L. That is, in this case, the following (Equation 1-1) can be used as the relational expression between L and P1. (Formula 1-1) L=P1 In this case, in this embodiment, the parameter P1 may be set to 60 in advance.

[0037] As another example of the parameter P1, the parameter P1 may specify the number L of required data d from a calculation formula using the number n of variables. In this case, the following (Formula 1-2) can be used as the relational formula between L and P1. (Formula 1-2) L=n×P1

[0038] In this case, in this embodiment, since n is 6, the parameter P1 may be preset to 10. By using such a formula, the number L of required data d can be specified when the user sets the parameter P1 without needing to take into account the number n of variables.

[0039] In this embodiment, the parameter P1 is defined using (Equation 1-2). In this case, the first calculation unit 313 calculates the index I1 based on the learning data D, the number of the multiple sensors 2, and the parameter P1.

[0040] Next, the relationship between the index I1 and the number L of required data d will be described. As an equation showing the index I1, for example, the following (Equation 2-1) and (Equation 2-2) can be adopted, which are divided into cases using the number L and the number m of data d included in the learning model D. (Formula 2-1) I1=1 (m≧L) (Formula 2-2) I1=m / L (m <L)

[0041] 5 is a diagram for explaining the index I1, and is a graph showing (Equation 2-1) and (Equation 2-2). As can be seen from these equations and graphs, the index I1 is 1 when the number m of data d included in the learning data D is L or more, and is a monotonically increasing linear function when it is smaller than L. When the number m of data d is 0, the index I1 is 0. In other words, the index I1 takes a value in the range from 0 to 1, and the closer to 1 the index I1 is, the more preferable it is.

[0042] [Second calculation section 314] The second calculation unit 314 calculates an index I2 (corresponding to a second index). The index I2 is an index indicating the uniformity of the distribution of the training data D. The second calculation unit 314 calculates the index I2 based on the training data D.

[0043] The second calculation unit 314 calculates the index I2 using a parameter P2 (corresponding to the second parameter). The parameter P2 is a parameter that is set in advance by a user. The parameter P2, the index I2, and the relationship between them will be described in detail below.

[0044] In this embodiment, the parameter P2 is a parameter indicating the number of divisions into which the range of values ​​included in the learning data D is divided. Note that in this embodiment, the number of sensors 2 is six, and therefore the number of variables that the data d has is six. However, for convenience of explanation, a case where the number of variables that the data d has is one (the number of sensors 2 is one) will be explained first. After that, an extension to a case where the number of variables that the data d has (the number of sensors 2) is multiple will be explained.

[0045] When calculating the data index I2, the second calculation unit 314 divides the range of the variable values ​​in the learning data D into P2 regions. Among the number of data d included in each of the multiple regions divided into P2 regions, the maximum number is defined as m_max, and the minimum number is defined as m_min. Then, the index I2 is defined by the following (Equation 3-1). (Formula 3-1) I2=1-(m_max-m_min) / m_max

[0046] For example, if the number of data d contained in each of the multiple regions divided into P2 is the same, m_max and m_min are equal, so index I2 will have a maximum value of 1. Also, the minimum value of index I2 is 0. In other words, the value of index I2 is in the range from 0 to 1, and the closer to 1, the higher the uniformity.

[0047] As described above, I2 is defined when the number of variables is 1. On the other hand, when the number of variables is multiple, for example, the definition of the index I2 may be expanded as follows.

[0048] When there are multiple variables, the index I2 is calculated for each of the multiple variables using the above (Equation 3-1). Then, the minimum value of the indexes I2 for each of the multiple variables can be set as the index I2 for the learning data D.

[0049] Note that the definition of the index I2 is not limited to this. For example, the average value of the indexes I2 for each of a plurality of variables may be set as the index I2 for the learning data D.

[0050] [Third calculation section 315] The third calculation unit 315 calculates an index I3 (corresponding to a third index). The index I3 is an index indicating how far a ratio r based on the number m_NG (the number of first data) of data d classified into the abnormal class and the number m_OK (the number of second data) of data d classified into the normal class is from a reference ratio R.

[0051] In this embodiment, the ratio r based on m_NG and m_OK is the ratio between m_NG and m_OK. However, this is not limited to this, and the ratio r based on m_NG and m_OK may be the ratio of m_NG or m_OK to the number of all data d, etc.

[0052] The third calculation unit 315 calculates the index I3 based on the learning data D and the parameter P3 (corresponding to the third parameter). The parameter P3, the index I3, and the relationship between them will be described in detail below.

[0053] The parameter P3 is a parameter that indicates the ratio R of the above-mentioned criteria, and is a parameter that is set in advance by the user.

[0054] The index I3 is defined by the following (Equation 4-1) and (Equation 4-2) for different cases. (Formula 4-1) I3=1 (m_NG / m_OK≧P3) (Formula 4-2) I3=m_NG / m_OK (m_NG / m_OK <P3)

[0055] 6 is a diagram for explaining the index I3, and is a graph showing (Formula 4-1) and (Formula 4-2). As can be seen from these formulas and graphs, the index I3 is 1 when the ratio r is equal to or greater than the ratio R, and is a monotonically increasing linear function when the ratio r is smaller than R. When the ratio r is 0, the index I3 is 0. In other words, the value of the index I3 is in the range of 0 to 1, and the closer to 1 the better.

[0056] In addition, when the learning model is based on unsupervised learning, it is not necessary to evaluate the index I3 for the learning data D for generating the learning model based on unsupervised learning. Therefore, in this case, the third calculation unit 315 does not calculate the index I3.

[0057] [Fourth calculation section 316] The fourth calculation unit 316 calculates an index I4 (corresponding to a fourth index) based on the calculated multiple indexes. The index I4 is an index for comprehensively evaluating the learning data D.

[0058] In this embodiment, the fourth calculation unit 316 sets the index I4 to the average value of the calculated indices. That is, when the learning model is based on unsupervised learning, the index I4 can be set to, for example, the average value of the indexes I1 and I2. On the other hand, when the learning model is based on supervised learning, the index I4 can be set to the average value of the indexes I1, I2, and I3.

[0059] Note that the present invention is not limited to the above example, and the fourth calculation unit 316 may calculate the index I4 using weighting factors W1 to W3 (corresponding to first to third weighting factors) for the indexes P1 to P3, respectively.

[0060] For example, the index I1 has a larger effect on the reliability of the learning model than the other indexes, so it is preferable that the weighting factor W1 for the index I1 is larger than the weighting factors W2 and W3 for the indexes I2 and I3.

[0061] By using such a weighting coefficient, it is possible to increase the contribution of index I1, which has the greatest effect on the reliability of the learning model, to index I4, thereby improving the accuracy of the overall evaluation of the learning data D by index I4.

[0062] In this embodiment, the fourth calculation unit 316 calculates the index I4 using weighting factors W1 to W3, where the weighting factors W1, W2, and W3 are set to 0.5, 0.2, and 0.3, respectively.

[0063] [Display section 318] The display unit 318 displays the calculated multiple indexes. The display unit 318 generates a display screen for displaying these indexes and displays them on the output device 304.

[0064] [Alarm section 317] The warning unit 317 issues a warning when at least one of the calculated indices is not within a predetermined range.

[0065] The predetermined range is a range that is set in advance for each of the indices I1 to I3. The user sets a range in advance for each of the indices I1 to I3. This range may be, for example, a range (tolerable range) for the generated learning model to have a minimum reliability, or may be a range (recommended range) for the generated learning model to have a desired reliability.

[0066] In this embodiment, the user sets the allowable range as a predetermined range. In this embodiment, the user sets the allowable range of the index I1 to 0.8 or more, the allowable range of the index I2 to 0.1 or more, and the allowable range of the index I3 to 0.8 or more.

[0067] The information processing device 3 according to the embodiment has been described above. According to the configuration of the information processing device 3 as described above, when generating a learning model, it is possible to evaluate the quality of the learning data D for generating the learning model. By considering the evaluation result of the quality of the learning data D, the user can estimate the reliability of the generated learning model.

[0068] If the evaluation results show that the generated learning model has sufficient reliability, the user can use the generated learning model for actual anomaly diagnosis with peace of mind.

[0069] On the other hand, if the evaluation result indicates that the generated learning model is not sufficiently reliable, the user can try to improve the learning data D by, for example, adding new data to the learning data D or deleting some data from the learning data D. Then, based on the improved learning data D, a learning model can be generated again and the quality of the new learning data can be evaluated.

[0070] <Processing from generating a learning model to evaluating the quality of the learning data> ==For MSPC== (Unsupervised) 7 is a flowchart explaining a process flow from when the information processing device 3 generates a learning model to when the information processing device 3 evaluates the quality of the learning data. This process includes steps S101 to S112. Note that, here, an example will be explained in which MSPC is selected as the learning model.

[0071] First, in step S101, the acquiring unit 310 acquires learning data. The learning data D shown in FIG.

[0072] Next, in step S102, the reception unit 311 receives a selection from the user as to whether the learning model to be generated is based on supervised learning or unsupervised learning. The reception unit 311 receives this selection via the setting screen 40.

[0073] 8 is a diagram showing an example of a setting screen 40. A user selects whether to use supervised learning or unsupervised learning via the setting screen 40. The user also selects a learning model, which will be described later, and inputs parameters P1 to P3 via the setting screen 40.

[0074] In the example of FIG. 8, the setting screen 40 shows a state in which the button 40a for selecting unsupervised learning is selected out of the buttons 40a and 40b.

[0075] Next, in step S103, the reception unit 311 receives a selection of a learning model from the user. The reception unit 311 receives the selection of a learning model from the user via the setting screen 40 shown in Fig. 8. The example in Fig. 8 shows a state in which MSPC is selected as the learning model by selecting MSPC from the drop-down list 40c.

[0076] After step S103, the process branches to step S104 and steps S105 to S112. The process of step S104 is a process for generating a learning model, and the processes of steps S105 to S112 are a process for evaluating the quality of the learning data D. In this embodiment, these processes are executed in parallel.

[0077] First, a process of generating a learning model in step S104 will be described. In step S104, the generation unit 312 generates the learning model selected in step S103 using the learning data D. In this example, the generation unit 312 generates a learning model based on MSPC.

[0078] Next, the process of evaluating the quality of the learning data D in steps S105 to S112 will be described.

[0079] In step 105, the reception unit 311 receives input of parameters P1 and P2 from the user via the setting screen 40 shown in Fig. 8. The user inputs parameters P1 and P2 in input windows 40e and 40f, respectively. Note that since parameter P3 is unnecessary in unsupervised learning, it is not possible to input parameter P3 in input window 40g on the setting screen 40 in Fig. 8.

[0080] In step S106, the first calculation unit 313 calculates the index I1 based on the learning data D and the parameter P1 whose input is accepted by the acceptance unit 311 in step S105.

[0081] In step S107, the second calculation unit 314 calculates the index I2 based on the learning data D, using the parameter P2 whose input was accepted by the acceptance unit 311 in step S105.

[0082] In step 108, the third calculation unit 315 calculates the index I3 based on the learning data D and the parameter P3 whose input was accepted by the acceptance unit 311 in step S105.

[0083] In addition, when the reception unit 311 receives a selection of unsupervised learning in step S100, the index I3 is not evaluated for the learning data D for generating a learning model based on unsupervised learning. Therefore, in this case, step S108 is omitted.

[0084] In step S109, the fourth calculation unit 316 calculates an index I4. In this embodiment, the index I4 is a weighted average value of the index I1 and the index I2.

[0085] In step S110, the display unit 318 displays the indices calculated in steps S106 to S109 as an output screen. Fig. 9 is a diagram showing an output screen 41, which is an example of an output screen displayed on the output device 304. In this example, the output screen 41 shows the values ​​of the indices I1, I2, and I4.

[0086] In this result, the value of index I1 is 0.5. As described above, the preset tolerance range for index I1 is 0.8 or more. In other words, the value of index I1 is not within the tolerance range. On the other hand, the value of index I2 is 0.11. The preset tolerance range for index I2 is 0.1 or more. In other words, the value of index I2 is within the tolerance range.

[0087] In this way, if either the calculated value of the index I1 or I2 is not within the allowable range (step S110: Y), the warning unit 317 issues a warning.

[0088] 9 shows a warning pop-up 41a. The pop-up 41a displays a warning that the index I1 is too small relative to the allowable range. Furthermore, the pop-up 41a displays a message recommending that the number of data included in the learning data D be increased to 60 or more.

[0089] If the values ​​of both the calculated indices I1 and I2 are within the allowable range (step S110: N), the process of evaluating the quality of the learning data D ends.

[0090] When the process of generating a learning model and the process of evaluating the quality of the learning data D are completed, the information processing device 3 ends the process.

[0091] In the above example, a case has been described in which the receiving unit 311 has received a selection of unsupervised learning in step S102. Below, a difference between the case in which the receiving unit 311 has received a selection of supervised learning in step S102 will be described.

[0092] ==Neural Networks==(Supervised) 10, the state in which supervised learning is selected is shown by selecting the button 40b for selecting supervised learning out of the buttons 40a and 40b on the setting screen 40. Note that here, an example in which a neural network is selected as the learning model will be described.

[0093] Next, in step S103, the receiving unit 311 receives a selection of a learning model from the user. The receiving unit 311 receives the selection of a learning model from the user via the setting screen 40 shown in Fig. 10. The example in Fig. 10 shows a state in which a neural network is selected as the learning model by selecting the neural network from the drop-down list 40d.

[0094] In step S105, the reception unit 311 receives input of parameters P1 to P3 from the user via the setting screen 40 shown in Fig. 10. The user inputs the parameters P1 to P3 in the input windows 40e to 40g, respectively.

[0095] In step S108, the third calculation unit 315 calculates the index I3. Step S108 is a process that is executed only when the reception unit 311 receives a selection of supervised learning.

[0096] In step S109, the fourth calculation unit 316 calculates an index I4. At this time, the index I4 is a weighted average value of the indexes I1, I2, and I3.

[0097] In step S110, the display unit 318 displays the indices calculated in steps S106 to S109 as an output screen. Fig. 11 is an example of an output screen displayed on the output device. In this example, the output screen 42 displays the values ​​of the indices I1, I2, I3, and I4.

[0098] In this result, the values ​​of the indexes I1 and I2 are similar to those in the example of FIG. 9. Meanwhile, the value of the index I3 is 0.71. As described above, the preset allowable range for the index I3 is 0.8 or more. In other words, the value of the index I3 is not within the allowable range.

[0099] In this way, if any of the calculated values ​​of the indices I1, I2, or I3 does not fall within the allowable range (step S110: Y), the warning unit 317 issues a warning.

[0100] 11 shows a warning pop-up 42a. The pop-up 42a displays a warning that the index I3 is too small relative to the allowable range. Furthermore, the pop-up 42a displays a message recommending that the number of data included in the abnormal class be increased among the learning data D.

[0101] If any of the calculated indices I1, I2, or I3 falls within the allowable range (step S110: N), the process of evaluating the quality of the learning data D ends.

[0102] According to the above-described information processing method, when generating a learning model, it is possible to evaluate the quality of the learning data D for generating the learning model. By considering the evaluation result of the quality of the learning data D, a user can estimate the reliability of the generated learning model.

[0103] ==Summary== As described above, the information processing device 3 of the embodiment includes an acquisition unit 310 that acquires learning data D for generating a learning model, a first calculation unit 313 that calculates an index I1 indicating the amount of learning data D based on the learning data D and a parameter P1 according to the type of learning model, and a second calculation unit 314 that calculates an index I2 indicating the uniformity of distribution of the learning data D based on the learning data D.

[0104] According to this configuration, when generating a learning model, it is possible to evaluate the quality of the learning data D for generating the learning model. By considering the evaluation result of the quality of the learning data D, a user can estimate the reliability of the generated learning model.

[0105] In the information processing device 3, the second calculation unit 314 calculates the index I2 using a parameter P2 indicating the number of divisions for dividing the range of values ​​included in the learning data D. The index I2 may vary depending on the number of divisions. Therefore, with this configuration, by using the parameter P2 within a range in which the index I2 does not vary significantly, it is possible to obtain an index I2 with improved reliability.

[0106] The information processing device 3 further includes a third calculation unit 315 that calculates an index I3, where the learning data D includes first data classified into a first class and second data classified into a second class, and the third calculation unit 315 calculates an index I3 indicating how far a ratio based on the number of first data and the number of second data is from a reference ratio, based on the learning data D and a parameter P3 indicating the reference ratio. With this configuration, the accuracy of evaluation of the learning data D for generating a learning model based on supervised learning is improved.

[0107] The information processing device 3 further includes a fourth calculation unit 316 that calculates, based on the calculated multiple indexes, a fourth index for evaluating the learning data D. With this configuration, the quality of the learning data D can be evaluated using one index, making the evaluation easier.

[0108] In the information processing device 3, the fourth calculation unit 316 calculates an index I4 using weighting factors W1 to W3 for the indexes I1 to I3, respectively, and the weighting factor W1 is greater than the weighting factors W2 and W3. Since the quality of the training data is relatively greatly influenced by the completeness of the training data D, such a configuration improves the accuracy of the quality evaluation.

[0109] The information processing device 3 includes a display unit 318 that displays the calculated index. With this configuration, the evaluation result of the learning data D can be easily understood.

[0110] The information processing device 3 further includes a warning unit 317 that issues a warning when at least one of the calculated multiple indexes is not within a predetermined range. With this configuration, when the evaluation result of the learning data is not favorable, the fact and the cause can be easily understood.

[0111] In the information processing device 3, the learning data D is the output of a plurality of sensors 2 provided in the device to be diagnosed, the learning model is a model generated based on MSPC, and the first calculation unit 313 calculates an index I1 based on the learning data D, the number of the plurality of sensors 2, and a parameter P1. In MSPC, the required number of data included in the learning data D is proportional to the number of sensors 2. Therefore, according to such a configuration, the index I1 taking the number of sensors 2 into consideration is calculated, improving the accuracy of the evaluation.

[0112] The information processing method of the embodiment includes the steps of acquiring learning data D for generating a learning model, calculating an index I1 indicating the amount of learning data D based on the learning data D and a parameter P1 corresponding to the type of learning model, and calculating an index I2 indicating the uniformity of distribution of the learning data D based on the learning data D.

[0113] According to this method, when generating a learning model, it is possible to evaluate the quality of the learning data D used to generate the learning model. By taking into account the evaluation result of the quality of the learning data D, a user can estimate the reliability of the generated learning model.

[0114] The information processing program of the embodiment causes a computer to realize an acquisition unit 310 that acquires learning data D for generating a learning model, a first calculation unit 313 that calculates an index I1 indicating the amount of learning data D based on the learning data D and a parameter P1 corresponding to the type of learning model, and a second calculation unit 314 that calculates an index I2 indicating the uniformity of distribution of the learning data D based on the learning data D.

[0115] According to such a program, when generating a learning model, it becomes possible to evaluate the quality of the learning data D for generating the learning model. By taking into account the evaluation result of the quality of the learning data D, a user can estimate the reliability of the generated learning model.

[0116] The above-mentioned embodiment is for the purpose of facilitating understanding of the present invention, and is not intended to limit the present invention. Furthermore, the present invention can be modified or improved without departing from the spirit of the present invention, and it goes without saying that the present invention includes equivalents thereof.

[0117] For example, in the above embodiment, the information processing device 3 is shown as not having a means for performing an abnormality diagnosis using the generated learning model, but may further include a diagnosis unit for performing this, an output unit for outputting the diagnosis result, etc. [Explanation of symbols]

[0118] 1: Information processing system 2: Sensor 3: Information processing device 300: Processor 301: Main memory 302:Auxiliary storage device 303: Input device 304: Output device 305: Communication equipment 310: Acquisition Department 311: Reception 312: Generation part 313: 1st calculation section 314:Second calculation section 315:Third Calculation Department 316: 4th calculation section 317:Alarm section 318: Display section

Claims

1. An acquisition unit that acquires learning data for generating a learning model; a first calculation unit that calculates a first index indicating the amount of the learning data based on the learning data and a first parameter corresponding to the type of the learning model; a second calculation unit that calculates a second index indicating uniformity of distribution of the training data based on the training data; An information processing device comprising:

2. 2. The information processing device according to claim 1, The second calculation unit calculates the second index using a second parameter indicating a number of divisions into which a range of values ​​included in the learning data is divided. Information processing device.

3. 3. The information processing device according to claim 1, Further comprising a third calculation unit that calculates a third index, The learning data is The data includes first data classified into a first class and second data classified into a second class, The third calculation unit is calculating the third index indicating how far the ratio based on the number of the first data and the number of the second data is from a reference ratio, based on the learning data and a third parameter indicating the reference ratio; Information processing device.

4. 4. The information processing device according to claim 3, and a fourth calculation unit that calculates a fourth index for evaluating the learning data based on the calculated plurality of indexes. Information processing device.

5. 5. The information processing device according to claim 4, The fourth calculation unit is calculating the fourth index using first to third weighting factors for the first to third indexes, respectively; the first weighting factor is greater than the second and third weighting factors; Information processing device.

6. The information processing device according to any one of claims 1 to 5, a display for displaying the calculated metrics; Information processing device.

7. The information processing device according to any one of claims 1 to 6, and a warning unit that issues a warning when at least one of the calculated multiple indexes is not within a predetermined range. Information processing device.

8. The information processing device according to any one of claims 1 to 7, the learning data is output from a plurality of sensors provided in the device to be diagnosed, The learning model is a model generated based on MSPC, The first calculation unit is Calculating the first index based on the learning data, the number of the plurality of sensors, and the first parameter. Information processing device.

9. Obtaining training data for generating a training model; calculating a first index indicating the amount of the learning data based on the learning data and a first parameter corresponding to the type of the learning model; calculating a second index indicating uniformity of distribution of the training data based on the training data; An information processing method comprising:

10. On the computer, An acquisition unit that acquires learning data for generating a learning model; a first calculation unit that calculates a first index indicating the amount of the learning data based on the learning data and a first parameter corresponding to the type of the learning model; a second calculation unit that calculates a second index indicating uniformity of distribution of the training data based on the training data; An information processing program that realizes this.

Images