JP2014074994A - Evaluation support method, information processor, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To efficiently determine an evaluation function for evaluating data including values of a plurality of evaluation indexes.SOLUTION: A storage part 11 stores a plurality of data samples 13 which respectively include values 14 of two or more evaluation indexes. An operation part 12 receives an input of superiority/inferiority information 17 showing relative superiority/inferiority in the plurality of data samples 13 about at least a portion of the plurality of data samples 13. The operation part 12 determines whether weight Wis settable such that a function value 16 suiting the superiority/inferiority relation between the plurality of data samples 13 corresponding to the superiority/inferiority information 17 is calculated when the values 14 of the two or more evaluation indexes of each data sample 13 are substituted for a variable Xabout at least one evaluation function among a plurality of evaluation functions 15 respectively including the variable Xand the weight Wcorresponding to two or more evaluation indexes. The operation part 12 selects an evaluation function 15 determined that the weight Wcan be set among the plurality of evaluation functions 15.

Description

  The present invention relates to an evaluation support method, an information processing apparatus, and a program.

  Currently, computer simulation is used in various fields as the computing power of computers increases. For example, in order to support the design of mechanical products and electronic products, evaluation indices such as response speed and manufacturing cost when parameters such as temperature and pressure are set to certain values are predicted through computer simulation. As a result, it becomes easy to find a combination of appropriate parameters and appropriate evaluation indexes, and it is possible to efficiently design mechanical products and electronic products by reducing the number of times of producing prototypes.

  When analyzing the results of computer simulation, mathematical optimization techniques are often used. The mathematical optimization technique includes numerical analysis that receives a numerical value as an input and outputs a numerical value as a calculation result, and mathematical expression processing that receives a mathematical expression as an input and processes the mathematical expression symbolically. Examples of formula processing methods include Cylindrical Algebraic Decomposition (CAD) and Quantifier Elimination (QE). By using mathematical expression processing technology, for example, it is possible to determine the presence or absence of an optimal solution that satisfies a constraint condition, or to calculate a region or boundary that satisfies a specific condition on a graph.

  When there is a trade-off relationship between a plurality of evaluation indexes, multi-objective optimization may be performed using the mathematical optimization technique as described above to try to optimize the entire plurality of evaluation indexes. For example, with respect to mathematical expression processing technology, a method has been proposed in which a plurality of evaluation functions are expressed by a polynomial and the plurality of evaluation functions are mathematically processed by a QE method in order to support the design of the slider shape of the hard disk. By using mathematical expression processing, it is possible to calculate and display an area that can be taken by a plurality of evaluation index values. In addition, regarding numerical analysis technology, a method for calculating an optimal solution that minimizes a weighted sum of values of a plurality of evaluation indexes using a result of repeatedly executing computer simulation has been proposed.

  If mathematical expression processing technology is applied, the number of executions of computer simulation can be reduced, so that the calculation load can be reduced compared to the case of applying numerical analysis technology. However, when the mathematical expression processing technique is applied, an approximate expression is calculated using data obtained by computer simulation in order to obtain a mathematical expression that is a target of mathematical expression processing. Therefore, the accuracy of approximation may be a problem when applying mathematical expression processing technology. Therefore, when it is difficult to obtain a solution with the expected accuracy by the mathematical expression processing technique, the above numerical analysis technique can be applied.

JP 2009-169557 A

  By the way, it is often difficult to uniformly determine whether or not certain data including a plurality of evaluation index values is better than other data. For example, there are two evaluation indexes that are in a trade-off relationship, and through computer simulation, the value of one evaluation index is slightly disadvantageous compared to the other data, but the value of the other evaluation index is large and advantageous Is acquired. At this time, whether or not the acquired data can be said to be better than the other data depends on the product design policy and the user's intention to judge the quality.

  For this reason, when evaluating data using an evaluation function that outputs an evaluation index for comprehensively judging whether the data is good or bad, an evaluation function that matches the user's intention is determined. For example, an appropriate evaluation function is determined by evaluating several data samples while changing the form of the evaluation function (degree, number of terms, etc.) and the weight included in the weighted sum. However, such a method for determining the evaluation function is performed by the user on a trial and error basis based on their own experience. Therefore, in such an evaluation function determination method, whether or not the determined evaluation function is appropriate depends on the ability of the user. Further, such an evaluation function determination method has a large work load for determining the evaluation function.

  Accordingly, in one aspect, the present invention provides an evaluation support method, an information processing apparatus, and a program capable of efficiently determining an evaluation function for evaluating data including a plurality of evaluation index values. With the goal.

  In one aspect, an evaluation support method executed by a computer is provided. The computer obtains a plurality of data samples each including two or more evaluation index values. In addition, the computer accepts input of superiority / inferiority information indicating relative superiority or inferiority among the plurality of data samples for at least some of the plurality of data samples. Further, the computer assigns values of two or more evaluation indexes of each data sample to at least one evaluation function among a plurality of evaluation functions each including a variable and a weight corresponding to two or more evaluation indexes. It was determined whether the weight can be set so that a function value that fits the superiority or inferiority relationship between multiple data samples according to superiority information is calculated, and it was determined that the weight could be set from multiple evaluation functions Select an evaluation function.

  In one aspect, an evaluation function for evaluating data including a plurality of evaluation index values can be efficiently determined.

It is the figure which showed the example of the information processing apparatus which concerns on 1st Embodiment. It is the figure which showed the example of the hardware of the information processing apparatus which concerns on 2nd Embodiment. It is the figure which showed the example of the information processing apparatus which concerns on 2nd Embodiment. It is a figure explaining single objective optimization and multi-objective optimization. It is a figure explaining the determination method of superiority or inferiority with respect to evaluation data according to the second embodiment. It is a figure explaining the change method of superiority or inferiority according to the second embodiment. It is the figure explaining the structure of the data table which concerns on 2nd Embodiment, and matching of superiority. It is a figure explaining the production | generation method of the constraint equation which concerns on 2nd Embodiment. It is the figure which showed the flow of the process by the information processing apparatus which concerns on 2nd Embodiment. It is the figure which showed the flow of the process about the production | generation process of the evaluation function which concerns on 2nd Embodiment. It is the figure which showed the flow of the process about the optimization calculation which concerns on 2nd Embodiment. It is the figure shown about the calculation method of the appropriate parameter with respect to each of several weight group. It is the figure which showed the flow of a process about the parameter calculation which concerns on 2nd Embodiment. It is the figure which showed the flow of the process about the optimization calculation which concerns on the modification of 2nd Embodiment. It is the figure which showed the example of the information processing apparatus which concerns on 3rd Embodiment. It is a figure explaining the superiority or inferiority quantification method which concerns on 3rd Embodiment. It is the figure which showed the flow of the process by the information processing apparatus which concerns on 3rd Embodiment. It is the figure which showed the flow of the process about the calculation process of the constraint equation which concerns on 3rd Embodiment.

Hereinafter, embodiments will be described with reference to the drawings.
[First Embodiment]
First, the first embodiment will be described.

FIG. 1 is a diagram illustrating an example of an information processing apparatus according to the first embodiment. As illustrated in FIG. 1, the information processing apparatus 10 according to the first embodiment includes a storage unit 11 and a calculation unit 12.
The storage unit 11 may be a volatile storage device such as a Random Access Memory (RAM), or may be a nonvolatile storage device such as a Hard Disk Drive (HDD) or a flash memory. The calculation unit 12 may be a processor such as a central processing unit (CPU) or a digital signal processor (DSP). The computing unit 12 may be an electronic circuit other than a processor such as an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA). For example, the calculation unit 12 executes a program stored in the storage unit 11 or another memory.

  The storage unit 11 stores a plurality of data samples 13A,..., 13C each including two or more evaluation index values 14A,. Moreover, the calculating part 12 receives the input of the superiority / inferiority information 17 which shows the relative superiority in the some data sample 13A, ..., 13C about at least one part of several data sample 13A, ..., 13C.

In addition, the calculation unit 12 sets each data sample 13A,..., 13C as the evaluation index X _j for at least one evaluation function among the plurality of evaluation functions 15 each including two or more evaluation indices X _j and weights W _j. .., 16C that match the superiority or inferiority relationship between the plurality of data samples 13A,..., 13C according to the superiority / inferiority information 17 when the two or more evaluation index values 14A,. It is determined whether the weight W _j can be set.

The evaluation function 15 is, for example, an expression Q (X) expressed by the following expression (1) obtained by accumulating and adding the weight W _j and the evaluation index value X _{j in} the range of j from 1 to n. It may be set as follows. The evaluation function 15 may be set to include a quadratic term related to the value of the evaluation index, for example, as shown in the following formula (2). Further, the evaluation function 15 may be set so as to include, for example, a third-order or higher term relating to the value of the evaluation index.

Here, consider a case where superiority / inferiority information 17 is set so that the value of the evaluation function becomes smaller as the data sample becomes better.
If the data sample 13C is superior to the data sample 13A, the function value Q _C corresponding to the data sample 13C is smaller than the function value Q _A corresponding to the data sample 13A. In this case, superiority / inferiority information 17 as shown in the following formula (3) is obtained. Conversely, if the data sample 13A is superior to the data sample 13C, the function value Q _C corresponding to the data sample 13C is larger than the function value Q _A corresponding to the data sample 13A. In this case, superiority information 17 as shown in the following formula (4) is obtained.

  Similarly, superiority / inferiority information 17 is obtained for other relationships relating to the plurality of data samples 13A,..., 13C. The arithmetic unit 12 receives input of such superiority information 17.

The computing unit 12 selects the evaluation function 15 that is determined to be able to set the weight W _j from the plurality of evaluation functions 15. For example, the arithmetic unit 12 may determine whether it is possible to set a weight W _j such that the function values 16A,... Further, the calculation unit 12 may select the evaluation function 15 having a lower order from the evaluation functions 15 that can set the weights for calculating the function values 16A,. Good.

  According to the information processing apparatus 10 according to the first embodiment, an evaluation function capable of setting a weight such that a function value that matches superiority information is calculated is selected. Therefore, it is possible to efficiently determine an evaluation function for evaluating data including a plurality of evaluation index values.

[Second Embodiment]
Next, a second embodiment will be described.
FIG. 2 is a diagram illustrating an example of hardware of the information processing apparatus according to the second embodiment.

  As illustrated in FIG. 2, the information processing apparatus 100 includes, for example, a CPU 901, a RAM 902, an HDD 903, an image signal processing unit 904, an input signal processing unit 905, a disk drive 906, and a communication interface 907. The CPU 901 is an example of the calculation unit 12 according to the first embodiment. The RAM 902 and the HDD 903 are an example of the storage unit 11 according to the first embodiment.

  The CPU 901 is a processor including an arithmetic unit that executes instructions described in a program. The CPU 901 loads at least a part of the program and data stored in the HDD 903 into the RAM 902 and executes instructions described in the program. Note that the CPU 901 may include a plurality of processor cores. The information processing apparatus 100 may be equipped with a plurality of CPUs 901. In this case, the information processing apparatus 100 can execute processes in parallel.

  The RAM 902 is a volatile memory for temporarily storing programs executed by the CPU 901 and data used for processing. Note that the information processing apparatus 100 may have a different type of memory from the RAM 902. Further, the information processing apparatus 100 may include a plurality of memories.

  The HDD 903 is an example of a nonvolatile storage device that stores programs such as an operating system (OS), firmware, or application software, data used for processing, and the like. Note that the information processing apparatus 100 may include a different type of storage device from the HDD 903, such as a flash memory or a solid state drive (SSD). Further, the information processing apparatus 100 may include a plurality of storage devices.

  The image signal processing unit 904 is controlled by the CPU 901 and outputs an image to the display device 91 connected to the information processing apparatus 100. The display device 91 is a display device such as a Cathode Ray Tube (CRT) display, a Liquid Crystal Display (LCD), a Plasma Display Panel (PDP), or an Organic Electro-Luminescence Display (OELD).

  The input signal processing unit 905 acquires an input signal from the input device 92 connected to the information processing apparatus 100 and notifies the CPU 901 of the input signal. As the input device 92, for example, a mouse, a keyboard, a touch panel, a touch pad, a trackball, a remote controller, a button switch, or the like can be used.

  The disk drive 906 is a reading device that reads programs and data recorded on the recording medium 93. As the recording medium 93, for example, a magnetic disk such as a Flexible Disk (FD) or HDD, an optical disk such as a Compact Disc (CD) or a Digital Versatile Disc (DVD), or a magneto-optical disk such as a Magneto-Optical disk (MO) is used. be able to. For example, the disk drive 906 is controlled by the CPU 901 and stores a program and data read from the recording medium 93 in the RAM 902 or the HDD 903.

  The communication interface 907 is an interface for communicating with other computers via the network 94. The communication interface 907 may be a wired interface or a wireless interface.

FIG. 3 is a diagram illustrating an example of an information processing apparatus according to the second embodiment.
As illustrated in FIG. 3, the information processing apparatus 100 includes a parameter input unit 101, a simulation execution unit 102, an dominance determination unit 103, a determination result output unit 104, a storage unit 105, and an dominance information input unit 106. In addition, the information processing apparatus 100 includes a constraint expression generation unit 107, a mathematical expression processing unit 108, an evaluation function generation unit 109, a weight calculation unit 110, and an optimization calculation unit 111.

  The parameter input unit 101, the simulation execution unit 102, the superiority / inferiority determination unit 103, the determination result output unit 104, and the superiority / inferiority information input unit 106 can be realized as a module of a program executed by the CPU 901. Further, some or all of the functions of the parameter input unit 101, the simulation execution unit 102, the superiority / inferiority determination unit 103, the determination result output unit 104, and the superiority / inferiority information input unit 106 can be realized as an electronic circuit instead of software. .

  In addition, the constraint expression generation unit 107, the mathematical expression processing unit 108, the evaluation function generation unit 109, the weight calculation unit 110, and the optimization calculation unit 111 can be realized as modules of a program executed by the CPU 901. Also, some or all of the functions of the constraint expression generation unit 107, the mathematical expression processing unit 108, the evaluation function generation unit 109, the weight calculation unit 110, and the optimization calculation unit 111 can be realized as an electronic circuit instead of software. It is. The storage unit 105 is a storage area secured in the RAM 902 or the HDD 903.

  The parameter input unit 101 receives parameters used for calculating the evaluation index. For example, the parameter may be input by the user using the input device 92 or may be acquired from an external terminal device or server device via the network 94. Further, the parameter may be, for example, a parameter related to the design of the evaluation object, a parameter that defines the setting contents in the manufacturing process of the evaluation object, a parameter that defines the actual operation condition of the evaluation object, or the like.

  The evaluation index may be, for example, an index that represents the performance of the evaluation target, an index that represents the cost that occurs during the manufacture of the evaluation target, or the like. Examples of performance evaluation indexes include durability, power consumption, and operating environment (temperature, humidity, atmospheric pressure). In addition, as an evaluation index related to cost, there are cost, manufacturing period, yield, and the like. In the second embodiment, a multi-objective optimization problem that considers a plurality of evaluation indexes that are in a trade-off relationship with each other, such as a relationship between performance and cost, is considered.

  Multi-objective optimization is often used in design and evaluation scenes in the manufacturing field, such as hard disk head design, semiconductor memory design, automobile collision safety design, and engine performance design. For example, multi-objective optimization is used for optimizing a substrate material, a substrate shape, a wiring pattern, and the like in the design of a semiconductor memory. Multi-objective optimization is used for optimizing body materials and body shapes in collision safety design for automobiles. Multi-objective optimization is used to optimize gasoline quantity and injection timing in engine performance design.

  The parameters input to the parameter input unit 101 are stored in the storage unit 105. The parameters stored in the storage unit 105 are read by the simulation execution unit 102. Note that the parameters input to the parameter input unit 101 may be directly input to the simulation execution unit 102.

The simulation execution unit 102 simulates the evaluation object and the environment in which the object is placed based on the parameters, and calculates an evaluation index for the object and the environment. For example, the simulation execution unit 102 calculates a value of an evaluation index indicating the performance of the head and a value of an evaluation index indicating the manufacturing cost in response to input of a parameter that defines the shape of the head. In addition, the simulation execution unit 102 calculates an evaluation index value indicating the amount of soot discharged from the engine or the amount of NO _x in accordance with the input of the parameter.

  A plurality of evaluation index values calculated by the simulation execution unit 102 (hereinafter sometimes referred to as evaluation data) are stored in the storage unit 105 in association with parameters. The evaluation data stored in the storage unit 105 is read by the superiority / inferiority determination unit 103. Note that the evaluation data may be directly input to the superiority / inferiority determination unit 103. The storage unit 105 stores a plurality of evaluation data calculated based on a plurality of parameters input to the parameter input unit 101.

  The superiority / inferiority determining unit 103 determines the superiority or inferiority of each evaluation data stored in the storage unit 105. For example, when one evaluation data has a better value than other evaluation data for at least one evaluation index, it is assumed that the one evaluation data is better than the other evaluation data. Further, when one evaluation data is better than all other evaluation data, the superiority / inferiority determination unit 103 determines that the one evaluation data is excellent evaluation data. Further, the superiority / inferiority determination unit 103 determines that evaluation data other than excellent evaluation data is inferior evaluation data.

The determination result by the superiority / inferiority determination unit 103 is input to the determination result output unit 104. The result of determination by the superiority / inferiority determination unit 103 is stored in the storage unit 105.
The determination result output unit 104 presents superiority or inferiority of each evaluation data to the user. For example, the determination result output unit 104 displays the superiority or inferiority of each evaluation data on the display device 91 via the image signal processing unit 904. As a display method, for example, a method of displaying an object indicating each evaluation data on a map and giving text data indicating superiority or inferiority to each evaluation data can be considered. In addition, a display method in which each object is color-coded according to superiority or inferiority can be considered. A method of displaying a table in which each evaluation data is associated with text data indicating superiority or inferiority is also conceivable.

  A user who refers to information indicating superiority or inferiority of each evaluation data may change the superiority or inferiority of certain evaluation data. In this case, the superiority or inferiority information input unit 106 receives information on evaluation data in which superiority or inferiority is changed by the user.

  For example, when the user changes the evaluation data determined to be inferior evaluation data by the superiority / inferiority determination unit 103 to superior evaluation data, the information of the evaluation data and the change content are input to the superiority / inferiority information input unit 106. . Further, when the evaluation data determined to be superior evaluation data by the superiority / inferiority determination unit 103 is changed to evaluation data with inferior evaluation data, the information of the evaluation data and the content of the change are input to the superiority / inferiority information input unit 106. . Note that the information input to the superiority / inferiority information input unit 106 is stored in the storage unit 105 in association with the evaluation data information corresponding to the information.

  The constraint expression generation unit 107 generates a constraint expression based on each evaluation data stored in the storage unit 105 and the superiority or inferiority of each evaluation data. This constraint expression is a conditional expression expressing the superiority or inferiority of each evaluation data using the values of a plurality of evaluation indexes included in each evaluation data. For example, the constraint expression generation unit 107 uses an evaluation function expressed by a weighted sum of values of a plurality of evaluation indexes in order to quantify the superiority or inferiority of each evaluation data. The constraint expression generation unit 107 also generates a constraint expression that expresses a condition that is satisfied by the value of the evaluation function corresponding to each evaluation data.

As an example, let us consider a case where the value of the evaluation function of evaluation data inferior to the value of the evaluation function of excellent evaluation data becomes large. For example, when there are excellent evaluation data A, inferior evaluation data B, and inferior evaluation data C, the constraint expression generation unit 107 determines the relationship between the evaluation data A and B and the relationship between the evaluation data A and C. Are generated as two constraint expressions. When the value of the evaluation function corresponding to the evaluation data A is expressed as Q _A and the value of the evaluation function corresponding to the evaluation data _B is expressed as Q _B , the constraint expression expressing the relationship between the evaluation data A and B is It is expressed by (5). The evaluation when the value of the evaluation function corresponding to the data C and Q _C, constraint formula expressing the relationship between the evaluation data A and C is expressed by the following equation (6).

  As described above, the constraint expression generation unit 107 generates a constraint expression for a combination between excellent evaluation data and inferior evaluation data. At this time, the constraint equation generation unit 107 may generate constraint equations for all possible combinations, or may generate constraint equations for some combinations. For example, the constraint expression generation unit 107 relates to all possible combinations for a set of evaluation data including all evaluation data whose superiority or inferiority has been changed by the user, all excellent evaluation data, or some or all inferior evaluation data. A constraint expression may be generated. In this way, by generating constraint equations for all evaluation data for which the user has changed superiority or inferiority, it is possible to reflect all the user's intentions in the calculation results.

Further, the constraint expression generation unit 107 acquires evaluation function information from the evaluation function generation unit 109. For example, the constraint expression generation unit 107 acquires evaluation function information as shown in the following expression (7). However, in the following formula (7), W _j is a variable representing the weight, and X _j is a variable representing the value of the evaluation index. However, X represents a set of variables X ₁ ,..., X _n representing evaluation index values.

The constraint expression generation unit 107 that has acquired the evaluation function information applies values and evaluation functions of a plurality of evaluation indexes corresponding to each evaluation data to the constraint expression. For example, when the evaluation index values (A ₁ , A ₂ ) corresponding to the excellent evaluation data A and the evaluation index values (B ₁ , B ₂ ) corresponding to the inferior evaluation data B are obtained, a constraint expression is generated. The unit 107 generates a constraint equation as shown in the following equation (8) based on the above equations (5) and (7). Similarly, the constraint expression generation unit 107 generates a constraint expression based on the evaluation function for other evaluation data. Information on a plurality of constraint expressions generated by the constraint expression generation unit 107 is input to the mathematical expression processing unit 108.

  The mathematical expression processing unit 108 uses mathematical expression processing to determine whether there is a set of weights that conforms to all the constraint expressions generated by the constraint expression generation unit 107. For example, CAD or QE can be used as the mathematical expression processing. For example, the explanation of QE is described in [Hirokazu Anai, Kazuhiro Yokoyama, “QE calculation algorithm and its application”, University of Tokyo Press].

QE is a technique for generating a logical expression that does not include a quantifier symbol from a logical expression that includes a quantifier symbol using a computer. For example, when QE is applied to the logical formula shown in the following formula (9), the logical formula shown in the following formula (10) is obtained. Note that x ₁ , x ₂ , and x ₃ with a quantifier symbol are bound variables, and y without a quantifier symbol is a free variable. As in this example, when QE is applied, a logical expression that does not include a quantifier symbol can be generated from a logical expression that includes a quantifier symbol.

  Also, as shown in the following formula (11), if QE is executed by adding existence symbols to all variables included in the logical expression, is there a set of variables that satisfy the constraint conditions included in the logical expression? It can be determined whether or not. For example, when QE is applied to the logical expression shown in the following expression (11), it is determined that there are variables x and y that satisfy the constraint conditions included in the logical expression. Even for a more complicated problem, it is possible to determine whether there is a set of variables that satisfy the constraint conditions included in the logical expression using QE. By using this method, it is possible to determine whether or not there is a set of weights that matches all the constraint expressions generated by the constraint expression generation unit 107.

  The mathematical expression processing unit 108 determines whether there is a set of weights that conforms to all the constraint expressions, and inputs the determination result to the constraint expression generation unit 107 and the evaluation function generation unit 109. If there is a set of weights that matches all the constraint equations, the evaluation function generation unit 109 inputs information on the evaluation function used to generate the constraint equation to the weight calculation unit 110. In addition, the constraint formula generation unit 107 inputs the constraint formula information to the weight calculation unit 110. In addition, the constraint formula generation unit 107 stores the constraint formula information and the evaluation function information in the storage unit 105.

On the other hand, when there is no set of weights that matches all the constraint expressions, the evaluation function generation unit 109 changes the form of the evaluation function. At this time, the evaluation function generation unit 109 changes the original evaluation function to an evaluation function including a higher-order term. For example, when the original evaluation function is the evaluation function shown in the above equation (7), the evaluation function generation unit 109 uses the evaluation function including the quadratic term as shown in the following equation (12). Change to However, W _j and W _ij represent weights. Note that the evaluation function generation unit 109 may change the original evaluation function to an evaluation function including a third-order or higher term as shown in the following equation (13).

  Information on the evaluation function after the change is input to the constraint expression generation unit 107. When the evaluation function is changed, the constraint expression generation unit 107 generates a constraint expression based on the changed evaluation function. Information on the constraint expression generated by the constraint expression generation unit 107 is input to the mathematical expression processing unit 108. The mathematical expression processing unit 108 determines whether there is a set of weights that matches all the constraint expressions. If there is a set of weights that matches all the constraint equations, the evaluation function generation unit 109 inputs information on the evaluation function used to generate the constraint equation to the weight calculation unit 110. In addition, the constraint formula generation unit 107 inputs the constraint formula information to the weight calculation unit 110. In addition, the constraint formula generation unit 107 stores the constraint formula information and the evaluation function information in the storage unit 105.

As described above, the evaluation function changing process and the constraint expression generating process based on the changed evaluation function are repeatedly executed until a set of weights that match all the constraint expressions is obtained.
The weight calculation unit 110 calculates a set of weights using an evaluation function. First, the weight calculation unit 110 considers a plurality of evaluation indexes included in the evaluation function as coefficients, and considers the weight sets W ₁ ,..., W _n as variables.

  In addition, the weight calculation unit 110 generates a plurality of coefficient groups using a method such as an experimental design method, a Latin hypersquare method, or a random sampling method. The experimental design method is a method of collecting experimental data of factors that affect the purpose under constraint conditions and analyzing the collected experimental data. However, experimental data may be collected by performing experiments or by performing computer simulations. The Latin supersquare method is an experimental design based on the Latin supersquare. The Latin supersquare is an m-row m-column (m is 2 or more) table in which m different values are arranged so that each value appears only once in each row and each column. The random sampling method is a method of randomly extracting data included in a certain data range.

  For example, when the Latin hypersquare method is used, the weight calculation unit 110 sets each value appearing in the table as a real number from −1 to 1, and arranges each value so that each value appears only once in each row and each column. Generate a super square. Then, the weight calculation unit 110 sets a real number group included in each row or each column of the Latin hypersquare to each coefficient group. When the random sampling method is used, the weight calculation unit 110 randomly extracts a real number included in the range from −1 to 1 n times, and sets the extracted n real numbers to one coefficient group. Further, the weight calculation unit 110 repeatedly executes this process to generate a plurality of coefficient groups.

  For each of the plurality of coefficient groups generated as described above, the weight calculation unit 110 calculates a set of weights that minimizes the evaluation function under the condition that the set of weights matches all constraint equations. In addition, the weight calculation unit 110 calculates, for each coefficient group, a set of weights that maximizes the evaluation function under the condition that the set of weights matches all constraint equations. At this time, since all constraint equations and evaluation functions are linear equations for each coefficient, a set of weights can be obtained by linear programming.

  In the above example, each coefficient is a real number included in the range from −1 to 1 in setting the coefficient group. This is because the ratio of the coefficients included in the coefficient group is important in calculating the weight set, and even if a plurality of coefficient sets obtained by multiplying each coefficient by a constant is used, it does not contribute to the calculation of the weight set. Therefore, the range that the coefficient can take can be arbitrarily set, but it is preferable to set the upper limit and the lower limit of each coefficient in a range from −1 to 1 as described above.

  A plurality of sets of weights calculated for each coefficient group by the weight calculation unit 110 are input to the optimization calculation unit 111. The optimization calculation unit 111 reads evaluation function information from the storage unit 105. In addition, the optimization calculation unit 111 calculates a parameter that minimizes the value of the evaluation function for each of the plurality of sets of weights. For example, the optimization calculation unit 111 prepares a plurality of parameter candidates that minimize the value of the evaluation function, and acquires evaluation data corresponding to each candidate. In this case, the optimization calculation unit 111 inputs the parameters of each candidate to the simulation execution unit 102 and acquires evaluation data calculated by the simulation execution unit 102.

  In addition, the optimization calculation unit 111 calculates the value of the evaluation function using the evaluation data corresponding to each candidate, and extracts the parameter having the smallest evaluation function value from the parameters prepared as candidates. For example, parameter extraction can be realized using an optimization algorithm such as the Nelder-Mead method, the Powell method, the steepest descent method, or the conjugate gradient method. In addition, the optimization calculation unit 111 outputs the extracted parameters as appropriate parameters. Note that the optimization calculation unit 111 may output a plurality of evaluation index values corresponding to appropriate parameters.

  The example of the information processing apparatus 100 according to the second embodiment has been described above. According to the information processing apparatus 100, the weight set is calculated when there is a set of weights that conforms to the constraint expression, and the evaluation function is changed when there is no set of weights that conforms to the constraint expression. For this reason, useless calculation such as continuing to search for a weight set when there is no weight set that matches the constraint equation is not performed, and it is possible to efficiently calculate a weight set that matches the constraint equation.

In addition, since the user's change regarding the superiority or inferiority of the evaluation data is received and the weight set is calculated using the constraint equation based on the superiority or inferiority after the change, the weight set reflecting the user's intention is obtained.
Also, according to the information processing apparatus 100, a plurality of weight sets are calculated using mathematical formula processing and linear programming, and parameter values that minimize the value of the evaluation function using the plurality of weight sets are obtained. Calculated. When calculating multiple sets of weights, computer simulation is performed to obtain multiple evaluation data, but the calculation of the parameter that minimizes the value of the evaluation function is nothing but single-objective optimization. The number of computer simulation executions can be kept low compared to the case of computerization.

  As a method of solving the multi-objective optimization problem, for example, a method of searching for an optimal solution from evaluation data obtained by repeatedly executing computer simulation using a genetic algorithm or the like is considered. Genetic algorithms are one of the effective optimization methods for problems that are difficult to deal with by classical mathematical programming. However, when a genetic algorithm is executed to obtain an optimal solution, computer simulation is executed many times in order to calculate a large number of evaluation data.

  On the other hand, according to the information processing apparatus 100 described above, since the calculation of the parameter that minimizes the value of the evaluation function is reduced to the single-objective optimization, the number of computer simulations is significantly higher than the method using a genetic algorithm or the like. It becomes possible to reduce it to less. As a result, according to the information processing apparatus 100, it is possible to keep the calculation load required for parameter optimization low compared to the case of performing multi-objective optimization as described above.

Hereinafter, considering the two evaluation indexes (evaluation index # 1 and evaluation index # 2) that are in a trade-off relationship with each other, the processing by the information processing apparatus 100 will be further described.
FIG. 4 is a diagram for explaining single-objective optimization and multi-objective optimization. FIG. 4A is a diagram for explaining single-objective optimization. FIG. 4B is a diagram for explaining multi-objective optimization. Single-objective optimization is a method of calculating a parameter that makes an evaluation index value corresponding to one objective an appropriate value (for example, a minimum value). On the other hand, multi-objective optimization is a method of calculating a parameter that optimizes a combination of values of a plurality of evaluation indexes respectively corresponding to a plurality of purposes.

  For example, when a smaller evaluation index value is better, single-objective optimization can be performed by calculating evaluation data corresponding to a plurality of parameters by computer simulation and selecting a parameter that minimizes the evaluation index value. . It is also possible to perform single-objective optimization by calculating an approximate expression that approximates a plurality of evaluation data, and calculating a parameter that minimizes the value of the approximate expression. According to the single-objective optimization, for example, a parameter that minimizes the value of the evaluation index # 1 and a parameter that minimizes the value of the evaluation index # 2 can be obtained.

  However, when the evaluation indices # 1 and # 2 are in a trade-off relationship with each other, the parameter that minimizes the value of the evaluation index # 1 may not be a parameter that optimizes the combination of the evaluation indices # 1 and # 2. Many. Therefore, multi-objective optimization is performed to calculate a parameter that optimizes the combination of evaluation indices # 1 and # 2.

  For example, as shown in FIG. 4B, when at least the value of evaluation index # 1 or the value of evaluation index # 2 has a value of one evaluation data smaller than the values of all other evaluation data In addition, it is determined that the one evaluation data is excellent evaluation data. In this case, it can be said that a parameter group corresponding to a set of excellent evaluation data is an appropriate parameter in multi-objective optimization. As described above, in multi-objective optimization, excellent evaluation data is extracted by simultaneously considering the values of a plurality of evaluation indexes, and a parameter corresponding to the extracted excellent evaluation data is obtained as an appropriate parameter.

  As one method for realizing multi-objective optimization, a method of repeatedly executing computer simulation while changing parameters, calculating a large amount of evaluation data, and selecting appropriate evaluation data from the large number of evaluation data can be considered. . For example, a method of searching for an optimal solution using a genetic algorithm or the like is considered. Genetic algorithms are one of the effective optimization methods for problems that are difficult to deal with by classical mathematical programming. When an optimal solution is obtained by executing a genetic algorithm, computer simulation is executed many times in order to calculate a large number of evaluation data. However, computer simulation is a processing with a high calculation load. Therefore, it is preferable to reduce the number of executions of computer simulation as much as possible.

  As a method for quantifying the superiority or inferiority of the evaluation data, for example, a method using a weighted sum of the value of the evaluation index # 1 and the value of the evaluation index # 2 can be considered. For example, the value of the evaluation function that is a weighted sum obtained by adding the product of the first weight and the value of the evaluation index # 1 and the product of the second weight and the value of the evaluation index # 2 is superior or inferior to the evaluation data. I think that represents. In this case, the first and second weights are calculated so that, for example, the value of the evaluation function corresponding to excellent evaluation data is smaller than the value of the evaluation function corresponding to inferior evaluation data. If the first and second weights are calculated, an appropriate parameter can be obtained by obtaining a parameter that minimizes the evaluation function.

  When calculating the first and second weights, a computer simulation is executed to obtain a plurality of evaluation data. However, the process for obtaining the parameter that minimizes the value of the evaluation function is nothing other than single-objective optimization, and the number of times of computer simulation execution is reduced compared to the method of performing multi-objective optimization without using the evaluation function as described above. Can be suppressed.

FIG. 5 is a diagram illustrating a method for determining superiority or inferiority with respect to evaluation data according to the second embodiment. The superiority or inferiority of the evaluation data can be determined by a method as shown in FIG. 5, for example. 5 (A) is a diagram showing a method of determining the superiority or inferiority of the evaluation data D _2. Figure 5 (B) is a diagram showing a method of determining the superiority or inferiority of the evaluation data D _1.

The region hatched in FIG. 5 (A) indicates a region in which values of at least metrics # 1 value or metric # 2 becomes a better value than the evaluation data D _2. The region hatched in FIG. 5 (B) shows a region where the value of the least metrics # 1 value or metric # 2 is better than the evaluation data D _1.

In the example of FIG. 5A, evaluation data D ₁ exists as evaluation data better than the evaluation data D ₂ . Further, in the example of FIG. 5 (B), other evaluation data no better than the evaluation data D _1. Accordingly, the evaluation data D ₁ is determined to be excellent evaluation data, and the evaluation data D ₂ is determined to be inferior evaluation data.

  Thus, the superiority or inferiority of each evaluation data can be determined based on the relationship between at least two evaluation data. However, the superiority or inferiority of the evaluation data determined by such a method may not coincide with the superiority or inferiority of the evaluation data considered by the user.

For example, in the example of FIG. 5, if the evaluation index # 1 is good, it may be determined that the evaluation data is excellent even if the evaluation index # 2 is bad. For example, in the design of engine performance, a user that emphasizes the discharge amount of soot than emissions of the NO _x can be good evaluation index indicating the emission of the NO _x, at worst evaluation index indicating a discharge amount of soot In some cases, the evaluation data having these evaluation indices is considered inferior. However, for users who place importance on evaluation index # 1, evaluation data with poor evaluation index # 1 is inferior evaluation data even if evaluation index # 2 is good. Therefore, it is preferable that the superiority or inferiority of the evaluation data can be changed according to the user's intention.

FIG. 6 is a diagram for explaining a method for changing superiority or inferiority according to the second embodiment.
Information indicating superiority or inferiority of the evaluation data is stored in the storage unit 105 in association with the information of each evaluation data. As shown in the example of FIG. 6A, when the user changes the superiority or inferiority of the evaluation data D ₂ to superior evaluation data, the superiority / inferiority information input unit 106 evaluates the evaluation among the information stored in the storage unit 105. Information indicating superiority or inferiority corresponding to the data D ₂ is changed from “inferior” to “excellent”. Conversely, as shown in the example of FIG. 6B, when the user changes the superiority / inferiority of the evaluation data D ₃ to the evaluation data inferior to the evaluation data, the superiority / inferiority information input unit 106 stores the information stored in the storage unit 105. among changes the information that indicates the relative merits corresponding to the evaluation data D ₃ from "excellent" to "poor".

FIG. 7 is a diagram for explaining the structure of the data table and the correspondence between superiority and inferiority according to the second embodiment.
As illustrated in FIG. 7, the storage unit 105 stores a data table that associates identification information (number), each parameter value, evaluation data, and information indicating superiority or inferiority. The information indicating the superiority or inferiority of each evaluation data includes the determination result by the superiority / inferiority determination unit 103 (“excellence (determination)”) and information indicating the superiority or inferiority of each evaluation data input by the user via the superiority / inferiority information input unit 106 (“ Dominance (manual) "). FIG. 7A shows the contents of the data table corresponding to FIG. FIG. 7B shows the contents of the data table corresponding to FIG. The constraint expression generation unit 107 generates a constraint expression using information as shown in FIG.

FIG. 8 is a diagram for explaining a constraint equation generation method according to the second embodiment. As an example, consider a method of generating a constraint equation based on the relationship between the three evaluation data D ₄ , D ₅ , and D ₆ shown in FIG. In the example of FIG. 8, the value of the evaluation index # 1 included in the evaluation data D ₄ is 11, the value of the evaluation index # 2 is 12. The value of the evaluation index # 1 included in the evaluation data D ₅ is 18, the value of the evaluation index # 2 is 16. The value of the evaluation index # 1 included in the evaluation data D ₆ is 17, the value of the evaluation index # 2 is 19. The evaluation data D ₄ is excellent evaluation data, and the evaluation data D ₅ and D ₆ are inferior evaluation data.

In this case, the constraint equation is generated based on the relationship between the evaluation data D ₄ and the evaluation data D ₅ and the relationship between the evaluation data D ₄ and the evaluation data D ₆ . For example, when the evaluation function is given by the above equation (7), the constraint equation generation unit 107 represents the following equation (14) based on the relationship between the evaluation data D ₄ and the evaluation data D _5. Generate constraint expressions. In addition, the constraint equation generation unit 107 generates a constraint equation represented by the following equation (15) based on the relationship between the evaluation data D ₄ and the evaluation data D ₆ . However, W ₁ is a weight corresponding to the evaluation index # 1, and W ₂ is a weight corresponding to the evaluation index # 2.

If there is evaluation data other than the evaluation data D ₄ , D ₅ , and D ₆ , constraint expressions are similarly generated for the other evaluation data.
When the constraint expression is generated as described above, the mathematical expression processing unit 108 determines whether there is a set of weights that matches all the constraint expressions. If there is no set of weights that matches all the constraint equations, the constraint equation is generated again after changing the evaluation function. When there is a set of weights that matches all the constraint expressions, the weight calculation unit 110 calculates the set of weights using the evaluation function used for generating the constraint expressions. An appropriate parameter is calculated using the set of weights calculated by the weight calculation unit 110.

Hereinafter, the flow of processing by the information processing apparatus 100 will be described.
FIG. 9 is a diagram illustrating a flow of processing by the information processing apparatus according to the second embodiment. Note that a plurality of parameters are input to the information processing apparatus 100 via the parameter input unit 101 before the processing is started. The number of input parameters and the numerical range of each parameter are set by the user. Each parameter is generated using, for example, an experimental design method, a Latin hypersquare method, or a random sampling method.

  (S101) The simulation execution unit 102 executes computer simulation for each parameter, and calculates a plurality of evaluation data each including a plurality of evaluation index values. Each parameter and each evaluation data are stored in the storage unit 105.

  (S102) The superiority / inferiority determining unit 103 determines the superiority or inferiority of each evaluation data stored in the storage unit 105. For example, when one evaluation data has a better value than other evaluation data for at least one evaluation index, it is assumed that the one evaluation data is better than the other evaluation data. Further, when one evaluation data is better than all other evaluation data, the superiority / inferiority determination unit 103 determines that the one evaluation data is excellent evaluation data. Further, the superiority / inferiority determination unit 103 determines that evaluation data other than excellent evaluation data is inferior evaluation data.

  (S103) The determination result output unit 104 presents the result of determination by the superiority / inferiority determination unit 103 to the user. For example, the determination result output unit 104 may present to the user a table in which information included in each evaluation data is associated with a determination result for each evaluation data. Further, the determination result output unit 104 may present to the user an image obtained by plotting a plurality of evaluation data on a graph so that the expression varies depending on the determination result.

  (S104) The superiority / inferiority information input unit 106 determines whether the superiority or inferiority of the evaluation data has been changed by the user. When the user changes the superiority or inferiority of the evaluation data, the process proceeds to S105. On the other hand, when the user does not change the superiority or inferiority of the evaluation data, the process proceeds to S106.

  (S105) The superiority / inferiority information input unit 106 stores information indicating superiority or inferiority of the evaluation data changed by the user in the storage unit 105. When information indicating superiority or inferiority of the evaluation data changed by the user is stored in the storage unit 105, the process proceeds to S106.

  (S106) The constraint equation generation unit 107 generates a constraint equation based on information indicating superiority or inferiority of each evaluation data stored in the storage unit 105. At this time, the constraint expression generation unit 107 generates a constraint expression using information on the evaluation function set by the evaluation function generation unit 109. Also, the mathematical expression processing unit 108 determines whether there is a set of weights that match the constraint expression by mathematical expression processing. If there is no set of weights that match the constraint expression, the evaluation function is reset, the constraint expression is regenerated, and redetermination is performed by mathematical expression processing. If there is a set of weights that matches the constraint expression, the process proceeds to S107.

  (S107) The weight calculation unit 110 calculates a set of weights conforming to the constraint equation using the constraint equation and the evaluation function information. Further, the optimization calculation unit 111 calculates an appropriate parameter using the weight set and the evaluation function calculated by the weight calculation unit 110.

  (S108) The optimization calculation unit 111 outputs an appropriate parameter and evaluation data corresponding to the parameter. For example, the optimization calculation unit 111 presents appropriate parameters and evaluation data to the user. In addition, the optimization calculation unit 111 stores appropriate parameters and evaluation data in the storage unit 105.

When the process of S108 ends, the series of processes shown in FIG. 9 ends.
FIG. 10 is a diagram illustrating a process flow of the evaluation function generation process according to the second embodiment. The process flow shown in FIG. 10 corresponds to the process of S106 shown in FIG.

  (S111) The evaluation function generation unit 109 sets a weighted sum of values of a plurality of evaluation indexes as an evaluation function. For example, the evaluation function generation unit 109 sets a weighted sum in which the order of each term is 1 for the value of each evaluation index as the evaluation function.

  (S112) The constraint equation generation unit 107 sets a constraint equation expressing the superiority or inferiority of the evaluation data using the evaluation function. For example, the constraint expression generation unit 107 sets a constraint expression that expresses a condition in which the value of the evaluation function corresponding to excellent evaluation data is smaller than the value of the evaluation function corresponding to evaluation data that is inferior.

  (S113) The mathematical expression processing unit 108 uses mathematical expression processing to determine whether there is a set of weights that conforms to all constraint expressions. For example, the mathematical expression processing unit 108 uses each weight included in each constraint expression as a binding variable and executes QE on the logical expression including the logical product of all the constraint expressions.

(S114) If there is a set of weights that matches all the constraint equations, the process proceeds to S116. On the other hand, if there is no set of weights that matches all the constraint equations, the process proceeds to S115.
(S115) The evaluation function generation unit 109 sets a new evaluation function by adding higher-order terms to the evaluation function used in the determination of S113. For example, when the original evaluation function is set as in the above equation (7), the evaluation function generation unit 109 newly sets an evaluation function as shown in the above equation (12). When the evaluation function is reset, the process proceeds to S112.

  (S116) Information on the evaluation function used in the determination in S113 and information on the constraint equation set using the evaluation function are output. For example, evaluation function information and constraint equation information are input to the weight calculation unit 110. Further, the evaluation function information and the constraint expression information are stored in the storage unit 105.

When the process of S116 ends, the series of processes shown in FIG. 10 ends.
FIG. 11 is a diagram illustrating a flow of processing for the optimization calculation according to the second embodiment. The process flow shown in FIG. 11 corresponds to the process of S107 shown in FIG. In addition, before the start of processing, information indicating the number of coefficient groups is input to the information processing apparatus 100 via the parameter input unit 101. For example, the number of coefficient groups is set by the user.

(S121) The weight calculator 110 considers each evaluation index of the evaluation function as a coefficient.
(S122) The weight calculation unit 110 generates a plurality of coefficient groups used to calculate a set of weights. For example, the weight calculation unit 110 assumes that each coefficient is a real number included in the range of −1 to 1, and generates a plurality of coefficient groups using an experimental design method, a Latin hypersquare method, a random sampling method, or the like. The weight calculation unit 110 generates the number of coefficient groups indicated by the information according to the information indicating the number of coefficient groups input via the parameter input unit 101. Note that the process of S122 may be replaced with the process of S121.

  (S123) The weight calculation unit 110 calculates, for each coefficient group generated in S122, a set of weights that minimizes the value of the evaluation function into which the value of each coefficient is substituted. For example, the weight calculation unit 110 uses a linear programming method to calculate a weight set that satisfies all constraint equations and minimizes the evaluation function. The set of weights obtained by the process of S123 is stored in the storage unit 105.

  (S124) For each coefficient group generated in S122, the weight calculation unit 110 calculates a set of weights that maximizes the value of the evaluation function into which the value of each coefficient is substituted. For example, the weight calculation unit 110 uses a linear programming method to calculate a weight set that maximizes the evaluation function that is a set of weights that conforms to all constraint equations. The set of weights obtained by the process of S124 is stored in the storage unit 105. Note that the process of S124 may be replaced with the process of S123.

  (S125) The optimization calculation unit 111 acquires a plurality of sets of weights obtained in the processes of S123 and S124, and calculates a parameter that minimizes the evaluation function for each set of weights. For example, the optimization calculation unit 111 calculates parameters using an optimization algorithm such as the Nelder-Mead method, the Powell method, the steepest descent method, or the conjugate gradient method.

(S126) The optimization calculation unit 111 outputs the parameter obtained by the process of S125 and the evaluation data corresponding to the parameter.
When the process of S126 ends, the series of processes shown in FIG. 11 ends.

FIG. 12 is a diagram showing an appropriate parameter calculation method for each of a plurality of weight sets.
When the weight set that minimizes the value of the evaluation function and the weight set that maximizes the value of the evaluation function are obtained for a certain coefficient group, the optimization calculation unit 111 repeatedly performs computer simulation while changing the parameter. The parameter that minimizes the value of the evaluation function using the tuple is calculated.

For example, as shown in FIG. 12 (A), the evaluation index # for the combination of 1 weight W _A and metrics # 2 of weight W _B, a set of weights that minimizes the evaluation function (minimum # 1, the minimum value # 2) and a set of weights that maximize the evaluation index (maximum value # 1, maximum value # 2) are obtained. However, the minimum value # 1 and the maximum value # 1 correspond to a certain coefficient group, and the minimum value # 2 and the maximum value # 2 correspond to another coefficient group. In this case, evaluation with respect to the maximum value # 1 function L _1, the evaluation function L ₃ for the minimum value # 1, evaluated against a maximum value # 2 function L _2, the evaluation function L ₄ for the minimum value # 2 Is decided.

When the evaluation function is determined as described above, the optimization calculating unit 111 uses, for example, the evaluation data D ₁₁ , D ₁₂ , and D using the evaluation function L ₃ while changing the parameters as shown in FIG. ₁₃ is sequentially calculated, and evaluation data D ₁₃ that minimizes the value of the evaluation function L ₃ is calculated. Further, the optimization calculation unit 111 sequentially calculates the evaluation data D ₂₁ , D ₂₂ , and D ₂₃ using the evaluation function L ₁ while changing the parameters, and obtains the evaluation data D ₂₃ that minimizes the value of the evaluation function L _1. calculate.

For example, the calculation of the evaluation data D ₁₁ , D ₁₂ , and D _{13 is} a calculation that plots the evaluation data while changing the parameters in a space that uses the values of the evaluation indices # 1 and # 2 and the value of the evaluation function as coordinate axes. It corresponds to. The same applies to the calculation of the evaluation data D ₂₁ , D ₂₂ , D ₂₃ .

Further, the evaluation data can be calculated in the same manner for the evaluation functions L ₂ and L ₄ while changing the parameters. And evaluation data D ₁₃ and the evaluation data D _23, the evaluation data is obtained that minimizes the value of the evaluation function can be obtained, for example, an area d of the appropriate evaluation data. In addition, the optimization calculation unit 111 can calculate an appropriate parameter range based on the evaluation data area d.

  By applying a calculation method based on single-objective optimization as described above, it is possible to calculate appropriate parameters with fewer computer simulations compared to multi-objective optimization based on genetic algorithms, etc. .

  FIG. 13 is a diagram showing a flow of processing for parameter calculation according to the second embodiment. The process flow shown in FIG. 13 corresponds to the process of S125 shown in FIG. Further, the series of processing shown in FIG. 13 is executed for each set of weights. Further, before starting the processing, information indicating the maximum number of repeated calculations is input to the information processing apparatus 100 via the parameter input unit 101. For example, the maximum number of repeated calculations is set by the user.

  (S131) The optimization calculation unit 111 acquires parameters and evaluation data. For example, the optimization calculation unit 111 acquires parameters and evaluation data stored in the storage unit 105. In addition, the optimization calculation unit 111 adds the acquired parameters and evaluation data to the calculation data list.

  (S132) The optimization calculation unit 111 calculates parameters (parameter candidates) that are highly likely to minimize the value of the evaluation function, based on the parameters and the evaluation data included in the calculation data list. For example, the optimization calculation unit 111 calculates the value of the evaluation function using each evaluation data included in the calculation data list, and calculates an approximate expression representing the relationship between the parameter and the value of the evaluation function. In addition, the optimization calculation unit 111 calculates a parameter value that minimizes the calculated approximate expression, and sets the calculated value as a parameter candidate.

  (S133) The optimization calculation unit 111 inputs a parameter candidate to the simulation execution unit 102 to execute computer simulation, and calculates evaluation data corresponding to the parameter candidate. The evaluation data calculated by the simulation execution unit 102 is stored in, for example, the storage unit 105 and read by the optimization calculation unit 111.

(S134) The optimization calculation unit 111 adds the evaluation data calculated by the simulation execution unit 102 and the parameters corresponding to the evaluation data to the calculation data list.
(S135) The optimization calculation unit 111 determines whether the number of executions of the processing of S132 to S134 has reached the maximum number of iterations. When the number of executions of the processes of S132 to S134 reaches the maximum number of repeated calculations, the process proceeds to S136. On the other hand, if the number of executions of the processes of S132 to S134 has not reached the maximum number of repeated calculations, the process proceeds to S132.

  (S136) The optimization calculation unit 111 outputs the parameter last added to the calculation data list and the evaluation data corresponding to this parameter. For example, the optimization calculation unit 111 stores parameters and evaluation data in the storage unit 105.

When the process of S136 ends, the series of processes shown in FIG. 13 ends.
By the way, the processing flow shown in FIG. 11 can be replaced with the processing flow shown in FIG.

  FIG. 14 is a diagram illustrating a flow of processing for optimization calculation according to a modification of the second embodiment. The process flow shown in FIG. 14 corresponds to the process of S107 shown in FIG. Further, before starting the process, information indicating the number of weight candidates is input to the information processing apparatus 100 via the parameter input unit 101. For example, the number of weight candidates is set by the user.

(S141) The weight calculation unit 110 calculates the minimum value of each weight under conditions that match the constraint equation.
(S142) The weight calculation unit 110 calculates the maximum value of each weight under conditions that match the constraint equation. Note that the process of S142 may be replaced with the process of S141.

  (S143) The weight calculator 110 selects each weight by the number of weight candidates within each weight range defined by the maximum value and the minimum value of each weight, and generates a plurality of weight set candidates. For example, the weight calculation unit 110 generates a plurality of sets of weights using an experimental design method, a Latin hypersquare method, a random sampling method, or the like.

(S144) The weight calculation unit 110 extracts a set of weights that match all the constraint equations from the plurality of sets of weights generated in the process of S143.
(S145) The optimization calculation unit 111 acquires a plurality of sets of weights extracted in the process of S144, and calculates a parameter that minimizes the evaluation function for each set of weights. For example, the optimization calculation unit 111 calculates parameters using an optimization algorithm such as the Nelder-Mead method, the Powell method, the steepest descent method, or the conjugate gradient method. Note that the processing in S145 is the same as the processing in S125 shown in FIG.

(S146) The optimization calculation unit 111 outputs the parameter obtained by the process of S145 and evaluation data corresponding to the parameter.
When the process of S146 ends, the series of processes shown in FIG. 14 ends.

  In addition, according to the optimization calculation shown in FIG. 11, even if the range of an appropriate weight is non-linear, an appropriate weight can be calculated. Further, according to the optimization calculation shown in FIG. 14, even if the number of weights is large, it is difficult for the calculation accuracy to decrease. Further, according to the optimization calculation shown in FIG. 14, it is possible to reduce the influence of the instruction from the user on the calculation result.

  The second embodiment has been described above. According to the second embodiment, the weight set is calculated when there is a weight set that matches the constraint equation, and the evaluation function is changed when there is no weight set that matches the constraint equation. The For this reason, useless calculation such as continuing to search for a weight set when there is no weight set that matches the constraint equation is not performed, and it is possible to efficiently calculate a weight set that matches the constraint equation.

  In addition, since the user's change operation regarding superiority or inferiority of the evaluation data is accepted and the weight set is calculated using the constraint equation based on the superiority or inferiority after the change, the weight set reflecting the user's intention is obtained. Also, a plurality of weight sets are calculated using mathematical processing and linear programming, and a parameter value that minimizes the value of the evaluation function is calculated using the plurality of weight sets. Therefore, it is possible to reduce the number of computer simulations that are required to calculate appropriate parameters. As a result, it is possible to keep the calculation load required for parameter optimization low.

  By the way, in the above description, it is assumed that the information processing apparatus 100 executes all the processes including the computer simulation, but the computer simulation is executed by another information processing apparatus different from the information processing apparatus 100. Also good. In this case, the information processing apparatus 100 calculates an appropriate parameter using a result of computer simulation input from another information processing apparatus. Such a modification is also included in the second embodiment.

The following modifications are also included in the second embodiment.
For example, if there are too many constraint expressions, a set of weights that match all the constraint expressions may not be obtained, but if the number of constraint expressions is too small, sufficient calculation accuracy may not be obtained. . In view of this, a method for adjusting the number of evaluation data used for generating constraint expressions so as to obtain an appropriate number of constraint expressions is proposed as a modified example.

  Further, when it is determined whether or not there is a set of weights that match all the constraint expressions using mathematical expression processing, it may be assumed that the mathematical expression processing operation does not end within an appropriate time. Therefore, if the calculation of the mathematical expression processing does not end within a certain time, the method of stopping the mathematical expression processing, reducing the number of constraint expressions subject to the calculation, and re-executing the mathematical expression calculation is a variation. Propose as.

  In the above description, when there is no set of weights that match all the constraint equations, a method of increasing the order of the evaluation function and regenerating the constraint equation has been proposed. It is also conceivable to estimate a constraint equation that is an impediment factor and to exclude the constraint equation. According to this method, when there are too many constraint expressions and a set of weights that match all the constraint expressions cannot be obtained, it is possible to efficiently remove factors that make it impossible to obtain a set of evaluation functions. Become. In addition, it is easy to obtain a set of weights that match all constraint equations without increasing the order of the evaluation function.

  In the above description, when calculating the appropriate parameters, the process of calculating the evaluation data by computer simulation and the process of calculating the value of the evaluation function using the evaluation data and the set of weights are repeatedly executed. A method to determine the correct parameters was proposed. However, instead of this method, for example, a single-purpose evaluation function indicating the relationship between the value of the evaluation index and the parameter is calculated for each evaluation index, and a function obtained by weighting a plurality of single-purpose evaluation functions is obtained. A method may be employed in which an appropriate parameter is determined by executing mathematical expression processing on the subject. According to this method, the number of executions of computer simulation can be reduced.

  In addition, when calculating an appropriate parameter, a method using a combination of the above-described method for determining an appropriate parameter based on repeated execution of computer simulation and the above-described method for determining an appropriate parameter based on mathematical expression processing may be considered. For example, a method for executing an appropriate parameter determination method based on repeated execution of a computer simulation when an appropriate parameter determination method based on mathematical expression processing is executed and an appropriate parameter cannot be obtained by this method is considered. It is done. According to this method, the number of executions of computer simulation can be reduced as compared with a case where only an appropriate parameter determination method based on repeated execution of computer simulation is used.

[Third Embodiment]
Next, a third embodiment will be described.
FIG. 15 is a diagram illustrating an example of an information processing device according to the third embodiment. As illustrated in FIG. 15, the information processing apparatus 200 includes a parameter input unit 201, a simulation execution unit 202, an dominance determination unit 203, a determination result output unit 204, a storage unit 205, and an dominance information input unit 206. The information processing apparatus 200 includes a constraint expression generation unit 207, a mathematical expression processing unit 208, an evaluation function generation unit 209, a weight calculation unit 210, an optimization calculation unit 211, and a superiority / inferiority level calculation unit 212.

  The information processing apparatus 200 can be realized using the same hardware as the information processing apparatus 100 according to the second embodiment. Further, the parameter input unit 201, the simulation execution unit 202, the superiority / inferiority determination unit 203, the determination result output unit 204, and the superiority / inferiority information input unit 206 can be realized as modules of a program executed by the CPU 901. Further, some or all of the functions of the parameter input unit 201, the simulation execution unit 202, the superiority / inferiority determination unit 203, the determination result output unit 204, and the superiority / inferiority information input unit 206 can be realized as an electronic circuit instead of software. .

  Further, the constraint expression generation unit 207, the mathematical expression processing unit 208, the evaluation function generation unit 209, the weight calculation unit 210, the optimization calculation unit 211, and the superiority / inferiority level calculation unit 212 can be realized as modules of programs executed by the CPU 901. In addition, some or all of the functions of the constraint expression generation unit 207, the mathematical expression processing unit 208, the evaluation function generation unit 209, the weight calculation unit 210, the optimization calculation unit 211, and the superiority level calculation unit 212 are not software, but an electronic circuit It is also possible to realize. The storage unit 205 is a storage area secured in the RAM 902 or the HDD 903.

  However, the parameter input unit 201, the simulation execution unit 202, the superiority / inferiority determination unit 203, and the determination result output unit 204 are the parameter input unit 101, the simulation execution unit 102, the superiority / inferiority determination unit 103 according to the second embodiment, and The substantially same process as the determination result output unit 104 is executed. Also, the superiority / inferiority information input unit 206, the evaluation function generation unit 209, the weight calculation unit 210, and the optimization calculation unit 211 are respectively the superiority / inferiority information input unit 106, the evaluation function generation unit 109, and the weight calculation according to the second embodiment. Unit 110 and optimization calculation unit 111 execute substantially the same processing.

  Therefore, description of substantially the same items as those of the second embodiment will be omitted, and description will be made focusing on items that are different from those of the second embodiment. Hereinafter, the constraint expression generation unit 207, the mathematical expression processing unit 208, and the superiority / inferiority level calculation unit 212 will be mainly described.

  The superiority / inferiority level calculation unit 212 calculates the degree to which each evaluation data is superior to other evaluation data based on the relationship between at least two evaluation data. For example, when there are m other evaluation data better than a certain evaluation data, the superiority / inferiority level calculation unit 212 sets the superiority / inferiority level for the one evaluation data to m. The superiority / inferiority level calculation unit 212 calculates the superiority / inferiority level of each evaluation data for all inferior evaluation data. In addition, the superiority / inferiority level calculation unit 212 stores the calculated superiority / inferiority level of each evaluation data in the storage unit 205.

When the above superiority / inferiority level is used, the degree of superiority or inferiority for each evaluation data can be handled quantitatively. Moreover, it becomes possible to classify the evaluation data based on the superiority / inferiority level, such as an evaluation data group with superiority level m _1, an evaluation data group with superiority level m _2, an evaluation data group with superiority level m ₁ or higher. . Further, when the evaluation data is classified based on the superiority / inferiority level, a constraint expression can be generated for each classified evaluation data group.

For example, the constraint expression generation unit 207 generates a constraint expression that expresses the relationship between each evaluation data included in the evaluation data group having the superiority / inferiority level m ₀ and each excellent evaluation data. However, m ₀ is the initial value of the maximum superiority level or the superiority level set by the user in advance.

  The mathematical expression processing unit 208 determines whether or not there is a set of weights that matches all the constraint expressions generated by the constraint expression generation unit 207 by mathematical expression processing. The result of determination by the mathematical expression processing unit 208 is notified to the constraint expression generation unit 207 and the evaluation function generation unit 209. When there is no set of weights that match all the constraint expressions, the constraint expression generation unit 207 causes the evaluation function generation unit 209 to generate a new evaluation function. For example, the evaluation function generation unit 209 generates a new evaluation function having a higher order than the original evaluation function, and inputs the new evaluation function to the constraint expression generation unit 207.

The constraint expression generation unit 207 to which a new evaluation function has been input generates a constraint expression that expresses the relationship between each evaluation data included in the evaluation data group having the superiority level m ₀ and each excellent evaluation data. In addition, the mathematical expression processing unit 208 determines whether there is a set of weights that matches all the constraint expressions generated by the constraint expression generation unit 207 by mathematical expression processing. In this way, while changing the evaluation function until there is a set of weights that match all the constraint expressions, the update process of the constraint expressions and the determination process that determines the presence or absence of a weight set that matches all the constraint expressions Is repeatedly executed.

On the other hand, when there is a set of weights that matches all the constraint equations, the constraint equation generation unit 207 changes the combination of constraint equations used for weight calculation. For example, the constraint expression generation unit 207 adds a constraint expression that expresses the relationship between each evaluation data included in the evaluation data group with the superiority level m ₁ and each excellent evaluation data. However, m ₁ is a value smaller than m ₀ . For example, assume that m ₁ is m ₀ −1. When the combination of constraint expressions is changed, the mathematical expression processing unit 208 determines again whether or not there is a weight set that matches all the constraint expressions generated by the constraint expression generation unit 207 by mathematical expression processing.

When there is a set of weights that matches all the constraint equations, the constraint equation generation unit 207 changes the combination of constraint equations used for weight calculation. For example, the constraint expression generation unit 207 adds a constraint expression that expresses the relationship between each evaluation data included in the evaluation data group with the superiority level m ₂ and each excellent evaluation data. However, m ₂ is assumed to be a value smaller than m _1. For example, m ₂ is assumed to be m ₁ −1. When the combination of constraint expressions is changed, the mathematical expression processing unit 208 determines again whether or not there is a weight set that matches all the constraint expressions generated by the constraint expression generation unit 207 by mathematical expression processing.

  The constraint expression generation unit 207 and the formula processing unit 208 repeatedly execute the addition of the constraint formula and the determination by the formula processing as described above until it is determined that there is no weight set that matches all the constraint formulas. When it is determined that there is no weight set that matches all the constraint expressions, the constraint expression generation unit 207 determines the constraint expressions and evaluation functions at the time when it is determined that there exists a weight set that matches all the constraint expressions. Information is input to the weight calculator 210. Further, the constraint expression generation unit 207 stores the constraint expression and the evaluation function information in the storage unit 205.

  By the above iterative process, a combination of constraint expressions and an evaluation function from which a set of weights is obtained are determined. As described above, the constraint equation generation unit 207 according to the third embodiment does not target all inferior evaluation data from the beginning when generating a constraint equation, and weights that conform to the constraint equation A constraint equation is generated by selecting an evaluation data group that can easily obtain a set of. According to this method, the order of the evaluation function can be kept low as compared with the constraint equation generation method according to the second embodiment.

  Heretofore, an example of the information processing apparatus 200 according to the third embodiment has been described. According to the information processing apparatus 200, a weight set is calculated when there is a weight set that conforms to the constraint equation, and the constraint equation is changed when there is no weight set that conforms to the constraint equation. For this reason, useless calculation such as continuing to search for a weight set when there is no weight set that matches the constraint equation is not performed, and it is possible to efficiently calculate a weight set that matches the constraint equation. In addition, a set of weights reflecting the user's intention is calculated. In addition, since the order of the evaluation function can be kept low, it is possible to reduce the load of mathematical expression processing, weight calculation, and optimization calculation.

  Also, according to the information processing apparatus 200, a plurality of weight sets are calculated using mathematical formula processing and linear programming, and parameter values that minimize the value of the evaluation function using the plurality of weight sets are obtained. Calculated. When calculating multiple sets of weights, computer simulation is performed to obtain multiple evaluation data, but the calculation of the parameter that minimizes the value of the evaluation function is nothing but single-objective optimization. The number of computer simulation executions can be kept low compared to the case of computerization. As a result, it is possible to keep the calculation load required for parameter optimization low.

  Hereinafter, considering the two evaluation indices (evaluation index # 1 and evaluation index # 2) that are in a trade-off relationship with each other, the processing by the information processing apparatus 200 will be further described. Note that the smaller the value of the evaluation index, the better.

FIG. 16 is a diagram for explaining a superior / inferior quantification method according to the third embodiment. In the example of FIG. 16 (A), the other evaluation data (solid circles) better than the evaluation data D ₁₁ is present one. Therefore, the superiority / inferiority level of the evaluation data D ₁₁ is 1. Further, in the example of FIG. 16 (B), the other evaluation data (solid circles) better than the evaluation data D ₁₂ is present five. Therefore, superiority levels of evaluation data D ₁₂ is 5.

As shown in FIG. 16, by using the superiority / inferiority level, the degree of superiority or inferiority of each evaluation data can be quantified. Moreover, it becomes possible to classify the evaluation data based on the superiority / inferiority level, such as an evaluation data group with superiority level m _1, an evaluation data group with superiority level m _2, an evaluation data group with superiority level m ₁ or higher. . Further, when the evaluation data is classified based on the superiority / inferiority level, a constraint expression can be generated for each classified evaluation data group.

Hereinafter, the flow of processing by the information processing apparatus 200 will be described.
FIG. 17 is a diagram illustrating a flow of processing performed by the information processing apparatus according to the third embodiment. Note that a plurality of parameters are input to the information processing apparatus 200 via the parameter input unit 201 before the processing is started. The number of input parameters and the numerical range of each parameter are set by the user. Each parameter is generated using, for example, an experimental design method, a Latin hypersquare method, or a random sampling method.

  (S201) The simulation execution unit 202 executes a computer simulation for each parameter, and calculates a plurality of evaluation data each including a plurality of evaluation index values. Each parameter and each evaluation data are stored in the storage unit 205.

  (S202) The superiority / inferiority determination unit 203 determines the superiority or inferiority of each evaluation data stored in the storage unit 205. For example, when one evaluation data has a better value than other evaluation data for at least one evaluation index, it is assumed that the one evaluation data is better than the other evaluation data. Moreover, when one certain evaluation data is better than all the other evaluation data, the superiority / inferiority determination unit 203 determines that the one evaluation data is excellent evaluation data. Moreover, the superiority / inferiority determination unit 203 determines that evaluation data other than excellent evaluation data is inferior evaluation data.

  (S203) The determination result output unit 204 presents the result of determination by the superiority / inferiority determination unit 203 to the user. For example, the determination result output unit 204 may present to the user a table in which information included in each evaluation data is associated with a determination result for each evaluation data. Further, the determination result output unit 204 may present to the user an image obtained by plotting a plurality of evaluation data on a graph so that the expression varies depending on the determination result.

  (S204) The superiority / inferiority information input unit 206 determines whether the superiority or inferiority of the evaluation data has been changed by the user. If the user changes the superiority or inferiority of the evaluation data, the process proceeds to S205. On the other hand, when the user does not change the superiority or inferiority of the evaluation data, the process proceeds to S206.

  (S205) The superiority / inferiority information input unit 206 stores information indicating superiority or inferiority of the evaluation data changed by the user in the storage unit 205. When information indicating superiority or inferiority of the evaluation data changed by the user is stored in the storage unit 205, the process proceeds to S206.

  (S206) The superiority / inferiority level calculation unit 212 calculates the degree to which each evaluation data is superior to other evaluation data based on the relationship between at least two evaluation data. For example, when there are m other evaluation data better than a certain evaluation data, the superiority / inferiority level calculation unit 212 sets the superiority / inferiority level for the one evaluation data to m. The superiority / inferiority level calculation unit 212 calculates the superiority / inferiority level of each evaluation data for all inferior evaluation data. In addition, the superiority / inferiority level calculation unit 212 stores the calculated superiority / inferiority level of each evaluation data in the storage unit 205.

  (S207) The constraint expression generation unit 207 generates a constraint expression based on information indicating superiority or inferiority of each evaluation data stored in the storage unit 205 and information indicating superiority or inferiority level. At this time, the constraint expression generation unit 207 generates a constraint expression using information on the evaluation function set by the evaluation function generation unit 209. Also, the mathematical expression processing unit 208 determines whether or not there is a set of weights that match the constraint expression by mathematical expression processing. If there is no set of weights that match the constraint expression, the evaluation function is reset, the constraint expression is regenerated, and redetermination is performed by mathematical expression processing. If there is a set of weights that matches the constraint equation, the process proceeds to S208.

  (S208) The weight calculation unit 210 calculates a set of weights that match the constraint equation using the constraint equation and evaluation function information. Further, the optimization calculation unit 211 calculates appropriate parameters using the weight set and evaluation function calculated by the weight calculation unit 210.

  (S209) The optimization calculation unit 211 outputs an appropriate parameter and evaluation data corresponding to the parameter. For example, the optimization calculation unit 211 presents appropriate parameters and evaluation data to the user. In addition, the optimization calculation unit 211 stores appropriate parameters and evaluation data in the storage unit 205.

When the process of S209 ends, the series of processes illustrated in FIG. 17 ends.
FIG. 18 is a diagram illustrating a flow of processing for the calculation processing of the constraint equation according to the third embodiment. The process flow shown in FIG. 18 corresponds to the process of S207 shown in FIG.

(S211) The constraint equation generation unit 207 sets the maximum value of the superiority / inferiority level to N.
(S212) The constraint expression generation unit 207 generates a constraint expression that expresses the relationship between each evaluation data included in the evaluation data group having the superiority / inferiority level N and excellent evaluation data.

  (S213) The mathematical formula processing unit 208 determines whether there is a set of weights that matches all the constraint formulas generated by the constraint formula generation unit 207 in the processing of S212. If there is a set of weights that matches the constraint equation, the process proceeds to S215. On the other hand, if there is no set of weights matching the constraint equation, the process proceeds to S214.

  (S214) The evaluation function generation unit 209 generates a new evaluation function having a higher order than the original evaluation function. In addition, the evaluation function generation unit 209 inputs new evaluation function information to the constraint expression generation unit 207.

(S215) The constraint equation generation unit 207 decrements N by one.
(S216) The constraint expression generation unit 207 generates a constraint expression that expresses a relationship between each evaluation data included in the evaluation data group having the superiority / inferiority level of N or more and excellent evaluation data.

  (S217) The mathematical expression processing unit 208 determines whether or not there is a set of weights that matches all the constraint expressions generated by the constraint expression generation unit 207 in the process of S216. If there is a set of weights that matches the constraint equation, the process proceeds to S215. On the other hand, if there is no set of weights matching the constraint equation, the process proceeds to S218.

  (S218) The constraint expression generation unit 207 outputs a constraint expression when it is finally determined that there is a set of weights that matches all the constraint expressions. For example, the constraint expression generation unit 207 stores the set constraint expression information in the storage unit 205. In addition, the constraint equation generation unit 207 inputs information on the set constraint equation to the weight calculation unit 210.

When the process of S218 ends, the series of processes shown in FIG. 18 ends.
The third embodiment has been described above. According to the third embodiment, the weight set is calculated when a set of weights that match the constraint formula exists, and the constraint formula is changed when there is no set of weights that match the constraint formula. The For this reason, useless calculation such as continuing to search for a weight set when there is no weight set that matches the constraint equation is not performed, and it is possible to efficiently calculate a weight set that matches the constraint equation. In addition, a set of weights reflecting the user's intention is calculated.

  In addition, since the order of the evaluation function can be kept low, it is possible to reduce the load of mathematical expression processing, weight calculation, and optimization calculation. Further, when calculating a plurality of sets of weights, a computer simulation is executed to obtain a plurality of evaluation data, but the calculation of a parameter that minimizes the value of the evaluation function is nothing other than single-objective optimization, Compared with multi-objective optimization, the number of computer simulations can be reduced. As a result, it is possible to keep the calculation load required for parameter optimization low.

DESCRIPTION OF SYMBOLS 10 Information processing apparatus 11 Memory | storage part 12 Calculation part 13A, 13C Data sample 14A, 14C Evaluation index value 15 Evaluation function 16A, 16C Function value 17 Superiority information 100, 200 Information processing apparatus 101, 201 Parameter input part 102, 202 Simulation execution Unit 103, 203 superiority / inferiority determination unit 104,204 determination result output unit 105,205 storage unit 106,206 superiority / inferiority information input unit 107,207 constraint expression generation unit 108,208 mathematical expression processing unit 109,209 evaluation function generation unit 110,210 weight Calculation unit 111, 211 Optimization calculation unit 212 Superiority level calculation unit

Claims (8)

An evaluation support method executed by a computer,
Retrieve multiple data samples, each containing two or more metrics values,
For at least some of the plurality of data samples, accepting input of superiority / inferiority information indicating relative superiority or inferiority among the plurality of data samples;
When at least one evaluation function among a plurality of evaluation functions each including a variable and a weight corresponding to the two or more evaluation indexes is substituted with the value of the two or more evaluation indexes of each data sample It is determined whether the weight can be set so that a function value matching the superiority or inferiority relationship between the plurality of data samples according to the superiority / inferiority information is calculated, and the weight can be set from the plurality of evaluation functions An evaluation support method for selecting a judged evaluation function. The evaluation support method according to claim 1, wherein in the selection, when it is determined that the weight cannot be set for one evaluation function, the determination is performed for another evaluation function having a higher order than the one evaluation function. By calculating the initial value of superiority or inferiority of each data sample by comparing the values of the two or more evaluation indices between the plurality of data samples, and outputting the calculated initial value of superiority or inferiority,
The evaluation support method according to claim 1, wherein the input of the superiority / inferiority information is an input for correcting the initial value of superiority / inferiority for at least a part of the plurality of data samples. In the determination, by substituting the values of the two or more evaluation indexes of each data sample into the variable, a plurality of expressions including the weight corresponding to the plurality of data samples are calculated, and the plurality of expressions are used. 4. The evaluation support according to claim 1, wherein a constraint equation expressing a superiority or inferiority relationship between the plurality of data samples is calculated, and whether the weight has a solution in the constraint equation is determined. Method. The evaluation support method according to claim 4, wherein the weight included in the selected evaluation function is calculated using the constraint equation used when determining whether the weight can be set for the selected evaluation function. By comparing the values of the two or more evaluation indices between the plurality of data samples, the plurality of data samples are classified into three or more superiority levels,
The evaluation support method according to any one of claims 1 to 5, wherein in the determination, it is determined whether the weight can be set by excluding some data samples from the plurality of data samples according to the superiority / inferiority level. . A storage unit for storing a plurality of data samples each including two or more evaluation index values;
Accepting input of superiority / inferiority information indicating relative superiority or inferiority among the plurality of data samples for at least some of the plurality of data samples, each including a variable and a weight corresponding to the two or more evaluation indices For at least one evaluation function among the evaluation functions, when the values of the two or more evaluation indices of each data sample are substituted into the variable, the superiority / inferiority relationship between the plurality of data samples according to the superiority / inferiority information is met. An information processing apparatus comprising: an arithmetic unit that determines whether the weight can be set so that a function value to be calculated is calculated, and selects an evaluation function that is determined to be able to set the weight from the plurality of evaluation functions. On the computer,
Retrieve multiple data samples, each containing two or more metrics values,
For at least some of the plurality of data samples, accepting input of superiority / inferiority information indicating relative superiority or inferiority among the plurality of data samples;
When at least one evaluation function among a plurality of evaluation functions each including a variable and a weight corresponding to the two or more evaluation indexes is substituted with the value of the two or more evaluation indexes of each data sample It is determined whether the weight can be set so that a function value matching the superiority or inferiority relationship between the plurality of data samples according to the superiority / inferiority information is calculated, and the weight can be set from the plurality of evaluation functions A program that executes the process of selecting the evaluated evaluation function.

Images