Prediction device, material design system, prediction method, and prediction program

Abstract

The development efficiency of new materials is improved. The prediction device includes: a section determination unit that acquires a learning data set for generating a learning completion model, and determines a plurality of sections for dividing attribute values on the basis of a frequency distribution of the attribute values calculated among a plurality of data included in the learning data set; an evaluation unit that evaluates the suitability of the data to be predicted with respect to an opposite index by determining which of the plurality of sections an attribute value calculated between the data to be predicted and the plurality of data is divided; and a display unit that displays a predicted value predicted by the learning completion model and an evaluation result of the evaluation unit in association with each other by inputting data of the prediction target.

Description

Technical Field

[0001] The present invention relates to a prediction device, a material design system, a prediction method and a prediction program. Background Art

[0002] In recent years, in the development of new materials, attempts have been made to use learning prediction models generated by machine learning to predict material properties. By using this learning prediction model, the process of material generation and material property verification experiments can be reduced, thereby improving the efficiency of new material development.

[0003] On the other hand, in the case of a learned prediction model generated by machine learning, it is known that the prediction accuracy will decrease when the prediction target data (input data) significantly deviates from the learning data used in the machine learning.

[0004] On the other hand, for example, the following Non-Patent Document 1 proposes defining the application range of a prediction model (the range of prediction target data that can achieve a desired prediction accuracy).

[0005] <Prior Art Literature>

[0006] <Non-patent literature>

[0007] Non-patent document 1: Igor I.Baskin, Natalia Kireeva, and Alexandre Varnek, "The One-Class Classification Approach to Data Description and to ModelsApplicability Domain", Nol.Inf.2010,29,581p-587p Summary of the invention

[0008] <Problems to be Solved by the Invention>

[0009] However, in the development of new materials, even if material properties are predicted for prediction target data that is close to learning data, the prediction target data lacks unpredictability, and the possibility of obtaining a useful new material is low.

[0010] On the other hand, if the prediction object data deviates significantly from the learning data, the prediction accuracy is inherently low. Therefore, even if the material properties that meet the specified conditions are predicted, the predicted material properties are rarely obtained during actual generation, and the possibility of obtaining beneficial new materials is also low.

[0011] For the above reasons, in the development of new materials, if we want to improve the development efficiency, we believe that the key is to define the application scope of the prediction model based on the two opposite indicators of unpredictability and prediction accuracy.

[0012] The present invention aims to improve the efficiency of developing new materials.

[0013] <Methods used to solve the problem>

[0014] A prediction device according to a first aspect of the present invention includes:

[0015] An interval determination unit that obtains a learning data set for generating a learning completion model and determines a plurality of intervals for dividing the attribute value based on a frequency distribution of the attribute value calculated between a plurality of data included in the learning data set;

[0016] an evaluation unit that evaluates the suitability of the data of the prediction target with respect to an opposite indicator by determining into which of the plurality of intervals a property value calculated between the data of the prediction target and the plurality of data is divided; and

[0017] The display unit displays the predicted value predicted by the learned model by inputting the data of the prediction target in association with the evaluation result of the evaluation unit.

[0018] A second aspect of the present invention is a prediction device according to the first aspect, wherein:

[0019] The interval determination unit calculates summary statistics for attribute values ​​calculated between the plurality of data and determines lower limit values ​​or upper limit values ​​for defining the attribute values ​​of the plurality of intervals.

[0020] A third aspect of the present invention is a prediction device according to the second aspect, wherein:

[0021] The interval determination unit determines three or more intervals that do not overlap each other.

[0022] A fourth aspect of the present invention is a prediction device according to the third aspect, wherein:

[0023] The evaluation unit evaluates the suitability of the data to be predicted based on which interval, among the three or more intervals, the attribute value calculated between the data to be predicted and the plurality of data is divided into, which interval is close to an interval including a predetermined summary statistic.

[0024] A fifth aspect of the present invention is a prediction device according to the third aspect, wherein:

[0025] The evaluation unit excludes the data of the prediction object from the data to be input into the learning completion model when it is determined that the attribute value calculated between the data of the prediction object and the multiple data is divided into an interval farthest from the interval containing the specified summary statistic among the three or more intervals.

[0026] A sixth aspect of the present invention is the prediction device according to the third aspect, wherein:

[0027] The evaluation unit selects the data of the prediction object as data input to the learning completion model when it is determined that the attribute value calculated between the data of the prediction object and the multiple data is divided into the Rth interval among the three or more intervals that is close to the interval containing the specified summary statistic.

[0028] A seventh aspect of the present invention is the prediction device according to the first aspect, further comprising:

[0029] a learning data attribute value calculation unit that calculates distances between a plurality of data included in the learning data set; and

[0030] The learning data minimum attribute value extraction unit extracts the minimum distance from the distances with other data calculated for each of the plurality of data.

[0031] The section determination unit determines a plurality of sections for dividing the minimum distance based on the extracted frequency distribution of the minimum distance.

[0032] An eighth aspect of the present invention is a prediction device according to the seventh aspect, wherein:

[0033] The learning data attribute value calculation unit calculates the distance between the i-th (1≤i≤N) data among the N data (N is an arbitrary integer) included in the learning data set and the (N-1) data other than the i-th data.

[0034] The learning data minimum attribute value extraction unit extracts the minimum distance from among the (N-1) distances calculated for the i-th data.

[0035] A ninth aspect of the present invention is the prediction device according to the seventh aspect, further comprising:

[0036] A prediction target data attribute value calculation unit calculates the distance between the prediction target data and the plurality of data; and

[0037] The prediction target data minimum attribute value extraction unit extracts the minimum distance among the distances between the prediction target data and the plurality of data.

[0038] The evaluation unit evaluates the suitability of the data of the prediction target with respect to an opposite index by determining into which of the plurality of intervals the minimum distance extracted for the data of the prediction target is divided.

[0039] A material design system according to a tenth aspect of the present invention includes:

[0040] The prediction device according to the first aspect; and

[0041] A material design device generates material design data by inputting data of a prediction object whose attribute value calculated between the plurality of data is judged by the evaluation unit of the prediction device to be divided into a specified interval and whose prediction value satisfies the specified conditions is predicted by the learning completion model of the prediction device.

[0042] The eleventh aspect of the present invention is a material design system according to the tenth aspect, wherein:

[0043] It also has a learning device that generates a learning completion model based on a learning data set.

[0044] The prediction device predicts the prediction value by inputting the data of the prediction object into the learned model generated by the learning device.

[0045] The prediction method according to the twelfth aspect of the present invention is for executing the following steps by a computer:

[0046] An interval determination step of obtaining a learning data set for generating a learning completion model and determining a plurality of intervals for dividing the attribute value based on a frequency distribution of the attribute value calculated between a plurality of data included in the learning data set;

[0047] an evaluation step of evaluating the suitability of the data of the prediction target with respect to an opposite indicator by determining into which of the plurality of intervals the attribute value calculated between the data of the prediction target and the plurality of data is divided; and

[0048] The display step displays the predicted value predicted by the learned model by inputting the data of the prediction object in association with the evaluation result in the evaluation step.

[0049] The prediction program according to the thirteenth aspect of the present invention is configured to execute the following steps by a computer:

[0050] An interval determination step of obtaining a learning data set for generating a learning completion model and determining a plurality of intervals for dividing the attribute value based on a frequency distribution of the attribute value calculated between a plurality of data included in the learning data set;

[0051] an evaluation step of evaluating the suitability of the data of the prediction target with respect to an opposite indicator by determining into which of the plurality of intervals the attribute value calculated between the data of the prediction target and the plurality of data is divided; and

[0052] The display step displays the predicted value predicted by the learned model by inputting the data of the prediction object in association with the evaluation result in the evaluation step.

[0053] <Effects of the Invention>

[0054] According to the present invention, the efficiency of developing new materials can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a diagram showing an example of the system configuration of the material design system.

[0056] Figure 2 This is a diagram showing an example of the functional configuration of a learning device and a prediction device.

[0057] Figure 3 This is a diagram showing an example of the hardware configuration of a learning device and a prediction device.

[0058] Figure 4 This is a diagram showing a specific example of processing performed by a learning unit of a learning device.

[0059] Figure 5 This is a diagram showing a specific example of processing performed by the learning data attribute value calculation unit of the prediction device.

[0060] Figure 6 This is a diagram showing a specific example of processing performed by the interval determination unit of the prediction device.

[0061] Figure 7 This is a diagram showing a specific example of processing performed by a prediction target data attribute value calculation unit of a prediction device.

[0062] Figure 8 This is a diagram showing a specific example of processing performed by the evaluation unit of the prediction device.

[0063] Fig. 9 This is a diagram showing a specific example of processing performed by the prediction unit of the prediction device.

[0064] Fig.10 This is a diagram showing a specific example of processing performed by the display unit of the prediction device.

[0065] Fig.11 It is a flowchart which shows the flow of a learning process.

[0066] Fig.12 It is a flowchart which shows the flow of prediction processing.

[0067] Fig.13 This is a diagram showing Example 1.

[0068] Fig.14 This is a diagram showing Example 2.

[0069] Fig.15 This is a diagram showing Example 3.

[0070] Fig.16 It is a diagram showing a verification example of the predicted characteristics. DETAILED DESCRIPTION

[0071] Hereinafter, each embodiment will be described with reference to the drawings. In addition, in this specification and the drawings, the same reference numerals are given to components having substantially the same functional configuration, and repeated descriptions are omitted.

[0072] [First embodiment]

[0073] <System structure of the Material Design system>

[0074] First, the system configuration of the material design system according to the first embodiment will be described. Figure 1 is a diagram showing an example of the system structure of the material design system. Figure 1 As shown, the material design system 100 includes a material property experimental device 110 , a learning device 120 , a prediction device 130 , a material design device 140 , a material generation device 150 , and a material property experimental device 160 .

[0075] The material property experiment device 110 is a device for obtaining the characteristic value of "material property" related to the material with a known structural formula of "material data" through experiments. The characteristic value of each material obtained by the experimenter 111 using the material property experiment device 110 is input into the learning device 120 together with the corresponding structural formula as learning data. In addition, the learning data can also be composed based on the information of a known database.

[0076] The learning device 120 performs machine learning on a prediction model (a model for predicting a characteristic value of a material based on a structural formula of the material) using a learning data set including the acquired plurality of learning data, and generates a learned prediction model. In addition, the learning device 120 notifies the prediction device 130 of the learning data set used when generating the learned prediction model and the model parameters of the learned prediction model.

[0077] The prediction device 130 has a learned prediction model in which the model parameters notified from the learning device 120 are set, and predicts the prediction characteristics (an example of a prediction value) of the prediction target data input by the designer 131 .

[0078] In addition, the prediction device 130 analyzes the learning data set notified by the learning device 120 and quantitatively calculates the degree to which the material data of each learning data deviates from the material data of other learning data, thereby calculating the attribute values ​​between the material data.

[0079] Furthermore, the prediction device 130 calculates the frequency distribution of the attribute values ​​between the calculated material data, and thereby determines a plurality of intervals for dividing the attribute values ​​(specifically, defines the lower limit value and the upper limit value of the attribute value in each interval).

[0080] In addition, the prediction device 130 calculates the property value of the material data of the prediction target data by quantitatively calculating the degree to which the material data of the prediction target data deviates from the material data of each learning data. Furthermore, the prediction device 130 evaluates the appropriateness of the prediction target data by determining which of the multiple determined intervals the property value of the material data of the prediction target data is divided into, and outputs the evaluation result in association with the prediction characteristic.

[0081] In addition, the appropriateness of the prediction object data mentioned here refers to the appropriateness of the prediction object data in the development of new materials relative to the two opposite indicators of unpredictability and prediction accuracy. A high appropriateness of the prediction object data indicates a state in which the unpredictability of the prediction object data relative to the learning data and the prediction accuracy related to the prediction characteristics of the prediction object data are balanced (in this embodiment, the appropriateness is expressed as a "level" (see below for details)).

[0082] The prediction characteristics of the prediction target data and the evaluation results of the prediction target data output by the prediction device 130 are notified to the designer 131. Thus, the designer 131 can understand the prediction characteristics of the prediction target data and whether the material generated based on the prediction target data is likely to be a new useful material.

[0083] The material design device 140 is a device for generating material design data. The designer 131 generates material design data by selecting prediction target data that is highly likely to become a useful new material and inputting the data into the material design device 140 .

[0084] The material generating device 150 is a device for actually generating the material based on the generated material design data.

[0085] The material property experiment device 160 is a device for obtaining, through verification experiments, material properties of a new material actually generated by the material generation device 150. The material properties of the new material obtained by the experimenter 161 using the material property experiment device 160 are notified to the designer 131.

[0086] <Functional Structure of Learning Device and Prediction Device>

[0087] The functional configurations of the learning device 120 and the prediction device 130 of the various devices constituting the material design system 100 are described below. Figure 2 Provide explanation. Figure 2 This is a diagram showing an example of the functional configuration of a learning device and a prediction device.

[0088] A learning program is installed in the learning device 120 , and by executing the program, the learning device 120 functions as a learning data acquisition unit 201 and a learning unit 202 .

[0089] The learning data acquisition unit 201 acquires the learning data input by the experimenter 111 and stores the data in the learning data storage unit 205 .

[0090] The learning unit 202 performs machine learning on the prediction model and generates a learned prediction model using a learning data set including a plurality of learning data stored in the learning data storage unit 205. The model parameters of the learned prediction model generated by the learning unit 202 are notified to the prediction device 130 together with the learning data set used when generating the learned prediction model.

[0091] The prediction device 130 is installed with a prediction program. The prediction device 130 executes the program to function as a learning data attribute value calculation unit 211, an interval determination unit 212, a prediction target data acquisition unit 221, a prediction unit 222, a prediction target data attribute value calculation unit 223, an evaluation unit 224, and a display unit 225.

[0092] The learning data attribute value calculation unit 211 calculates attribute values ​​between material data by analyzing the learning data set notified by the learning device 120 and quantitatively calculating how much the material data of each learning data deviates from the material data of other learning data.

[0093] The interval determination unit 212 determines a plurality of intervals for dividing the attribute values ​​(specifically, defines the lower limit value and the upper limit value of the attribute value in each interval) by calculating the frequency distribution of the attribute values ​​between the material data calculated by the learning data attribute value calculation unit 211 .

[0094] The prediction target data acquisition unit 221 acquires the prediction target data input by the designer 131 , and notifies the prediction unit 222 and the prediction target data attribute value calculation unit 223 .

[0095] The prediction unit 222 has a learned prediction model in which the model parameters notified from the learning unit 202 are set, and predicts the prediction characteristics of the prediction target data input by the designer 131 .

[0096] The prediction target data attribute value calculation unit 223 acquires each learning data of the learning data set notified by the learning device 120 , and acquires the prediction target data notified by the prediction target data acquisition unit 221 .

[0097] Furthermore, the prediction target data attribute value calculation unit 223 calculates the attribute value of the material data of the prediction target data by quantitatively calculating how much the material data of the prediction target data deviates from the material data of each learning data.

[0098] The evaluation unit 224 determines into which of the multiple intervals determined by the interval determination unit 212 the attribute value of the material data of the prediction target data calculated by the prediction target data attribute value calculation unit 223 is divided. In this way, the evaluation unit 224 evaluates the appropriateness of the prediction target data and notifies the display unit 225 of the evaluation result.

[0099] In addition, the evaluation unit 224 pre-sets an exclusion object, and when the calculated attribute value is divided into a specified interval, the prediction object data will be excluded from the prediction object of the prediction unit 222 (that is, it will not be input into the learning completion prediction model by the prediction unit 222). Alternatively, the evaluation unit 224 excludes the prediction object data from the display object of the display unit 225 (that is, it will not be displayed by the display unit 225) when the calculated attribute value is divided into a specified interval.

[0100] The display unit 225 displays the prediction characteristic of the prediction target data predicted by the prediction unit 222 and the evaluation result of the prediction target data evaluated by the evaluation unit 224 in association with each other.

[0101] As described above, the material design system 100 according to the first embodiment,

[0102] The prediction device 130 is provided to predict the prediction characteristics of the prediction target data. Thus, according to the first embodiment, the prediction target data that does not predict the prediction characteristics that satisfy the predetermined conditions can be excluded from the objects of material production and verification experiments.

[0103] · The prediction device 130 is provided to output the evaluation result of the prediction target data. Therefore, according to the first embodiment, when the prediction device 130 predicts the prediction characteristics that meet the prescribed conditions, the prediction target data with a low possibility of becoming a useful new material can be excluded from the objects of material generation and verification experiments.

[0104] As a result, according to the material design system 100 of the first embodiment, it is possible to narrow down the prediction target data for material generation and verification experiments, thereby improving the efficiency of new material development.

[0105] <Hardware Configuration of Learning Device and Prediction Device>

[0106] The following describes the hardware configuration of the learning device 120 and the prediction device 130. Since the learning device 120 and the prediction device 130 have the same hardware configuration, reference is made to the hardware configuration of the learning device 120 and the prediction device 130. Figure 3 The hardware structures of the learning device 120 and the prediction device 130 will be described together.

[0107] Figure 3 FIG. 1 is a diagram showing an example of the hardware configuration of a learning device and a prediction device. Figure 3 As shown, the learning device 120 and the prediction device 130 include a processor 301, a memory 302, an auxiliary storage device 303, an I / F (interface) device 304, a communication device 305, and a drive device 306. In addition, the hardware of the learning device 120 and the prediction device 130 are connected to each other via a bus 307.

[0108] The processor 301 includes various computing devices such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), etc. The processor 301 reads various programs (for example, a learning program, a prediction program, etc.) to the memory 302 and executes them.

[0109] The memory 302 includes a main storage device such as a ROM (Read Only Memory) and a RAM (Random Access Memory). The processor 301 and the memory 302 form a so-called computer, and the processor 301 executes various programs read out to the memory 302 , thereby realizing the various functions described above.

[0110] The auxiliary storage device 303 stores various programs and various data used when the processor 301 executes the various programs. For example, the learning data storage unit 205 is implemented in the auxiliary storage device 303.

[0111] The I / F device 304 is a connection device connected to an operation device 311 as an example of a user interface device and a display device 312. The communication device 305 is a communication device for communicating with an external device (not shown) via a network.

[0112] The drive device 306 is a device for setting the recording medium 313. The recording medium 313 mentioned here includes a medium that records information optically, electrically or magnetically, such as a CD-ROM, a floppy disk, a magneto-optical disk, etc. In addition, the recording medium 313 may also include a semiconductor memory that records information electrically, such as a ROM, a flash memory, etc.

[0113] In addition, various programs installed in the auxiliary storage device 303 are installed by, for example, setting the distributed recording medium 313 in the drive device 306 and reading various programs recorded in the recording medium 313 through the drive device 306. Alternatively, various programs installed in the auxiliary storage device 303 can also be downloaded from the network through the communication device 305 and installed.

[0114] <Specific Example of Processing by Each Component of the Learning Device>

[0115] Next, a specific example of processing performed by each unit (here, the learning unit 202 ) of the learning device 120 will be described.

[0116] (1) Specific example of processing performed by the learning unit 202

[0117] First, a specific example of the processing performed by the learning unit 202 of the learning device 120 will be described. Figure 4 1 is a diagram showing a specific example of processing performed by the learning unit of the learning device. Figure 4 As shown, the learning unit 202 includes an input unit 401 , a prediction model 402 , and a comparison / change unit 403 .

[0118] in addition, Figure 4 The learning data set 400 in the example includes "data number", "material data" and "material properties" as information items. The "data number" stores a number for identifying each learning data, the "material data" stores a structural formula of each material, and the "material properties" stores a characteristic value of each material. Figure 4 The example shows a case where N learning data (N is an arbitrary integer) are stored.

[0119] The input unit 401 reads the structural formula of the material included in the “material data” of each learning data from the learning data set 400 , and inputs it into the prediction model.

[0120] When the structural formula of a material included in the “material data” of each learning data is input, the prediction model 402 outputs output data corresponding to the material characteristics.

[0121] The comparison / changing unit 403 updates the model parameters of the prediction model 402 so that the output data output by the prediction model 402 approaches the characteristic values ​​of the material included in the “material characteristics” of the learning data.

[0122] Thus, in the learning unit 202 , a learned prediction model capable of predicting the material characteristics of the prediction target data can be generated based on the material data of the prediction target data.

[0123] <Specific Example of Processing by Each Component of the Prediction Device>

[0124] The following describes a specific example of processing performed by each part of the prediction device 130 (here, the learning data attribute value calculation unit 211, the interval determination unit 212, the prediction target data attribute value calculation unit 223, the evaluation unit 224, the prediction unit 222, and the display unit 225).

[0125] (1) Specific example of processing performed by the learning data attribute value calculation unit 211

[0126] First, a specific example of the processing performed by the learning data attribute value calculation unit 211 will be described. Figure 5 1 is a diagram showing a specific example of processing performed by the learning data attribute value calculation unit of the prediction device. Figure 5 As shown, the learning data attribute value calculation unit 211 includes a mutual attribute value calculation unit 501 and a learning data minimum attribute value extraction unit 502 .

[0127] The mutual attribute value calculation unit 501 reads out a number for identifying each piece of learning data and a structural formula of each material from the “data number” and “material data” of the learning data set 400 including N pieces of learning data.

[0128] In addition, the mutual attribute value calculation unit 501 generates the structural formula of the material included in the "material data" of the N learning data read out,

[0129] ·The structural formula of the material contained in the "material data" of the i-th (1≤i≤N) learning data and

[0130] · Combinations of structural formulas of materials contained in the “material data” of (N-1) other learning data except the i-th learning data, and for all combinations, calculating the attribute values ​​between the material data. The attribute value between the material data mentioned here refers to a value indicating the extent to which the structural formula of the material contained in the “material data” of the i-th learning data deviates from the structural formula of the material contained in the “material data” of other learning data except the i-th learning data. Specifically, it refers to the distance between the structural formula of the material contained in the “material data” of the i-th learning data and the structural formula of the material contained in the “material data” of other learning data except the i-th learning data. Alternatively, it refers to the density between the structural formula of the material contained in the “material data” of the i-th learning data and the structural formula of the material contained in the “material data” of other learning data except the i-th learning data, etc.

[0131] exist Figure 5In the table 511, numbers for identifying each learning data are respectively indicated on the horizontal axis and the vertical axis. The table 511 shows all combinations of material data of each learning data included in the learning data set 400.

[0132] In the table 511 , the area where the numbers written on the horizontal axis intersect the numbers written on the vertical axis indicates the attribute values ​​between the material data calculated for the corresponding combination of the material data of the learning data.

[0133] For example, in the area where the data number on the horizontal axis = "2" and the data number on the vertical axis = "1" intersect, the data for

[0134] · Material data of the learning data of data number = "2" = "Structural formula 2",

[0135] The property value (=0.1) between the material data calculated for the combination of material data = "Structural Formula 1" of the learning data of data number = "1".

[0136] The learning data minimum attribute value extraction unit 502 extracts the minimum attribute value (for example, the minimum distance) for each material from the attribute values ​​between the material data for all combinations calculated by the mutual attribute value calculation unit 501 .

[0137] exist Figure 5 In FIG. 5 , the table 512 is a table showing the minimum attribute value extracted for each material by the learning data minimum attribute value extraction unit 502 in association with a number for identifying each learning data.

[0138] For example, in the case of learning data identified by data number = "1", the minimum attribute value of the material is "0.1".

[0139] Furthermore, the minimum attribute value of each material extracted by the learning data minimum attribute value extraction unit 502 is notified to the interval determination unit 212 .

[0140] (2) Specific example of processing performed by the interval determination unit 212

[0141] A specific example of the processing performed by the interval determination unit 212 will be described below. Figure 6 1 is a diagram showing a specific example of processing performed by the interval determination unit of the prediction device. Figure 6 As shown, the interval determination unit 212 includes a frequency distribution generation unit 601 and an interval calculation unit 602 .

[0142] The frequency distribution generating unit 601 obtains the minimum attribute value of each material notified by the learning data minimum attribute value extracting unit 502, and generates a frequency distribution. Figure 6In the figure, reference numeral 611 is an example of a frequency distribution generated by the frequency distribution generating unit 601, the horizontal axis represents the attribute value, and the vertical axis represents the number of data.

[0143] The interval calculation unit 602 calculates summary statistics based on the minimum attribute value of each material obtained by the frequency distribution generation unit 601. Specifically, the interval calculation unit 602 calculates

[0144] The median value of the minimum property value of each material,

[0145] ·The first quartile of the minimum property value of each material (Q 25 ),

[0146] The third quartile of the minimum property value of each material (Q 75 ),

[0147] The interquartile range (IQR = Q 75 -Q 25 ).

[0148] Then, the interval calculation unit 602 determines coefficients (α1, α2) required for determining a plurality of intervals for dividing each attribute value based on the calculated summary statistic.

[0149] exist Figure 6 In the example, the coefficient α1 is determined to be "1.5", the coefficient α2 is determined to be "3.0", and the frequency distribution (symbol 611) generated by the frequency distribution generating unit 601 is determined.

[0150] · Become "attribute value < (third quartile (Q 75 ))” is the upper limit of the interval into which the attribute value is divided.

[0151] · Become "(the third quartile (Q 75 ))≤Attribute value<(Add the third quartile (Q 75 The lower and upper limits of the interval into which the attribute value of ")" is divided.

[0152] · It becomes “(the third quartile (Q 75 ) value) ≤ attribute value < (the value of α2 times the interquartile range (IQR) plus the third quartile (Q 75 The lower and upper limits of the interval into which the attribute value of ")" is divided.

[0153] · It becomes “(the third quartile (Q 75 )≤attribute value” is the lower limit value of the interval into which the attribute value is divided.

[0154] In addition, Figure 6 In the example, the following is specified:

[0155] · The interval containing the specified summary statistic (here, the “median value”) is defined as “level 1”,

[0156] · The first interval that is close to the interval containing the specified summary statistic is designated as "level 2",

[0157] · The second interval that is close to the interval containing the specified summary statistic is designated as "level 3",

[0158] The interval farthest from the interval containing the specified summary statistic is designated as "level 4".

[0159] Here, the relationship between the above grades and two opposing indicators in new material development (unpredictability and prediction accuracy) can be summarized as follows:

[0160] The prediction accuracy decreases in the order of "level 1" → "level 2" → "level 3" → "level 4". That is, the level indicates the degree of improvement of the prediction accuracy.

[0161] Unpredictability increases in the order of "level 1" → "level 2" → "level 3" → "level 4". That is, the level indicates the degree of decrease in unpredictability.

[0162] As described above, it can be said that each level simultaneously expresses two contradictory indicators (unpredictability and prediction accuracy) in the development of new materials.

[0163] (3) Specific example of processing performed by the prediction target data attribute value calculation unit 223

[0164] A specific example of the processing performed by the prediction target data attribute value calculation unit 223 will be described below. Figure 7 1 is a diagram showing a specific example of processing performed by the prediction object data attribute value calculation unit of the prediction device. Figure 7 As shown, the prediction target data attribute value calculation unit 223 includes a mutual attribute value calculation unit 701 and a prediction target data minimum attribute value extraction unit 702 .

[0165] The mutual attribute value calculation unit 701 reads out the number for identifying each learning data and the structural formula of each material from the “data number” and “material data” of the learning data set 400 including the N learning data notified by the learning device 120. In addition, the mutual attribute value calculation unit 701 reads out the number for identifying the prediction target data and the structural formula of the material from the “data number” and “material data” of the prediction target data 700 notified by the prediction target data acquisition unit 221.

[0166] In addition, the mutual attribute value calculation unit 701 generates the structural formula of the material included in the "material data" of the read N learning data.

[0167] ·The structural formula of the material contained in the "material data" of the i-th (1≤i≤N) learning data and

[0168] ·Predict the combination of the structural formulas of the materials contained in the "material data" of the prediction object data 700, and calculate the attribute values ​​between the material data for all the combinations. The attribute value between the material data mentioned here refers to a value indicating the degree to which the structural formula of the material contained in the "material data" of the i-th learning data and the structural formula of the material contained in the "material data" of the prediction object data 700 deviate from each other. Specifically, it refers to the distance between the structural formula of the material contained in the "material data" of the i-th learning data and the structural formula of the material contained in the "material data" of the prediction object data 700. Alternatively, it refers to the density between the structural formula of the material contained in the "material data" of the i-th learning data and the structural formula of the material contained in the "material data" of the prediction object data 700, etc.

[0169] exist Figure 7 In the table 711, the horizontal axis is marked with numbers for identifying each learning data included in the learning data set 400, and the vertical axis is marked with numbers for identifying the prediction target data. Table 711 shows all combinations of material data of each learning data included in the learning data set 400 and material data of the prediction target data.

[0170] In Table 711, the area where the numbers marked on the horizontal axis intersect the numbers marked on the vertical axis represents the attribute values ​​between the material data calculated for the combination of the corresponding material data of the learning data and the corresponding material data of the prediction target data.

[0171] For example, in the area where the data number on the horizontal axis = "2" and the data number on the vertical axis = "X" intersect, the data corresponding to

[0172] · Material data of the learning data of data number = "2" = "Structural formula 2",

[0173] The property value (=0.1) between the material data calculated for the combination of material data = "structural formula X" of the prediction target data of data number = "X".

[0174] The prediction target data minimum attribute value extraction unit 702 extracts the minimum attribute value from the attribute values ​​between all the combinations of material data calculated by the mutual attribute value calculation unit 701 .

[0175] exist Figure 7 In the table 712 , the minimum attribute value extracted by the prediction target data minimum attribute value extraction unit 702 is shown in association with the number for identifying the prediction target data. Figure 7 The example shows the case where the minimum attribute value is "0.1".

[0176] In addition, the minimum attribute value extracted by the prediction target data minimum attribute value extraction unit 702 is notified to the evaluation unit 224 .

[0177] (4) Specific example of processing performed by the evaluation unit 224

[0178] A specific example of the processing performed by the evaluation unit 224 will be described below. Figure 8 1 is a diagram showing a specific example of processing performed by the evaluation unit of the prediction device. Figure 8 As shown, the evaluation unit 224 includes an attribute value acquisition unit 801 and an evaluation result output unit 802 .

[0179] The attribute value acquisition unit 801 acquires the minimum attribute value notified by the prediction target data minimum attribute value extraction unit 702 , and determines to which of the plurality of intervals notified by the interval determination unit 212 the minimum attribute value is classified. Figure 8 The example shows a case where the minimum attribute value calculated for the structural formula (structural formula X) of the material contained in the "material data" of the prediction object data 700 is determined to be divided into the Rth interval (R=2) close to the interval containing the specified summary statistic (central value).

[0180] The evaluation result output unit 802 makes an evaluation on the suitability of the structural formula (structural formula X) of the material included in the "material data" of the prediction target data 700 in the new material development, based on the result that the interval determined by the attribute value acquisition unit 801 is the Rth interval (R=2).

[0181] The level of unpredictability is Rth (R=2),

[0182] The prediction accuracy is the Rth (R=2)

[0183] That is, the evaluation result output unit 802 evaluates the suitability (level) of the prediction target data according to the interval into which the attribute value of the material data of the prediction target data is classified, which is close to the interval containing the predetermined summary statistic (median value).

[0184] In addition, the evaluation result output unit 802 notifies the display unit 225 of the evaluation result (grade).

[0185] In addition, the evaluation result output unit 802 determines whether the minimum property value calculated for the structural formula (structural formula X) of the material included in the "material data" of the prediction target data 700 is classified into a predetermined interval. Specifically, it determines whether the evaluation result (level) made on the appropriateness of the structural formula (structural formula X) of the material included in the "material data" of the prediction target data 700 meets the exclusion target. When the evaluation result output unit 802 determines that it meets the exclusion target, it notifies the prediction unit 222 and / or the display unit 225.

[0186] (5) Specific example of processing performed by the prediction unit 222

[0187] A specific example of the processing performed by the prediction unit 222 will be described below. Fig. 9 1 is a diagram showing a specific example of processing performed by the prediction unit of the prediction device. Fig. 9 As shown, the prediction unit 222 includes a prediction target data input unit 901 , a learned prediction model 902 , and a prediction characteristic output unit 903 .

[0188] When the prediction target data acquisition unit 221 notifies the prediction target data 700 , the prediction target data input unit 901 inputs the structural formula (structural formula X) of the material included in the “material data” of the prediction target data 700 into the learned prediction model 902 .

[0189] In addition, when the evaluation result output unit 802 notifies the judgment result that the prediction target data input unit 901 meets the exclusion target, it can also be controlled not to input the structural formula (structural formula X) of the material included in the "material data" of the prediction target data 700.

[0190] The learned prediction model 902 is a prediction model after learning is completed, in which the model parameters calculated by the learning process performed by the learning unit 202 are set. When the structural formula (structural formula X) of the material included in the "material data" of the prediction target data 700 is input, the learned prediction model 902 predicts the prediction characteristics of the prediction target data 700.

[0191] The prediction characteristic output unit 903 notifies the display unit 225 of the prediction characteristic of the prediction target data 700 predicted by the learned prediction model 902 .

[0192] (6) Specific example of processing performed by the display unit 225

[0193] A specific example of the processing performed by the display unit 225 will be described below. Fig.10 FIG. 2 is a diagram showing a specific example of processing performed by the display unit of the prediction device. Fig.10As shown, the display unit 225 includes a display information acquisition unit 1001. The display information acquisition unit 1001 acquires the evaluation result notified by the evaluation unit 224 and the prediction characteristic notified by the prediction unit 222, and generates display data 1011.

[0194] like Fig.10 As shown, the display data 1011 includes “material data”, “predicted characteristics”, and “evaluation results” as information items. “Material data” stores “structural formula X”, which is the structural formula of the material included in the “material data” of the prediction target data 700, and “predicted characteristics” stores “characteristic value 2”, which is the predicted characteristics predicted for the prediction target data 700. Furthermore, “evaluation results” stores “level 2”, which is the evaluation result of evaluating the suitability of the structural formula (structural formula X) of the material included in the “material data” of the prediction target data 700 in the development of new materials.

[0195] <Flow of Learning Processing by Learning Device>

[0196] Next, the flow of the learning process performed by the learning device 120 will be described. Fig.11 It is a flowchart which shows the flow of a learning process.

[0197] In step S1101 , the learning device 120 acquires a learning data set including material properties of each material obtained by an experimenter 111 using the material property experiment device 110 .

[0198] In step S1102 , the learning device 120 performs a learning process on the prediction model using the learning data set, and generates a learned prediction model.

[0199] <Flow of prediction processing by prediction device>

[0200] The following describes the flow of prediction processing performed by the prediction device 130. Fig.12 It is a flowchart which shows the flow of prediction processing.

[0201] In step S1201 , the prediction device 130 acquires the learning data set used when the learning device 120 generates the learned prediction model and the model parameters of the learned prediction model.

[0202] In step S1202 , the prediction device 130 analyzes the learning data set and calculates the attribute values ​​between the material data.

[0203] In step S1203 , the prediction device 130 calculates the frequency distribution of the attribute values ​​between the calculated material data, and determines a plurality of intervals for dividing the attribute values.

[0204] In step S1204 , the prediction device 130 refers to the setting of the exclusion object.

[0205] In step S1205 , the prediction device 130 acquires prediction target data.

[0206] In step S1206 , the prediction device 130 calculates the attribute value of the material data of the prediction target data.

[0207] In step S1207 , the prediction device 130 determines into which of the determined plurality of intervals the calculated property value of the material data of the prediction target data is classified, thereby evaluating the suitability (level) of the prediction target data with respect to two opposite indices.

[0208] In step S1208, the prediction device 130 determines whether the evaluation result meets the exclusion target. If it is determined in step S1208 that the evaluation result does not meet the exclusion target (if "No" in step S1208), the process proceeds to step S1211.

[0209] On the contrary, when it is determined in step S1208 that the object is eligible for exclusion (if "YES" in step S1208), the process proceeds to step S1210.

[0210] In step S1209 , the prediction device 130 excludes the prediction target data from the prediction targets or display targets.

[0211] In step S1210 , the prediction device 130 predicts the prediction characteristics by inputting the prediction target data into the learned prediction model.

[0212] In step S1211 , the prediction device 130 displays display data in which the prediction target data, the prediction characteristics, and the evaluation results are associated with each other.

[0213] <Relationship between indicators and evaluation results in each embodiment>

[0214] Next, the results (prediction characteristics, evaluation results) of prediction processing on a plurality of prediction target data sets using the prediction device 130 are shown below, and the relationship between two opposite indices and the evaluation results and a verification example of the prediction characteristics are described.

[0215] Specifically, 100 molecules in the water solubility data set disclosed in the non-patent literature described later were used as the learning data set for learning, and the remaining unlearned molecules were used as the prediction target data to predict the water solubility and obtain the evaluation results. The results of the simultaneous processing of the molecules in the learning data set by changing them three times are described in the following Examples 1 to 3.

[0216] As a processing order, RDKIT was used to calculate descriptors based on SMILES (Simplified Molecular Line Input System) representing molecular structures, and the calculated descriptors were standardized and used as explanatory variables. Ridge regression was used as a prediction model. Standardization and ridge regression were performed using scikit-learn.

[0217] (1) Embodiment 1

[0218] Fig.13 is a diagram showing Example 1. Fig.13 As shown, in the first embodiment, by using multiple prediction target data,

[0219] Nitromethane,

[0220] Methanol,

[0221] ·sucrose,

[0222] Digoxin

[0223] The predicted characteristics are predicted, and the intervals into which the attribute values ​​are divided are determined, and the appropriateness is evaluated (see Table 1310). In addition, the material properties of the above-mentioned prediction object data (here, water solubility) are all known items, so they are shown in Table 1320 for the purpose of verifying the prediction accuracy.

[0224] According to Table 1310 , the relationship between the index of each prediction target data and the evaluation result is as follows.

[0225] Regarding the prediction target data evaluated as Level 1 and Level 2, a prediction property with high water solubility was obtained and good prediction accuracy was obtained, but the structural formula was simple and the unpredictability was low.

[0226] Regarding the prediction target data evaluated as level 3, a prediction characteristic with high water solubility was obtained, and good prediction accuracy was obtained, and the structural formula was relatively complex and unpredictability was high.

[0227] · Regarding the prediction target data evaluated as level 4, the prediction characteristics of extremely high water solubility were obtained, but in fact, the water solubility of the material was low, and the prediction accuracy was significantly low. In addition, the structural formula was complex and the unpredictability was high.

[0228] (2) Embodiment 2

[0229] Fig.14 is a diagram showing Example 2. Fig.14 As shown, in Embodiment 2, by using multiple prediction target data,

[0230] Erythritol,

[0231] Methanol,

[0232] ·lactose,

[0233] ·Lafite candy

[0234] The predicted characteristics are predicted, and the intervals into which the attribute values ​​are divided are determined, and the appropriateness is evaluated (see Table 1410). In addition, the material properties of the above-mentioned prediction object data (here, water solubility) are all known items, so they are shown in Table 1420 for the purpose of verifying the prediction accuracy.

[0235] According to Table 1410 , the relationship between the index of each prediction target data and the evaluation result is as follows.

[0236] Regarding the prediction target data evaluated as Level 1 and Level 2, a prediction property with high water solubility was obtained and good prediction accuracy was obtained, but the structural formula was simple and the unpredictability was low.

[0237] Regarding the prediction target data evaluated as level 3, a prediction characteristic with high water solubility was obtained, and good prediction accuracy was obtained, and the structural formula was relatively complex and unpredictability was high.

[0238] Regarding the prediction target data evaluated as level 4, a prediction characteristic with high water solubility was obtained, and good prediction accuracy was obtained, and the structural formula was complex and unpredictability was high.

[0239] (3) Embodiment 3

[0240] Fig.15 is a diagram showing Example 3. Fig.15 As shown, in Embodiment 3, by using multiple prediction target data,

[0241] Urea,

[0242] Methanol,

[0243] ·caffeine,

[0244] Digoxin

[0245] The predicted characteristics are predicted, and the intervals into which the attribute values ​​are divided are determined, and the appropriateness is evaluated (see Table 1510). In addition, the material properties of the above-mentioned prediction object data (here, water solubility) are all known items, so they are shown in Table 1520 for the purpose of verifying the prediction accuracy.

[0246] According to Table 1510, the relationship between the indicators of each prediction object data and the evaluation results is as follows.

[0247] Regarding the prediction target data evaluated as Level 1 and Level 2, a prediction property with high water solubility was obtained, and good prediction accuracy was obtained, but the structural formula was simple and the unpredictability was low.

[0248] Regarding the prediction target data evaluated as level 3, a prediction characteristic with high water solubility was obtained, and good prediction accuracy was obtained, and the structural formula was relatively complex and unpredictability was high.

[0249] · Regarding the prediction target data evaluated as level 4, the predicted characteristics of high water solubility were obtained, but in fact, the water solubility of the material was low, and the prediction accuracy was obviously low. In addition, the structural formula was complex and the unpredictability was high.

[0250] (4) Relationship between indicators and evaluation results and verification examples of prediction characteristics

[0251] According to the above Examples 1 to 3, it can be seen that prediction data having high water solubility prediction characteristics with good prediction accuracy and having a relatively complex structural formula (i.e., prediction data with a high possibility of becoming a useful new material) is likely to be evaluated as Level 3.

[0252] Therefore, the following will further verify how much higher the water solubility of the prediction target data (sucrose, lactose, caffeine) evaluated as level 3 in Examples 1 to 3 is compared with ordinary materials (in other words, whether it can be regarded as an objectively higher prediction characteristic).

[0253] Specifically, the frequency distribution of water solubility was generated based on the water solubility data set disclosed in the following non-patent document, and the water solubility of the prediction target data evaluated as level 3 in Examples 1 to 3 was verified.

[0254] [Non-patent document] JSDelaney, "Estimating Aqueous Solubility Directly from Molecular Structure", Journal of chemical information and computer sciences, p1000-p1005, May 24, 2004

[0255] Fig.16 is a diagram showing a verification example of the predicted characteristics. Fig.16 As shown, it can be seen that the water solubility of the prediction target data evaluated as level 3 in Examples 1 to 3 is also objectively high. That is, the prediction characteristics of the prediction target data output by the prediction device 130 and the evaluation results of the prediction target data can be said to accurately represent the possibility of becoming a useful new material.

[0256] <Conclusion>

[0257] As is clear from the above description, the prediction device of the first embodiment

[0258] A learning data set for generating a learning completion prediction model is obtained, and a plurality of intervals for dividing the attribute values ​​are determined based on the frequency distribution of the attribute values ​​calculated between a plurality of material data included in the learning data set.

[0259] By determining into which of a plurality of intervals a property value calculated between material data of the prediction target data and a plurality of material data included in the learning data set is classified, the suitability of the prediction target data with respect to two opposing indices is evaluated.

[0260] By inputting the prediction target data, the prediction characteristics predicted by the learned model are displayed in association with the evaluation results.

[0261] Thus, according to the first embodiment, even when predicting a prediction characteristic that satisfies a predetermined condition using a learned prediction model, prediction target data with a low possibility of becoming a useful new material can be excluded.

[0262] As a result, according to the prediction device of the first embodiment, it is possible to reduce the prediction target data for material generation and verification experiments, thereby improving the efficiency of new material development.

[0263] [Second embodiment]

[0264] In the first embodiment described above, each interval is determined by calculating the median, the first quartile, the third quartile, and the interquartile range as summary statistics and determining the coefficients α1 and α2. However, the method of determining each interval is not limited to this.

[0265] For example, as a summary statistic, you can also calculate

[0266] The average of the minimum property values ​​of each material,

[0267] The standard deviation of the minimum property values ​​for each material,

[0268] And multiple intervals are determined as,

[0269] The intervals into which the attribute values ​​are divided such that "attribute value < (mean value + β1 × standard deviation)" are obtained.

[0270] The attribute value is divided into intervals such that "(average value + β1 × standard deviation) ≤ attribute value < (average value + β2 × standard deviation)",

[0271] The attribute value is divided into intervals such that "(average value + β2 × standard deviation) ≤ attribute value < (average value + β3 × standard deviation)",

[0272] The intervals into which the attribute values ​​satisfy "(average value + β3 × standard deviation) ≤ attribute value" are divided.

[0273] In the first embodiment described above, a case is described where four sections that do not overlap with each other are determined. However, the number of sections to be determined is not limited to four, and may be three or more, as long as the sections do not overlap with each other.

[0274] In the first embodiment, the case where prediction target data are input to the prediction device 130 one by one is described, but a plurality of prediction target data may be input to the prediction device 130. In this case, the prediction device 130 may be configured to display prediction characteristics and evaluation results of only the prediction target data that has a high possibility of becoming a useful new material among the plurality of prediction target data.

[0275] In the first embodiment, the case where the exclusion object is set is described, but for example, the configuration may be such that the selection object is set.

[0276] When it is determined that the minimum property value calculated for the structural formula (structural formula X) of the material included in the "material data" of the prediction target data 700 is classified into the selection target interval (for example, the third interval close to the interval including the predetermined summary statistic), or

[0277] When the structural formula (structural formula X) of the material included in the "material data" of the prediction target data 700 is evaluated as a selection target (for example, level 3),

[0278] The structural formula (structural formula X) of the selected material can be controlled and input into the learning completion prediction model 902 .

[0279] Furthermore, in the above-mentioned first embodiment, a case where the learning device and the prediction device are realized by separate devices has been described, but the learning device and the prediction device may be realized as an integrated device.

[0280] In addition, the present invention is not limited to the configurations listed in the above embodiments, the combinations with other elements, etc. These points can be changed within the scope not departing from the gist of the present invention and can be appropriately determined according to the application mode.

[0281] This application claims priority based on Japanese Patent Application No. 2022-79269 filed on May 13, 2022, and all contents of the Japanese Patent Application are cited in this application.

[0282] Explanation of symbols

[0283] 100 Material Design System

[0284] 110 Material properties experimental device

[0285] 120 Learning Devices

[0286] 130 Prediction Device

[0287] 140 Material Design Device

[0288] 150 Material Generation Device

[0289] 160 Material properties experimental device

[0290] 201 Learning Data Acquisition Department

[0291] 202 Learning Department

[0292] 211 Learning Data Attribute Value Calculation Department

[0293] 212 interval determination unit

[0294] 221 prediction target data acquisition unit

[0295] 222 Forecast Department

[0296] 223 Prediction object data attribute value calculation unit

[0297] 224 Evaluation Department

[0298] 225 Display Department

[0299] 400 learning datasets

[0300] 700 Prediction object data

[0301] 902 Learning to complete the prediction model

Claims

1. A prediction device, include: An interval determination unit that obtains a learning data set for generating a learning completion model and determines a plurality of intervals for dividing the attribute value based on a frequency distribution of the attribute value calculated between a plurality of data included in the learning data set; an evaluation unit that evaluates the suitability of the data of the prediction target with respect to an opposite indicator by determining into which of the plurality of intervals a property value calculated between the data of the prediction target and the plurality of data is divided; and The display unit displays the predicted value predicted by the learned model by inputting the data of the prediction target in association with the evaluation result of the evaluation unit.

2. The prediction device according to claim 1, in, The interval determination unit calculates summary statistics for attribute values ​​calculated between the plurality of data and determines lower limit values ​​or upper limit values ​​for defining the attribute values ​​of the plurality of intervals.

3. The prediction device according to claim 2, in, The interval determination unit determines three or more intervals that do not overlap each other.

4. The prediction device according to claim 3, in, The evaluation unit evaluates the suitability of the data to be predicted based on which interval, among the three or more intervals, the attribute value calculated between the data to be predicted and the plurality of data is divided into, which interval is close to an interval including a predetermined summary statistic.

5. The prediction device according to claim 3, in, The evaluation unit excludes the data of the prediction object from the data to be input into the learning completion model when it is determined that the attribute value calculated between the data of the prediction object and the multiple data is divided into an interval farthest from the interval containing the specified summary statistic among the three or more intervals.

6. The prediction device according to claim 3, in, The evaluation unit selects the data of the prediction object as the data to be input into the learning completion model when it is determined that the attribute value calculated between the data of the prediction object and the multiple data is divided into the Rth interval among the three or more intervals that is close to the interval containing the specified summary statistic.

7. The prediction device according to claim 1, further comprising: include: a learning data attribute value calculation unit that calculates distances between a plurality of data included in the learning data set; and The learning data minimum attribute value extraction unit extracts the minimum distance from the distances with other data calculated for each of the plurality of data. The section determination unit determines a plurality of sections for dividing the minimum distance based on the extracted frequency distribution of the minimum distance.

8. The prediction device according to claim 7, in, The learning data attribute value calculation unit calculates the distance between the i-th (1≤i≤N) data among the N data (N is an arbitrary integer) included in the learning data set and the (N-1) data other than the i-th data. The learning data minimum attribute value extraction unit extracts the minimum distance from among the (N-1) distances calculated for the i-th data.

9. The prediction device according to claim 7, further comprising: include: A prediction object data attribute value calculation unit, which calculates the distance between the prediction object data and the plurality of data; and The prediction target data minimum attribute value extraction unit extracts the minimum distance among the distances between the prediction target data and the plurality of data. The evaluation unit evaluates suitability of the data of the prediction target with respect to an opposite index by determining into which of the plurality of intervals the minimum distance extracted for the data of the prediction target is divided.

10. A material design system, include: The prediction device according to claim 1; and A material design device generates material design data by inputting data of a prediction object judged by the evaluation unit of the prediction device as having attribute values ​​calculated between the multiple data divided into specified intervals and predicted values ​​satisfying specified conditions by the learning completion model of the prediction device.

11. The material design system according to claim 10, in, It also has a learning device that generates a learning completion model based on a learning data set. The prediction device predicts the prediction value by inputting the data of the prediction object into the learned model generated by the learning device.

12. A prediction method, for executing the following steps by a computer: An interval determination step of obtaining a learning data set for generating a learning completion model and determining a plurality of intervals for dividing the attribute value based on a frequency distribution of the attribute value calculated between a plurality of data included in the learning data set; an evaluation step of evaluating the suitability of the data of the prediction target with respect to an opposite indicator by determining into which of the plurality of intervals the attribute value calculated between the data of the prediction target and the plurality of data is divided; and The display step displays the predicted value predicted by the learned model by inputting the data of the prediction object in association with the evaluation result in the evaluation step.

13. A prediction program for causing a computer to execute the following steps: An interval determination step of obtaining a learning data set for generating a learning completion model and determining a plurality of intervals for dividing the attribute value based on a frequency distribution of the attribute value calculated between a plurality of data included in the learning data set; an evaluation step of evaluating the suitability of the data of the prediction target with respect to an opposite indicator by determining into which of the plurality of intervals the attribute value calculated between the data of the prediction target and the plurality of data is divided; and The display step displays the predicted value predicted by the learned model by inputting the data of the prediction object in association with the evaluation result in the evaluation step.

Images