JPH0668066A - Decision condition generating method - Google Patents

Abstract

PURPOSE:To use a small-sized computer by applying the version space method to the field to easily generate and correct a decision condition. CONSTITUTION:After a storage device 100 is initialized by a processor 200, the combination of attribute values of a positive instance in instance data 300 is set to the storage device, and the number of positive instances is set to '1', and combinations of attribute values of negative instances are set in the storage device. One is added to the number of negative instances of a combination of attribute values if this combination of attribute values exists in combinations of attribute values set as negative instances, and the number of negative instances is set. 'necessary' is set to the necessary/unnecessary column corresponding to the combination having the smallest number of attributes out of combinations of attributes whose number of negative instances is '0', and 'unnecessary' is set to the necessary/unnecessary column corresponding to the combination of all attribute values including combinations of attribute values, which are set to 'necessary' by the processing, out of combinations of attribute values whose number of negative instances is '0'. This operation is repeated to generate the combination of attribute values, which is included in positive instances but is not included in negative instances, as the decision condition.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a discriminating condition generating method. In detail, in some equipment, when some observation data are given, the success or failure of those attributes (whether the observed numerical data is in a specific range, or the discrete data belongs to a specific set). Whether or not) is suitable for a certain purpose of the facility (positive example) and whether or not (negative example) is determined by an attribute value and a determination result based on this attribute value. The present invention relates to a method of generating based on case data and contents of a storage device in which a combination of attribute values and the like are set, for example, a discrimination condition generating method applied to a failure diagnosis system of a plant or the like.

[0002]

2. Description of the Related Art Conventionally, this type of determination condition is incorporated in various types of equipment and receives status signals from various parts of the equipment, and is used to determine the presence or absence of equipment failure based on these status signals. Has been done. Here, even if the equipment is the same, the way in which the failure occurs differs depending on the installation environment and the operating method, so even when setting the conditions for determining failure diagnosis, there is a uniform The judgment condition cannot be set. Therefore, ideally, the cause of failure understood by the experts who maintain the equipment,
It is desirable to set the judgment conditions based on the appearance of phenomena at the time of failure individually for each equipment and incorporate it into the equipment.

Conventionally, in order to create a discrimination condition based on such expert know-how, data to be noticed at the time of discrimination (state signals from each part of the equipment, etc.) are determined, and specific numerical values of these data are used. Experts were required to determine the success or failure of the attributes of whether or not a certain condition has been met, and to sort out the correlation between the appearance of the success or failure of these attributes and the cause of failure in all cases. . For this reason, there are cases in which an erroneous determination condition is created due to the misunderstanding of an expert, and due to insufficient consideration, there are excesses and deficiencies in the types of attributes and failure cases.

Further, in order to verify the discriminant condition thus created, data is input and the discriminant result by the discriminant condition is compared with the expert's knowledge, and if there is an error, the preset attribute or discriminant condition is corrected. It was necessary to do. In this case, since the expert individually corrects the attribute and the judgment condition for the faulty failure case, there is a situation in which the correction work is not completed forever without noticing the mutual relation of the attribute and the judgment condition.

In order to solve these problems, some methods have conventionally been proposed for automating the correction of discrimination conditions (for example, Ginsberg, A., Weiss, S. and Politaki.
s, P., "SEEK 2: A Generalized Approach to Automatic
Knowledge Base Refinement "Proc. Of IJCAI-85, p
p. 367-374, 1985). These methods correct the discriminant condition that seems to be most effective after examining the mutual relations of all the discriminant conditions in the fault case having an error. However, in order for these methods to operate efficiently, it is necessary that the target discriminating conditions are well organized in advance, and they cannot be used to correct the conditions for establishing attributes. For this reason, the burden on the expert who first sorts out the attributes and the discrimination conditions remains large.

Therefore, in order to reduce this burden, some methods have been proposed in the past for automatically generating a discrimination condition based on a case (for example, the version space method, Mich.
ell, TM "Version Spaces: A Candidate Elimination
Approach to Rule Learning "Proc. Of 5th IJCAI, p
p. 305-310, 1977). These methods input a case composed of success or failure of some attributes and a judgment result by an expert, and generate a judgment condition for judging the cases without error. Note that these methods are called "learning from examples".

[0007]

When creating and refining a knowledge base using learning from examples, we provide the following three elements. System of attribute values that describe the condition part of the discriminant condition Set of target concepts that describe the execution part of the discriminant condition Case data consisting of positive and negative examples of the target concept Note that in the following, "attributes and their possible attributes The term “background knowledge” is used as appropriate to mean “system of values”.

However, the field where the method such as learning from the example is applied is generally in the following situation. (1) Background knowledge is not always organized on-site. For example, in the manpower allocation work, an expression such as "the load of this month is high" is used. However, the definition of the attribute value that the load is “high” is not always constant. That is, a person is said to be “high” when the load is “70% or more” with respect to the capacity upper limit of the person in charge.
Another person is "high" when "90% or more". This obviously causes confusion and reduces the work efficiency of knowledge base creation.

(2) The quality of case data collected from the site is not always good. In the case data, some attribute values and the goal concept inferred by the expert at that time are recorded. There is no guarantee that the expert will always give correct inference results. Also, not all required attribute values are recorded without error. Case data is a concrete example of a goal concept to be learned, and if this case data contains mistakes or things that are easy to understand as exceptions, the efficiency and reliability of knowledge base creation will decrease. To do.

(3) Non-gradual knowledge base changes are not welcome. The expert changes the definition of the attribute value or excludes the case data during the refinement process. At this time, generally, the form of the discriminant conditions created changes. This large change in shape makes it difficult for professionals to understand it. This is unfavorable for the cooperation of experts.

Because of the situation in the field, it is difficult to apply the learning from the examples that are based on the complete background knowledge and the case data without error. The present invention has been made to solve the above problems, and has the following objects. (1) Supporting the setting and adjustment of background knowledge by experts Background knowledge is important for ensuring the consistency of the entire knowledge base. This requirement is necessary to facilitate the incorporation of background knowledge of experts and to clearly define the meaning of attribute values. It is also necessary to allow an expert to control the learning process and to easily understand the discrimination condition obtained as a result of learning.

(2) Being able to detect peculiar case data Case data is important for creating and verifying discrimination conditions. This requirement is necessary for facilitating detection of erroneous or exceptional data from the case data and improving learning efficiency. (3) Minimize changes in discriminant conditions before and after refinement This requirement is necessary to make the process of refinement easy for experts to understand. It is also necessary so that the discriminant conditions can be gradually refined.

More specifically, according to the present invention, learning from an example (in particular, version space method) is applied in the field to easily generate a discrimination condition, and the discrimination processing is performed using the discrimination condition. The object of the present invention is to provide a discrimination condition generation method capable of immediately correcting the discrimination condition when there is an error in discrimination.

[0014]

In order to achieve the above object, the first aspect of the present invention has a plurality of cases each including attribute values of a plurality of types of attributes and a determination result for these attribute values, and at least one Corresponding to the combination of case data in which the discrimination result of two cases is positive, the combination of the attribute values in which the same attribute is not duplicated, and the combination of these attribute values, the discrimination result is positive (hereinafter, The number of positive examples as a value to determine whether it is itself or a combination of the attributes of the positive examples and the number of negative examples as the number of negative examples (hereinafter referred to as negative examples). And a storage device in which the necessity of necessity of the combination of the attribute value is set, respectively, and based on the case data, initialization of the storage device, setting of the number of positive cases and negative cases, and setting of necessity With the processing equipment The processing device sets the combination and necessity of attribute values in the storage device to “blank”, sets the positive example number and the negative example number to “0”, and initializes the storage device. All combinations of a plurality of attribute values in the positive example are set in the storage device, the number of positive cases corresponding to the positive example in the case data is set to “1”, and the plurality of attribute values in the negative example in the case data are set. The number of negative examples is set by setting combinations in the storage device, and for each combination of attribute values in the storage device, if each combination exists in the combination of attribute values set as a negative example, the number of negative examples is integrated. Is set, and "necessity" is set to the necessity or non-necessity corresponding to the combination having the smallest number of attributes among the combinations of the attribute values having the negative number of "0" in the storage device, Of attribute value combinations that are "0" Chi, "No" in necessity for all combinations of attribute values including a combination of attribute values necessity is set to "required" by the process
Is set, a combination of attribute values included in the positive example and not included in the negative example is generated as the determination condition for the positive example and the negative example.

According to a second aspect of the present invention, a storage device generated by the discriminating condition generating method of the first aspect of the invention, an input device to which an attribute value is input, an output device to which an inference result is output, and an input device is input. A processing device that compares the attribute value with the content of the storage device to perform inference, and corrects the content of the storage device according to the evaluation of the inference result, and the processing device stores the attribute value input from the input device. It is compared with a combination of attribute values whose necessity in the device is "necessary", and "positive" if the attribute values input from the input device include at least this combination of attribute values, and "negative" if not included. When the above inference result is evaluated to be correct, no processing is performed, and the above inference result is evaluated to be incorrect. When is input The attribute value stored in the input device is set in the attribute value combination field added in the storage device, and the number of positive cases is set to "0", the number of negative cases is set to "1", and "blank" is set for necessity. The number of negative examples corresponding to those included in the content of the newly added attribute value combination among the attribute value combinations previously set in the storage device is increased by "1",
If the corresponding necessity is "necessary", this is set to "blank", and then the contents of the attribute value combination whose necessity is "not required" are searched for in the storage device and the necessity is "necessary". If it does not include the contents of any other combination of attribute values, the necessity is “No”.
To "necessary", the inference result is evaluated as incorrect, and another conclusion "correct" is input from the input device, the attribute value combination field added in the storage device is added to the input device. The stored attribute value is set, and the number of positive examples is set to "1", the number of negative examples is set to "0", and the necessity is set to "necessary", and the contents of the storage device are corrected in this way. According to this, the discrimination conditions of the positive example and the negative example are generated.

[0016]

The present invention has the following actions. A. Support adjustment of background knowledge. (1) Detection of exceptional case data Since there are a small number of negative cases in the version space method,
Multiple nodes may occur that have common attribute value combinations but are not grouped as more generalized nodes. As a result, the created discrimination conditions are subdivided. Therefore, in the present invention, the most common nodes in the version space are examined, and if there are nodes that satisfy the following conditions, they are detected. -The number of included attributes is the same. • All but one attribute value is the same.

Next, a negative example in which these nodes are prevented from being collected as a more generalized node is detected.
Then, the experts are presented with the nodes that may be generalized and the attributes that they have in common. In addition, case data that impedes generalization is presented to experts as possible exceptions.

(2) Detection of unknown case data In order to evaluate the reliability of the learning result, the following method is generally adopted. That is, the case data is divided into learning data and test data, and first, the discrimination conditions are created using the learning data. Next, for each of the test case data, the attribute value is given to the discrimination condition, and the inference result is compared with the target concept in the case data. Then, the ratio of the number of case data items in which both match to the whole is taken as the evaluation value of the reliability of the discrimination condition.

In order to perform the above evaluation directly in the version space without performing inference, the test case data are associated with the nodes in the version space. If there are positive examples that cannot be associated with the current version space nodes, they are detected and presented to the expert along with the goal concept. Also, if there are negative examples associated with the nodes in the current version space, they are detected and presented to the expert together with the goal concept. At this time, the attribute values commonly included in the negative examples are also detected and presented together.

Each of the above-mentioned extended functions presents to the expert case data associated with a specific node in the version space. The expert attempts to remove the pointed case data from the currently associated nodes. For that purpose, there are the following two countermeasures. Remove the pointed out case data itself. The expert reviews the content of the case data pointed out and, if found to be incorrect or exceptional, corrects or deletes it.

The definition of background knowledge is changed to move the pointed out case data to another node. The expert reviews the definition of the indicated attribute value and changes the definition in order to realize the following movement. Move the minority case data pointed out by the method of the present invention to another node. Move the negative examples pointed out by the method of the invention out of the version space. Move the positive example pointed out by the method of the present invention into the version space. The method of the present invention presents a combination of attributes that the nodes of interest have in common. Some of these attributes are subject to definition changes.

B. Use on-site case data. (1) Permitting continuous attribute values The version space method accepts only discrete attribute values. Therefore, we divided the range of continuous attribute values into some discrete intervals so that we could include them in the background knowledge as new attribute values, and we were able to handle continuous attribute values as well. When a continuous attribute value is given, it is converted into a discrete attribute value corresponding to the section including the value.

(2) Acceptance of Missing Values If the attribute of the case data has a missing value, the version space method cannot be learned. Therefore, if there is a missing value, interpret that attribute as any value.
Values are not required to be set for all attributes of case data.

(3) Correspondence to a small number of case data If the nodes in the version space are deleted based on a sufficient amount of case data, and finally converge to one node, the version space method does learning. I can't finish. Therefore,
When the version space does not converge to one node even if all the input case data are used, the most general node is output as the learning result. Then, if a negative example was obtained during the operation of the knowledge base, the excessive generalization was corrected.

(4) Detection of erroneous case data If there are case data containing different target concepts with the same combination of attribute values, the version space method fails to learn. Therefore, before creating the version space, if there is inconsistent case data, it is detected and presented to the expert together with the goal concept. For example, consider the case where there are m positive examples and n negative examples with the same combination of attribute values, and m <n. At this time, the case data that is a positive example is presented to the expert as a possibility of an error. That is,
When a state in which a positive example and a negative example are simultaneously associated with one node in the version space is detected, the minor case data is presented to the expert.

(5) Automatic collection of case data Generally, it takes time and effort to collect case data. This is especially true when case data is created by calculating from certain data. Therefore, while performing the discrimination process based on the discriminant that was created once, an expert evaluates the discrimination result, and the result is automatically stored as case data.

(6) Selection of Usage Attribute Although some attributes included in the case data may not be directly necessary for the discrimination processing, it takes a lot of time to recreate the case data except them. Therefore, we made it possible to select only the required attributes without recreating the case data.

C. Used by a small computer. Since the conventional determination condition generation method is a batch processing method,
The processing time and the computer resources used cannot be controlled externally. For this reason, there are cases in which a small computer cannot be used to generate the determination condition within a certain period of time. Since the version space includes, as nodes, all of the combinations of attribute values that do not conflict with the case data, a large capacity is generally required. Therefore, from the viewpoint of using a small computer, the version space is created from one positive example. Then, when there is a positive example that is not included in the space, a new version space is created and only the difference from the original space is retained.

D. Efficiently perform discrimination processing on site. (1) Converting discriminant conditions into rules In order to perform discrimination processing at high speed, the discriminant conditions obtained in the form of version space are converted into the form of rules for use. (2) Correct the discriminant conditions on site Once the discriminant conditions are generated from the version space, they cannot be changed. Further, even if there is an error in the discrimination processing, the discrimination condition cannot be corrected. Therefore, we made it possible to input the expert's evaluation of the discrimination result and correct the discrimination condition on the spot.

[0030]

Embodiments of the present invention will be described below with reference to the drawings. 1 and 2 show a first embodiment of the first invention, and this embodiment is for generating a discrimination condition in a storage device from case data including a single positive example. In FIG. 1, the case data 300 includes an attribute column that stores attribute ₁ 310 ₁ to attribute _n 310 _n , and a determination result 330 that determines these as positive or negative.
And a determination result column for storing one case, one row constitutes one case. The storage device 100 is a field for storing a combination 110 of attribute values in which the values of the same attribute do not overlap, the case data 300.
For storing a positive case existing in the case, or a combination of the attributes of the positive case, that is, a column for storing the number 130 of positive cases, and a negative value that is an integrated value of the number of negative cases. A column for storing the number of examples 131 and a column for storing the necessity 150 are provided.

The processor 200 stores the attributes ₁ 310 ₁ to _n 310 _n and the discrimination result 330 stored in the case data 300.
Thus, the values are stored in the columns of the combination 110 of attribute values, the number 130 of positive examples, and the number 131 of negative examples of the storage device 100 in the following manner, and the values are set in the column 150 of necessity.

(1) Initialize the storage device 100. 1) Leave all the fields of the attribute value combination 110 blank. 2) Set "0" in the column for the number of positive examples 130. 3) Set "0" in the negative number 131 column. 4) Set "Blank" in the column of necessity 150.

(2) A positive example is set in the storage device 100. 1) A line whose discrimination result 330 is “correct” is searched from the case data 300, and its attributes ₁ 310 ₁ to _n 310 _n are read and set in the column of the attribute value combination 110 of the storage device 100. 2) Set "1" in the column of the number 130 of positive examples of the set 110 of attribute values. 3) Create all combinations of the read attribute ₁ 310 ₁ to attribute _n 310 _n , and add them sequentially to the line after the column of the attribute value combination 110 in 1).

(3) A negative example is set in the storage device 100. 1) From the case data 300, search for a line in which the determination result 330 is “negative”, read its attributes ₁ 310 ₁ to _n 310 _n , and set them in the column of the attribute value combination 110 of the storage device 100 so that they do not overlap. To do. 2) The combination of the read attribute ₁ 310 ₁ to attribute _n 310 _n includes the attribute value combination 110 of the storage device 100.
If it exists in the column of, the value in the column of the negative example number 131 is integrated by one. 3) The above operations 1) and 2) are repeated for all the remaining cases where the discrimination result 330 in the case data 300 is “negative”.

(4) Necessity Setting 1) In the storage device 100, the negative example number 131 is “0”, the necessity 150 column is “blank”, and the attribute value combination 110 column is A line having the smallest number of attributes is searched, and "necessity" is set in the column of necessity 150 of the line. 2) In the storage device 100, search for all rows in which the number of negative examples 131 is “0” and the column of necessity 150 is “blank” and includes the attribute combinations of the rows searched in 1) Then, “No” is set in the column of necessity 150 of those lines. 3) In the storage device 100, for a line in which the number of negative examples 131 is “0” and the column of necessity 150 is “blank”,
Repeat steps 1) and 2) above.

By performing the processes (1) to (4) as described above, for example, the attribute values a1 and b as shown in FIG.
The combination 110 of attribute values of the storage device 100, the number 130 of positive examples, the number 131 of negative examples, and the necessity 150 can be set from the case data 300 having 1, c1, ....

According to this embodiment, the discrimination result 330 of the case data 300 is included in the case where the discrimination result 330 is "positive", that is, the positive example, and is not included in the case where the discrimination result 330 is "negative", that is, the negative example. Attribute value combinations 110 (a1, c
1 and b1, c1) can be easily detected, and these can be generated as positive and negative discrimination conditions. Also, since not all combinations of attribute values are expanded,
The storage capacity required for the storage device 100 can be reduced, and a small computer can be used. After processing all the case data 300, the case data 300 is generated to generate the most general discriminant condition among the remaining combinations.
Learning can be completed with less. This has the effect of being able to use case data from the site.

FIG. 3 shows a second embodiment of the first invention. This embodiment has the same components as those of the first embodiment, but the processing contents of the processing device 200 are different, and the storage device 1 is selected from the case data 300 including a plurality of positive examples.
The determination condition is generated in 00. That is, the processing device 200 has the attribute ₁₃ stored in the case data 300.
From _{1 1} to attribute _n 310 _n and the determination result 330, values are stored in the respective fields of the combination 110 of attribute values, the number 130 of positive examples and the number 131 of negative examples of the storage device 100 in the following manner, and whether or not there is a need 150 Set a value in the column.

(1) Initialize the storage device 100. 1) Leave all the fields of the attribute value combination 110 blank. 2) Set "0" in the column for the number of positive examples 130. 3) Set "0" in the negative number 131 column. 4) Set "Blank" in the column of necessity 150.

(2) The first positive example is set in the storage device 100. 1) A line whose discrimination result 330 is “correct” is searched from the case data 300, and its attributes ₁ 310 ₁ to _n 310 _n are read and set in the column of the attribute value combination 110 of the storage device 100. 2) Set "1" in the column of the number 130 of positive examples of the set 110 of attribute values. 3) Create all combinations of the read attribute ₁ 310 ₁ to attribute _n 310 _n , and add them sequentially to the line after the column of the attribute value combination 110 in 1). 4) From the case data 300, a line in which the determination result 330 is “correct” is searched for, the attributes ₁ 310 ₁ to _n 310 _n are read, and the attributes that do not exist in the storage device 100 are the combination of the read attribute values. The value combination is the attribute value combination 11
Sequentially add to the lines after the 0 column. This case data 300
This operation is repeated for all the remaining rows in which the column of the discrimination result 330 in is “correct”.

(3) Negative examples are sequentially set in the storage device 100. 1) From the case data 300, search for a line in which the determination result 330 is “negative”, read its attributes ₁ 310 ₁ to _n 310 _n , and set them in the column of the attribute value combination 110 of the storage device 100 so that they do not overlap. To do. 2) The combination of the read attribute ₁ 310 ₁ to attribute _n 310 _n includes the attribute value combination 110 of the storage device 100.
If it exists in the column of, the value in the column of the negative example number 131 is integrated by one. 3) The above operations 1) and 2) are repeated for all the remaining cases where the discrimination result 330 in the case data 300 is “negative”.

(4) Necessity Setting 1) In the storage device 100, the negative example number 131 is "0", the necessity 150 column is "blank", and the attribute value combination 110 column is A line having the smallest number of attributes is searched, and "necessity" is set in the column of necessity 150 of the line. 2) In the storage device 100, search for all rows in which the number of negative examples 131 is “0” and the column of necessity 150 is “blank” and includes the attribute combinations of the rows searched in 1) Then, “No” is set in the column of necessity 150 of those lines. 3) In the storage device 100, for a line in which the number of negative examples 131 is “0” and the column of necessity 150 is “blank”,
Repeat steps 1) and 2) above.

By performing the processes (1) to (4), the attribute values a1, b1, c as shown in FIG. 3 are obtained.
From the case data 300 having 1, ...
110 attribute value combinations, 130 positive examples, 131 negative examples
Also, the necessity 150 can be set, and the determination condition can be generated.

According to this embodiment, a combination 110 of attribute values such that the discrimination result 330 of the case data 300 is included in the case of “positive” and is not included in the case of the discrimination result 330 of “negative”. Can be easily detected, and these can be generated as the discrimination conditions of the positive example and the negative example. Further, also in this embodiment, the same effect as that of the first embodiment can be obtained.

FIG. 4 shows a third embodiment of the first invention. In this embodiment as well, the processing contents in the processing device 200 are different from those in the first embodiment, and case data with missing values collected at the site. The determination condition is generated in the storage device 100 from 300. That is, the processing device 200 performs the process of detecting the missing value of the case data 300 and setting the content of the storage device 100 in addition to the processes (1) to (4) in the first embodiment.

* Missing value processing When there are missing values in the attributes ₁ 310 ₁ to _n 310 _n of the case data 300, the following processing is performed. The contents of the storage device 100 are generated by performing the processes (1) to (4) in the first embodiment. At this time, when the attribute 310 including the missing value is detected, the attribute is excluded and set in the storage device 100. For example, as shown in Fig. 4, the attributes in the first case
_If the value of ₂ is missing, the remaining attributes except this attribute _2
1 , combinations of values of attribute ₃ (a1, c1), (a1), (c
1) is created and set in the storage device 100.

According to this embodiment, in addition to the same effect as the first embodiment, even if the case data 300 collected at the site includes a missing value, this is allowed and the contents of the storage device 100 are set. However, it is possible to generate the determination condition for determining the positive example and the negative example.

FIGS. 5 and 6 show a fourth embodiment of the first invention, which is a modification of the first embodiment to which a process for generating a knowledge base 500 is added. The knowledge base 500 shown in FIG. 5 includes IF to T represented by production rules.
It is composed of HEN format rules.

The processing device 200 performs the following processing. (1) Generation of Storage Device 100 The processes (1) to (4) in the first embodiment are performed to set values in the storage device 100. (2) Generation of Knowledge Base 1) In the storage device 100, a line in the column of necessity 150 is “necessary” is searched. 2) Based on the combination 110 of the attribute values of the retrieved row, the condition part (IF part) of the rule is expressed by the attribute conjunctive form (form connected by and), and the rule conclusion part (THE
A rule that makes N part) "positive" is generated. 3) In the storage device 100, the above-mentioned operations 1) and 2) are repeated for all the remaining items whose “necessity” 150 column is “necessary”.

When the processes (1) to (4) in the first embodiment are performed, the storage device 100 having the contents as shown in FIG. 6 is generated, and further, the knowledge base generating operation described above is performed to thereby generate the knowledge base. 500 is generated. According to this embodiment, in addition to the effect of the first embodiment, there is an advantage that since the discriminating condition is a rule, it is possible to make inference at high speed and to use general inference software.
Therefore, it is possible to cooperate with the existing rule type system, and it is possible to efficiently perform the discrimination processing on the spot.

7 and 8 show a fifth embodiment of the first invention. In this embodiment, a use attribute designating section 350 is added to the first embodiment. In FIG. 7, the usage attribute designating unit 350 sets “necessary” indicating that it is used for generating the discrimination condition or “blank” indicating that it is not used in correspondence with the attributes ₁ 310 ₁ to _n 310 _n of the case data 300. It is composed of columns for storing.

The processing device 200 stores as follows.
Generate the contents of device 100. (0) Determining attributes to be used Attributes 310 for which "necessary" is set in the usage attribute designating section 350
Attributes of₁310₁~attribute _n310_nUse only. (1) Generation of Storage Device 100 Using the attribute 310 designated to be used, the first embodiment
Processing (1) to (4) in the storage device 100.
Generate content. For example, in the case of FIG.₂And attributes₃
The contents of the storage device 100 are generated using only the.

According to this embodiment, in addition to the same effects as the first embodiment, there is an effect that the time for generating the contents of the storage device 100 can be shortened and the capacity of the storage device 100 can be reduced. Furthermore, the knowledge base or the discrimination condition can be adjusted without recreating the case base 300.

9 to 12 show a sixth embodiment of the first invention. In this embodiment, a knowledge base can be generated from continuous value case data 300. The case data 300 in the first embodiment is used as case data having continuous values, and an attribute value definition unit 400 is added. It is a thing. In FIG. 9, the case data 300 includes attributes ₁ to _n.
Data ₁ 320 ₁ to data _n 32 having continuous values corresponding to
It has 0 _n . The attribute value definition unit 400 includes an attribute 410, a satisfaction condition 420, and an attribute value 430.

The processing device 200 sets the continuous value of the attribute 410 to the attribute value 43 in the processes (1) to (4) in the first embodiment.
The process of replacing with 0 is added to generate the contents of the storage device 100. That is, the processing device 200 performs the following processing. (0) Determination of attribute value 1) Read data 320 ₁ having continuous values from case data 300. 2) With respect to each data 320 ₁ , the satisfaction condition 420 of the attribute ₁ stored in the attribute value definition unit 400 is searched. 3) The attribute value 430 of the row whose data satisfies the condition 420
Is written in the column of attribute 310 ₁ of the case data 300. 4) Perform steps 1) to 3) on all case data.

(1) Generation of Storage Device 100 The processes (1) to (4) in the first embodiment are executed to generate the contents of the storage device 100. For example, case data 30
When 0 and the attribute value definition unit 400 are as shown in FIGS. 10 and 11, respectively, the continuous value of the attribute 410 is converted into the attribute value as shown in FIG.

According to this embodiment, in addition to the same effect as the first embodiment, the contents of the storage device 100 can be generated from the case data 300 of continuous values, and the attribute of taking continuous values can be used. There is an advantage that case data from the field can be used.

13 and 14 show a seventh embodiment of the first invention. In this embodiment, the output device 6 is the same as the first embodiment.
00 is added to the processing device 200, and a predetermined process is added to the processing device 200 to display the validity of the determination condition. In FIG. 13, the output device 600 displays the number of related positive example numbers 130 for the determination condition that the necessity 150 of the storage device 100 is “necessary”.

Processing device 200 performs the following processing. (0) Generation of storage device 100 The processes (1) to (4) in the first embodiment are performed. (1) Display of the number 130 of positive examples 1) Read the line in the storage device 100 whose necessity 150 is "necessary". 2) Including the combination 110 of attribute values in that row, the number of positive examples is 13
A row of the storage device 100 in which 0 is “1” and necessity 150 is “no” is searched. 3) Count the rows of the corresponding storage device 100. 4) The operations of 1) to 3) are repeated for all the rows in the storage device 100 whose necessity 150 is “necessary”, and the output device 6
The number of related positive example numbers 130 is displayed for the determination condition that is “necessary” in 00.

Note that FIG. 14 shows the display contents of the output device 600 with respect to the contents of the case data 300 and the storage device 100. According to this embodiment, in addition to the effect of the first embodiment, the frequency of use of the generated discriminating condition can be grasped by looking at the number of case data used for generating the discriminating condition. Further, the validity of the generated discriminant conditions can be ordered, and the discriminant conditions can be easily evaluated.

15 to 17 show an eighth embodiment of the first invention. This embodiment, as shown in FIG.
The case data 300 in the embodiment is provided with a storage field for the case number 360, an output device 600 for displaying the result of checking the case data 300 is added, and the storage device 100 has a positive example number 130A and a negative example number 131A. The storage column is provided.

The processing device 200 executes the following processing. (1) Generation of storage device 100 1) Values are set in the storage device 100 by the processes (1) to (4) in the first embodiment. 2) However, when the storage device 100 is initialized, the positive example number 1
A process of setting "0" is added to the fields of 30A and negative example number 131A. 3) From the case data 300, the attribute ₁ 310 ₁ to the attribute _n 310 _n whose determination result 330 is “negative” are read, and the storage device 1 is read.
At the same time as setting in the field of attribute value combination 110 of 00,
The case number 360 of the case data 300 is stored in the storage device 100.
The process of setting in the column of negative example number 131A is added. Similarly, the determination result 330 is “correct” from the case data 300.
Attribute ₁ 310 ₁ to attribute _n 310 _n are set and set in the column of the attribute value combination 110 of the storage device 100, and at the same time, the case number 360 of the case data 300 is set to the storage device 1
A process of setting in the column of the positive example number 130A of 00 is added.

(2) Checking and Displaying Case Data 1) In the storage device 100, a line in which the number 130 of positive examples is larger than “0” and the number 131 of negative examples is also larger than “0” is searched. 2) The line retrieved in 1) above is output to the output device 600. 3) In the storage device 100, the above operations 1) and 2) are repeated for the remaining lines except the line searched in 1).

By performing the processing as described in (1) above, the storage device 100 having the contents as shown in FIG. 16 is generated. Further, by performing the processing of (2), a display example 601 as shown in FIG. 17 is obtained on the output device 600. According to this embodiment, in addition to the effects of the first embodiment,
Inconsistencies included in the case data 300 can be detected and displayed, and erroneous case data can be detected.

18 to 20 show a ninth embodiment of the first invention. In this embodiment, as shown in FIG. 18, the analysis result storage unit 800 having a target combination number 810 column and a positive combination number 820 column and the contents of the analysis result storage unit 800 are added to the first embodiment. An output device 600 for displaying is added, and a column for storing the combination number 120 is provided in the storage device 100. With this configuration, the state of the storage device 100 can be analyzed and the determination condition can be displayed. Is.

Processing device 200 performs the following processing. (1) Generation of storage device 100 1) Values are set in the storage device 100 by the processes (1) to (4) in the first embodiment. 2) When the storage device 100 is initialized, the combination number 120
A process for setting a serial number from "1" is added to the column of.

(2) State analysis of the storage device 100 1) Select one row in which the column of necessity 150 of the storage device 100 is "necessary" and the attribute value combination 110 has the smallest number of attribute values. . At the same time, the column of necessity 150 is changed to “*”. 2) The number of attribute values of the attribute value combination 110 of the row selected in 1) above is the same, and the "attribute value number-1" attribute values of the attribute value combination 110 are the same, and The row of “necessity” in the column of necessity 150 of the storage device 100 is searched.

3) When a line satisfying the condition of 2) above is searched,
Perform the following processing. a. In the column of necessity 150 of the line meeting the condition of 2), "*" indicating that the line is selected is set. b. The combination number 120 of the row selected in 1) and the combination number 120 of the row searched in 2) are set in the positive combination number 820 column of the analysis result storage unit 800. c. A row having an attribute combination common to the combinations 1) and 2) and having a “necessity” in the necessity / non-necessity 150 column of the storage device 100 is searched. (I) If the search is successful, the combination number 12 of this searched line is entered in the column of the positive combination number 820 of the analysis result storage unit 800.
At the same time as setting 0, the column of necessity 150 is changed to “*”. Then, the search is repeated again. (Ii) If it cannot be detected in the storage device 100, it will be described later.
Perform step 5).

4) If no line satisfying the condition of 2) above is searched and there is a line of "necessity" in the column of necessity 150 of the storage device 100, the process from 1) is repeated. 5) When one or more rows satisfying the condition of 2) above are retrieved, a row matching the attribute value combination common to all the rows of the storage device 100 indicated by the positive combination number 820 of the analysis result storage unit 800 is searched. The combination number 120 is searched and set as the target combination number 810 of the analysis result storage unit 800. 6) The above operations 1) to 5) are repeated until there is no line in the storage device 100 where the column of necessity 150 is “necessary”. 7) Finally, "*" in the column of necessity 150 of the storage device 100 is returned to "necessary".

(3) Display of Analysis Result Based on the contents of the analysis result storage section 800, the attribute value combination 110 indicated by the target combination number 810 and the attribute value combination 110 indicated by the positive combination number 820 are displayed. , Are displayed on the output device 600.

By performing the process (1), the storage device 100 having the contents shown in FIG. 19 is generated, and by performing the processes (2) and (3), the process shown in FIG. 20 is performed. Such a display example 601 is displayed on the output device 600. According to this embodiment, in addition to the effect similar to that of the first embodiment, it is possible to easily detect the common part of the rules generated from the storage device 100 and reduce the load during the development of the knowledge base system. Has the effect of
Further, in this embodiment, since the combination that can unify the generated discrimination conditions can be detected, the discrimination conditions can be easily evaluated.

21 to 23 show a tenth embodiment of the first invention. In this embodiment, as shown in FIG. 21, a column for storing the negative combination number 830 is added to the analysis result storage unit 800 in the ninth embodiment, and a column for storing the case number 360 is added to the case data 300. Along with this, a column for storing the negative example number 131A is added to the storage device 100, whereby the state of the storage device 100 is analyzed and the discrimination conditions and cases associated therewith are displayed.

Processing device 200 performs the following processing. (1) Generation of storage device 100 1) A value is set in the storage device 100 based on the process (1) in the ninth embodiment. 2) When initializing the storage device 100, a negative example number 131A
A process of setting “0” in the column of is added. 3) The attribute ₁ 310 ₁ to attribute 310 _{n in} which the column of the discrimination result 330 is “negative” is read from the case data 300 and set in the column of the attribute value combination 110 of the storage device 100, and at the same time, the case number of the case data 300 Storage device 1
A process to be set in the column of negative example number 131A of 00 is added.

(2) State analysis of storage device 100 1) Storage device 100 is analyzed based on the method shown in the ninth embodiment.
And a value is set in the analysis result storage unit 800. 2) A line that includes the attribute value combination 110 of the storage device 100 indicated by the target combination number 810 of the analysis result storage unit 800 and the negative example number 131A is not “0” is searched, and the combination number 120 is analyzed. The negative combination number 830 of the result storage unit 800 is set.

(3) Display of Analysis Result Based on the contents of the analysis result storage unit 800, the attribute value combination 110 indicated by the target combination number 810, the attribute value combination 110 indicated by the positive combination number 820, and the negative The negative example number 131A of the row of the combination number 120 of the storage device 100 having the same value as the combination number 830 of No.

By performing the process (1), the storage device 100 having the contents shown in FIG. 22 is generated, and by performing the processes (2) and (3), the process shown in FIG. 23 is performed. Such a display example 601 is displayed on the output device 600. According to this embodiment, in addition to the effect similar to that of the first embodiment, it is possible to easily detect a negative case that hinders the commonization of the determination condition, so that exceptional case data can be detected. The effect is that the load at the time of developing a knowledge base system is reduced. Further, in this embodiment, since the combination that can unify the generated discrimination conditions can be detected, the discrimination conditions can be easily evaluated.

24 to 26 show an eleventh embodiment of the first invention. In this embodiment, as shown in FIG. 24, a column for storing a test identification 340 for identifying whether or not to use for a test is added to the case data 300 in the first embodiment, and the storage device 100 is established. Number 160, number of failures 1
61 and a column for storing the number of false establishments 162 are added, and an output device 600 is added. As a result, the state of the storage device 100 can be tested and the number of cases is displayed.

Processing device 200 performs the following processing. (1) Generation of Storage Device 100 1) Storage device 100 based on the method shown in the first embodiment.
Set the value to. It should be noted that only the rows in which the test identification 340 of the case data 300 is not "necessary" are used. 2) When the storage device 100 is initialized, a process of setting “0” is added to all the columns storing the established number 160, the unestablished number 161, and the falsely established number 162.

(2) Test the state of the storage device 100 1) From the case data 300, the test identification 340 is "necessary"
Attribute ₁ 310 ₁ to attribute _n 310 _n and the determination result 33
Read 0. 2) read the attribute ₁ 310 ₁ to the combination of the attribute _n 310 _n (i.e. a combination of attribute values), necessity 15 of the storage device 100
Compare the attribute value combinations 110 of the rows where 0 is “necessary”,
Make the following decisions. a. The combination of the read attribute ₁ 310 ₁ to attribute _n 310 _n is, includes a combination 110 of an attribute value, and the determination result 3
If 30 is “positive”, “1” is added to the column of the established number 160. b. The combination of the read attribute ₁ 310 ₁ to attribute _n 310 _n is, includes a combination 110 of an attribute value, and the determination result 3
If 30 is “negative”, “1” is added to the column of the number of false establishments 162. c. The above operations a and b are repeated for the row in the storage device 100 where the column of necessity 150 is "necessary". d. As a result of the operation c, if the determination result 330 is “correct” but the necessity 150 of the storage device 100 does not include one of the attribute value combinations 110 of “necessary”. For example, “1” is added to the column of the unsatisfied number 161. 3) The above operations 1) and 2) are performed on all the rows of the case data 300 for which the test identification 340 is “necessary”.

(3) Display of Number of Cases The number of cases is displayed on the output device 600 based on the established number 160, the unestablished number 161, and the falsely established number 162 of the storage device 100.

By performing the processing of (1), the storage device 100 as shown in FIG. 25 is created. Also,
By the processing of (2) and (3), the number of established storage devices 100 is 1
60, the number of non-established 161, the number of falsely established 162, the output device 60
When it is displayed at 0, it becomes as shown in FIG. According to this embodiment, the same effect as in the first embodiment can be obtained, and it is possible to evaluate how much test case data is satisfied as the performance of the discrimination condition, and the discrimination condition can be evaluated without recreating the case data. Since the test can be performed, the judgment condition can be effectively evaluated. Furthermore, it is possible to reduce the load when developing the knowledge base system.

27 to 29 show a twelfth embodiment of the first invention. In this embodiment, as shown in FIG.
The case number 36 is added to the case data 300 in the eleventh embodiment.
A column for storing 0 is added, and a column for storing the false positive example number 171 and the false negative example number 172 is added to the storage device 100. As a result, the storage device 100
The state of was tested and the case number and case number were displayed.

Processing device 200 performs the following processing. (1) Generation of storage device 100 1) Storage device 1 based on the method described in the eleventh embodiment
Set the value to 00. 2) When the storage device 100 is initialized, “0” is entered in the columns for storing the established number 160, the unestablished number 161, and the falsely established number 162.
Is added to the column for storing the non-established positive example number 171 and the false established negative example number 172. (2) Testing the state of the storage device 100 Based on the method described in the eleventh embodiment, the number of failures 16
Processing for setting the case number 360 in the columns of the positive positive example number 171 and the negative negative example number 172 is added for the case of integrating 1 and the case of accumulating the number of false cases 162, respectively. (3) Display of the case number In addition to the display of the number of cases according to the eleventh embodiment, a process for displaying the unsatisfied positive example number 171 and the falsely established negative example number 172 is added.

By the process (1), the storage device 100 as shown in FIG. 28 is generated. Also, (2),
By the processing of (3), the number of established 160, the number of unestablished 161,
FIG. 29 shows the case number displayed together with the falsely established number 162. According to this embodiment, the performance of each discrimination condition,
The case number 360 of the wrong case data 300 becomes clear. Also, the load at the time of developing the knowledge base system is reduced. Furthermore, since it is possible to detect a case in which the discrimination condition is erroneously discriminated, it is possible to detect unknown case data that was not used during learning.

30 to 32 show a first embodiment of the second invention. In this embodiment, the contents of the storage device 100 are used for inference and learning.
The storage device 100 shown in FIG. 30 is, for example, the first invention of the first invention.
It is generated by the embodiment. This storage device 10
0 is the combination 110 of attribute values and the number of positive examples is 13 as described above.
It is composed of columns for storing 0, the number of negative examples 131, and necessity 150. The processing device 200 inputs the value of the attribute from the input device 700, infers it according to the contents of the storage device 100,
The inference result is output to the output device 600. Further, the evaluation of the inference result is input from the input device 700, and the content of the storage device 100 is changed. Thus, the processing device 2
00 repeatedly inputs the value of the attribute from the input device 700 and outputs the inference result to the output device 600,
At the same time, the contents of the storage device 100 are changed.

Details of the processing in the processing device 200 will be described below. For example, the contents of the storage device 100 are shown in FIG.
It will be described in detail below. (1) Inference 1) Input an attribute value from the input device 700. For example, the following methods are available for inputting the attribute value. This embodiment is effective regardless of which method is adopted. The user of the system using this embodiment sets attribute values. -The system using this embodiment reads the attribute value in a file or the like. The system using this embodiment presents the attribute name to the user and reads the value. Here, it is assumed that a1, b2, and c1 are input as the values of attribute ₁ , attribute ₂ , and attribute ₃ , respectively.

2) The attribute value input from the input device 700 is
A comparison is made with the attribute value combination 110 of the row in the storage device 100 where the column of necessity 150 is “necessary”. If the input attribute value matches the content of the attribute value combination 110 or includes the content, an inference result of "correct" is output to the output device 6
Output to 00. If the input attribute value does not match the content of the attribute value combination 110 and does not include the content, the necessity / non-necessity 150 column in the storage device 100 indicates the next row. Compare with the attribute value combination 110. Then, if the input attribute value matches the content of the attribute value combination 110 or includes the content, the inference result “correct” is output to the output device 600.

If the input attribute value is the storage device 100
In the case where the column of necessity 150 does not match the contents of the attribute value combinations 110 of all the lines that are “necessary” and does not include the contents, the inference result “negative” is output to the output device 600. To do. Here, the input attribute values a1, b2, c1 are
Since the column of necessity 150 in the storage device 100 includes the contents a1 and c1 of the attribute value combination 110 of the first row that is “necessary”, the inference result “correct” is output to the output device 600. .

(2) Evaluation of inference result 3) Subsequently, evaluation of the inference result is input from the input device 700. At this time, it is assumed that the previously input attribute value has been saved. For example, the following methods are available for inputting the evaluation of the inference result. This embodiment is effective regardless of which method is used. -The system using this embodiment presents the inference result to the user and reads the evaluation. -The system using this embodiment determines the evaluation based on the information in the file or the like. If the inference result output to the output device 600 is correct, nothing is set in the input device 700. If the inference result is not correct, the correct conclusion is set in the input device 700. Here, it is assumed that the inference result “correct” for the attribute values a1, b2, c1 is evaluated as incorrect. In this case, the input device 700 is set to the correct conclusion “negative”.

(3) Correction of contents of storage device 4) When the inference result is evaluated as correct and nothing is set in the input device 700, no processing is performed. 5) When the inference result is evaluated to be incorrect and another conclusion is set in the input device 700, the following processing is performed. 6) When the conclusion set in the input device 700 is "negative", a line is added to the end of the storage device 100, and the attribute value combination 110 has the attribute value stored in the input device 700 set. To do. Then, "0" and "1" are set in the positive example number 130 and the negative example number 131 of the row, respectively.
Further, "blank" is set in the column of necessity 150 of that line.

7) A row in the storage device 100 is searched for a row having an attribute value combination 110 including the content of the attribute value combination 110 including the content of the attribute value combination 110 of the row added in 6) above. . Then, the value in the negative example number 131 column of that row is increased by "1". 8) If the content of the necessity 150 of the line searched in 7) above,
If it is "required", it is set to "blank". 9) A combination of attribute values including the contents of the combination 110 of the attribute values of the row added in 6) above by searching the rows of the storage device 100
The above processes 7) and 8) are repeated for all the rows having 0. When all the processing is completed, move to the next processing.

10) Search for a line in the storage device 100 and determine whether it is necessary 1
Search for the line in which the content of 50 is "No". Then, the content of the attribute value combination 110 in that row is “necessary” and the content of the necessity 150 is “necessary”
If the content of the attribute value combination 110 of any other row is included, no processing is performed and it is left as “no”. 11) If the content of the attribute value combination 110 of the row searched for in 10) above does not include the content of the attribute value combination 110 of any other row whose content of necessity 150 is "necessary" Changes the content of the necessity 150 from "no" to "necessary". 12) The contents of the storage device 100 are searched, and the processes of 10) and 11) above are repeated for all the lines in the contents of the necessity 150 which are “No”.

When all the processing is completed, the processing starting from the above 6) when the conclusion set in the input device 700 is "negative" is completed. In this example, the contents of the storage device 100 are changed as shown in FIG. 32 by the processes 6) to 12). That is, the attribute values a1, b2, c1 are in the last line.
Is added, and the number of positive examples and the number of negative examples are set to “0” and “1”, respectively. Then, the attribute values a1 and b
A1, c1 row, a1 row, c containing a combination of 2, c1
Negative numbers in the row of 1 are added by "1".
Further, the necessity of the lines a1 and c1 is changed from "necessary" to "blank".

13) When the conclusion set in the input device 700 is "correct", a line is added to the end of the storage device 100,
The attribute value stored in the input device 700 is set in the column of the attribute value combination 110. Then, "1" and "0" are set to the positive example number 130 and the negative example number 131 of the row, respectively. Further, "necessary" is set in the column of necessity 150 in that row. The above is the details of the processing in the processing device 200.

According to this embodiment, the same effect as that of the first embodiment of the first invention can be obtained, and inference can be performed using the contents of the storage device 100. By evaluating the inference result, If the inference is not correct, the storage device 100
There is an advantage that the contents of can be corrected immediately on the spot. That is, since the determination condition can be corrected based on the evaluation of the determination result during operation, the determination process can be efficiently performed on site.

FIG. 33 shows a second embodiment of the second invention. In this embodiment, the case data 300 is added to the first embodiment of the second invention so that the case data 300 can be accumulated while learning. In FIG. 33, the storage device 100 is constituted by, for example, the first embodiment of the first invention, and is the same as the storage device 100 according to the first embodiment of the second invention. Also, case data 30
0 consists of a storage column for attributes ₁ 310 ₁ to _n 310 _{n, and} a storage column for the discrimination result 330, all of which are initially blank.

As in the first embodiment of the second invention, the processing device 200 inputs the attribute value from the input device 700, infers it according to the contents of the storage device 100, and outputs the inference result to the output device 600. . Further, the evaluation of the inference result is input from the input device 700, and the content of the storage device 100 is changed. As described above, the processing device 200 includes the input device 7
The attribute value is input from 00 and the inference result is output to the output device 600.
Output to the storage device 10 at the same time.
The contents of 0 will be changed. The difference from the first embodiment of the second invention is that a process of storing the input attribute value and the evaluation of the inference result in the case data 300 is added.

Details of the processing in the processing device 200 will be described below. For example, the contents of the storage device 100 are shown in FIG.
It will be described in detail below. (1) Inference Since this process is the same as that described in the first embodiment of the second invention, detailed description thereof will be omitted. That is, the following processing is performed. 1) Input an attribute value from the input device 700. 2) The attribute value input from the input device 700 is stored in the storage device 10.
It is compared with the attribute value combination 110 of the row in which the column of necessity 150 in 0 is “necessary”.

(2) Evaluation of inference result The processing of 3) below is partly different from the first embodiment of the second invention. 3) Then, the evaluation of the inference result is input from the input device 700. At this time, it is assumed that the previously input attribute value has been saved. For example, the following methods are available for inputting the evaluation of the inference result. This embodiment is effective regardless of which method is used. -The system using this embodiment presents the inference result to the user and reads the evaluation. -The system using this embodiment determines the evaluation based on the information in the file or the like. If the inference result output to the output device 600 is correct, nothing is set in the input device 700. In this case, the attribute value stored in the input device 700 and the output device 600
The inference result output to each of the case data 300 is stored in the columns of attribute ₁ 310 ₁ to attribute _n 310 _n and the discrimination result 330 column in the last row of the case data 300.

If the inference result output to the output device 600 is incorrect, a correct conclusion is set in the input device 700. In this case, the attribute value stored in the input device 700 and the inference result set in the input device 700 are used as the attributes ₁ 310 ₁ to _n of the last line of the case data 300, respectively.
It is stored in the column of 310 _{n and} the column of the discrimination result 330.

(3) Correction of contents of storage device The following processing is the same as the processing 4) to 13) in the first embodiment of the second invention, and therefore detailed description thereof will be omitted.

According to this embodiment, in addition to the effects of the first embodiment of the second invention, case data for use in, for example, the first embodiment of the first invention is based on the evaluation of the discrimination result during operation. There is an advantage that 300 can be collected automatically. As a result, the discrimination conditions can be improved even during operation, and the discrimination processing can be efficiently performed on site.

34 to 38 show a third embodiment of the second invention. In this embodiment, a knowledge base 500 is added to the first embodiment of the second invention so that learning can be performed while inferring using the knowledge base 500. In FIG. 34, the storage device 100 and the knowledge base 500 are generated by, for example, the fourth embodiment of the first invention. The storage device 100 is, as described above, the combination 110 of attribute values, the number 130 of positive examples, the number 131 of negative examples, and necessity 1
It consists of 50 columns. In the knowledge base 500, among the combinations of attribute values in the storage device 100, the rules corresponding to the attribute value combinations 110 of the rows in which the column of necessity 150 is “necessary” are stored.

The processing device 200 inputs the value of the attribute from the input device 700, makes an inference according to the contents of the knowledge base 500, and outputs the inference result to the output device 600.
Furthermore, the evaluation of this inference result is input from the input device 700, and the contents of the storage device 100 and the knowledge base 500 are changed. As described above, the processing device 200 includes the input device 7
The attribute value is input from 00 and the inference result is output to the output device 600.
Output to the storage device 10 at the same time.
The contents of 0 and the knowledge base 500 are changed.

The details of the processing in the processing device 200 will be described below. For example, the storage device 100 and the knowledge base 5
It is assumed that the contents of 00 are as shown in FIGS. 35 and 36, respectively. (1) Inference 1) This process is the same as the process 1) in the first embodiment of the second invention. That is, the value of the attribute is input from the input device 700. The value of the attribute to be input is a
1, b2, c1. 2) This process is different from the first embodiment of the second invention. That is, the attribute value input from the input device 700 is compared with the combination of the attribute values of the rules stored in the knowledge base 500. If the input attribute value matches the content of the combination of the attribute values of the rule or includes the content, the inference result “correct” is output to the output device 600. If the input attribute value does not match the content of the attribute value combination of the rule and does not include the content, it is compared with the attribute value combination of the next rule stored in the knowledge base 500.

Then, if the input attribute value matches or contains the content of the combination of the attribute values of the rule,
The inference result “positive” is output to the output device 600. If the input attribute value does not match the content of the attribute value combinations of all the rules stored in the knowledge base 500 and does not include the content, the inference result “negative” is output to the output device 600. Output. Here, the input attribute value a
1, b2 and c1 include a1 and c1 which are the contents of the combination of the attribute values of the first rule stored in the knowledge base 500, the inference result of "correct" is output to the output device 600.
Output to.

(2) Evaluation of inference result This processing is the processing in the first embodiment of the second invention.
Since it is the same as 3), detailed description is omitted. (3) Correction of contents of storage device and knowledge base In this processing, first, 4) described in the first embodiment of the second invention is described.
~ Perform the same processing as 7). After that, the following process 8) is performed. 8) If the content of the necessity 150 of the line searched by the process of 7) is "necessary", it is set to "blank". Then, the rule corresponding to this row is deleted from the knowledge base 500.

Then, in the first embodiment of the second invention.
Perform steps 9) and 10). After that, the following 11) processing is executed. 11) If the content of the row combination 110 searched by the processing of 10) does not include the content of the attribute value combination 110 of any other row whose content of necessity 150 is “necessary” The content of the necessity 150 is changed from “No” to “Necessary”. Then, the rule corresponding to this row is added to the knowledge base 500. The subsequent processing is the same as the processing of 12) in the first embodiment of the second invention. That is, 12) the rows of the storage device 100 are searched, and the processes of 10) and 11) above are repeated for all the contents of the necessity 150, which is “NO”. When all the processing is completed, the processing starting from 6) when the conclusion set in the input device 700 is “negative” is completed.

In this example, by the processing of 6) to 12) above,
The contents of the storage device 100 and the knowledge base 500 are changed as shown in FIGS. 37 and 38, respectively. That is, the combination of the attribute values a1, b2, and c1 is added to the last line, and the number of positive examples and the number of negative examples are set to "0" and "1", respectively. Then, the negative example numbers of the rows of a1, c1, the rows of a1, and the rows of c1 including the combination of the attribute values a1, b2, c1 are respectively added by “1”. Further, the necessity of the lines a1 and c1 is changed from "necessity" to "no". Along with this, the corresponding rule is deleted from the knowledge base 500.

13) When the conclusion set in the input device 700 is "correct", a line is added to the end of the storage device 100, and the attribute stored in the input device 700 is added to the attribute value combination 110 column. Set the value. Then, "1" and "0" are set to the positive example number 130 and the negative example number 131 of the row, respectively. Further, "necessary" is set in the column of necessity 150 in that row. Then, the rule corresponding to this row is added to the knowledge base 500. The above is the details of the processing in the processing device 200.

According to this embodiment, in addition to the effect of the first embodiment of the second invention, since the discriminating condition is a rule, it is possible to cooperate with the existing rule type system and enable high speed inference. There is an effect that the discrimination processing can be efficiently performed on site and general software can be used.

39 to 41 show a fourth embodiment of the second invention. In this embodiment, an attribute value defining unit 400 is added to the first embodiment of the second invention, whereby inference is performed and learning is performed while using the contents of the storage device 100 using continuous values. It is a thing. In FIG. 39, the storage device 100 and the attribute value definition unit 400 are generated by, for example, the sixth embodiment of the first invention.
The configuration of the storage device 100 is the same as that described above, and the attribute value definition unit 400 uses the attribute 41 that stores the names of the attributes having continuous values.
0, Satisfaction condition 4 that stores the determination condition for the continuous value
20 and an attribute value 430 column that stores the attribute value of the corresponding attribute when the determination condition is satisfied.

The processing device 200 inputs the continuous value of the attribute from the input device 700, converts the continuous value into the attribute value according to the contents of the attribute value defining section 400, and makes an inference according to the contents of the storage device 100. Is output to the output device 600. Furthermore, the evaluation of the inference result is performed by the input device 7
00 to change the contents of the storage device 100. As described above, the processing device 200 repeatedly inputs the value of the attribute from the input device 700 and outputs the inference result to the output device 600, and simultaneously changes the content of the storage device 100.

The details of the processing in the processing device 200 will be described below. For example, it is assumed that the contents of the storage device 100 and the attribute value definition unit 400 are as shown in FIGS. 40 and 41, respectively. (1) Inference The process 1) below is different from the process 1) in the first embodiment of the second invention. 1) Input an attribute value from the input device 700. There are the following methods of inputting attribute values, for example, and the present embodiment is effective in any method. The user of the system using this embodiment sets attribute values. -The system using the present embodiment reads the value of an attribute in a file or the like. The system using this embodiment presents the attribute name to the user and reads the value.

Then, these values are set to the attribute value definition unit 400.
Convert to an attribute value using. That is, for each attribute, the line stored in the attribute value definition unit 400 is searched for in which the content of the attribute 410 matches the name of the corresponding attribute.
The value of the corresponding attribute input from the input device 700 is compared with the determination condition stored in the satisfaction condition 420 of the row. If the determination condition is met, the attribute value 4 of the line
The content of 30 is set as the attribute value of the corresponding attribute. If the determination condition is not satisfied, the line stored in the attribute value definition unit 400 is searched for the next item whose contents of the attribute 410 match the name of the corresponding attribute. Then, the comparison with the determination condition is repeated to finally determine the attribute value of the corresponding attribute.

The attribute value converted by using the attribute value defining section 400 will be referred to as an attribute value input from the input device 700 here for convenience. For example, attribute ₁ , attribute ₂ , attribute
_{It is} assumed that 6, 24 and 29 are input as the values of ₃ , respectively. For the attribute ₁ , the determination condition of the first row stored in the attribute value definition unit 400 is satisfied, and the attribute value a
Converted to 1. For attribute ₂ , the attribute value definition unit 40
The discriminating condition of the third row stored in 0 is not satisfied, but the fourth discriminating condition is satisfied, and the attribute value b2 is converted. For the attribute ₃ , the determination condition of the fifth row stored in the attribute value definition unit 400 is satisfied, and the attribute value c1 is converted. That is, from the input device 700 to a1, b
This is equivalent to inputting attribute values of 2 and c1. Next, the same processing as the processing 2) in the first embodiment of the second invention is performed.

(2) Evaluation of inference result, and (3) Correction of contents of storage device. These are the processes in the first embodiment of the second invention.
Perform the same processing as 3) to 13).

According to this embodiment, in addition to the effect of the first embodiment of the second invention, since an attribute that takes a continuous value can be used, it is possible to use on-site case data.

FIG. 42 shows a fifth embodiment of the second invention. In this embodiment, the case data 300 is added to the fourth embodiment of the second invention, whereby the case data is automatically accumulated while learning using continuous values. The storage device 100 and the attribute value definition unit 400 shown in FIG. 42 are generated by, for example, the sixth embodiment of the first invention, similarly to the fourth embodiment of the second invention.

Also in this embodiment, similarly to the fourth embodiment of the second invention, the processing device 200 inputs the continuous value of the attribute from the input device 700 and attributes the continuous value according to the contents of the attribute value defining section 400. The value is converted, inference is performed according to the contents of the storage device 100, and the inference result is output to the output device 600. Further, the evaluation of the inference result is input from the input device 700, and the content of the storage device 100 is changed.
As described above, the processing device 200 repeatedly inputs the value of the attribute from the input device 700 and outputs the inference result to the output device 600, and simultaneously changes the content of the storage device 100.

The difference of this embodiment from the fourth embodiment of the second invention is that a process of storing the continuous value of the input attribute and the evaluation of the inference result in the case data 300 is added. The details of the processing in the processing device 200 will be described below. (1) Inference First, the same processes 1) and 2) as in the fourth embodiment of the second invention are executed. (2) Evaluation of inference result The following processing 3) is different from the fourth embodiment of the second invention. 3) Input the evaluation of the inference result from the input device 700. At this time, it is assumed that the continuous value of the attribute previously input is saved. There are the following methods of inputting the inference result, for example, and the present embodiment is effective regardless of the method. -The system using this embodiment presents the inference result to the user and reads the evaluation. The system using the present embodiment determines the evaluation based on the information in the file or the like.

If the inference result output to the output device 600 is correct, nothing is set in the input device 700. In this case, the continuous values of the attributes stored in the input device 700 and the inference result set in the input device 700 are displayed in the columns of data ₁ 320 ₁ to data _n 320 _{n in} the last row of the case data 300, respectively. And the determination result 330 column.
If the inference result output to the output device 600 is incorrect, the correct conclusion is set in the input device 700. In this case, the continuous values of the attributes stored in the input device 700 and the inference result set in the input device 700 are displayed in the columns of data ₁ 320 ₁ to data _n 320 _{n in} the last row of the case data 300, respectively. And the determination result 330 column.

(3) Correction of contents of storage device In this processing, processing 4) to 4) are carried out in the same manner as in the fourth embodiment of the second invention.
13) is executed. The above is the details of the processing in the processing device 200.

According to this embodiment, in addition to the same effect as the fourth embodiment of the second invention, for example, the case data 300 for use in the sixth embodiment of the first invention can be automatically collected. There is. In addition, case data at the site can be used.

FIG. 43 shows a sixth embodiment of the second invention. In this embodiment, a knowledge base 500 is added to the fourth embodiment of the second aspect of the invention so that the knowledge base 500 can infer and learn using continuous values. In FIG. 43, the storage device 100
The knowledge base 500 and the knowledge base 500 are generated by, for example, combining the sixth and fourth embodiments of the first invention. The processing device 200 inputs the continuous value of the attribute from the input device 700, converts the continuous value into the attribute value according to the contents of the attribute value definition unit 400, infers according to the contents of the knowledge base 500, and outputs the inference result to the output device 600. Output to. Further, the evaluation of this inference result is input from the input device 700, and the contents of the storage device 100 and the knowledge base 500 are changed. As described above, the processing device 200 includes the input device 700.
The value of the attribute is input and the inference result is output to the output device 600, and the contents of the storage device 100 and the knowledge base 500 are changed at the same time.

Details of the processing in the processing device 200 will be described below. (1) Inference First, the same processing as the processing 1) in the fourth embodiment of the second invention is performed. Next, the following processing 2) different from the fourth embodiment of the second invention is executed. 2) The attribute value input from the input device 700 is used as the knowledge base 5
00 and the combination of the attribute values of the rule stored in 00.
If the input attribute value matches the attribute value combination of the rule or contains the content, the inference result “positive” is output to the output device 600. If the input attribute value does not match the content of the attribute value combination of the rule and does not include the content, it is compared with the attribute value combination of the next rule stored in the knowledge base 500.

If the input attribute value matches or contains the content of the combination of rule attribute values,
The inference result “positive” is output to the output device 600. If the input attribute value does not match the content of the combination of attribute values of all rules stored in the knowledge base 500 and does not include the content, the inference result “negative” is output to the output device 600. .

(2) Evaluation of inference result This process is described in 3) in the fourth embodiment of the second invention.
The process is the same as that of. (3) Correction of contents of storage device and knowledge base First, the same processes 4) to 7) as in the fourth embodiment of the second invention are performed. Next, the following process 8) different from the fourth embodiment is performed. 8) If the content of the necessity 150 of the line searched in the process 7) is "necessary"
If it is, then leave it blank. Then, the rule corresponding to this row is deleted from the knowledge base 500.

Then, the same processing 9), 10) as in the fourth embodiment of the second invention is performed. Further, the following processing 11) different from the fourth embodiment is performed. 11) If the content of the attribute value combination 110 of the row searched for in the processing 10) does not include the content of the attribute value combination 110 of any other row whose content of necessity 150 is “necessary” Changes the content of the necessity 150 from "no" to "necessary". Then, the rule corresponding to this row is added to the knowledge base 500.

Next, the same processing as in the fourth embodiment of the second invention is performed.
After performing 12), the following process 13) is performed. 13) When the conclusion set in the input device 700 is “correct”, a line is added to the end of the storage device 100, and the attribute value combination 110 stores the attribute values stored in the input device 700. Set. And the number of positive examples of that line is 130
And “1” and “0” are set in the negative example number 131, respectively. Further, "necessary" is set in the column of necessity 150 in that row. Then, the rule corresponding to this line is set to the knowledge base 5
Add to 00. The above is the details of the processing of the processing device 200.

According to this embodiment, in addition to the same effect as that of the fourth embodiment of the second invention, since the discriminating condition is a rule, it is possible to cooperate with the existing rule type system and enable high-speed inference. Reasoning software is available.

[0132]

As described in detail above, according to the first or second invention, since all combinations of attribute values are not expanded, the storage capacity required for the storage device is small, and the size is small. This has the effect that it can be used on a computer. In addition, after processing all the case data, the most general discriminant condition is generated among the remaining combinations, so learning can be completed even if the case data is small, and It has the advantage that it can be used effectively. In addition to the above effects, according to the second aspect of the present invention, it is possible to correct the discrimination condition based on the evaluation of the discrimination result during operation, which can contribute to efficient discrimination processing on site.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing a first embodiment of the first invention.

FIG. 2 is a configuration diagram showing a first embodiment of the first invention.

FIG. 3 is a configuration diagram showing a second embodiment of the first invention.

FIG. 4 is a configuration diagram showing a third embodiment of the first invention.

FIG. 5 is a configuration diagram showing a fourth embodiment of the first invention.

FIG. 6 is a configuration diagram showing a fourth embodiment of the first invention.

FIG. 7 is a configuration diagram showing a fifth embodiment of the first invention.

FIG. 8 is a configuration diagram showing a fifth embodiment of the first invention.

FIG. 9 is a configuration diagram showing a sixth embodiment of the first invention.

FIG. 10 is an explanatory diagram of case data according to the sixth embodiment of the first invention.

FIG. 11 is an explanatory diagram of an attribute value definition unit in the sixth embodiment of the first invention.

FIG. 12 is an explanatory diagram of case data according to the sixth embodiment of the first invention.

FIG. 13 is a configuration diagram showing a seventh embodiment of the first invention.

FIG. 14 is a configuration diagram showing a seventh embodiment of the first invention.

FIG. 15 is a configuration diagram showing an eighth embodiment of the first invention.

FIG. 16 is a configuration diagram showing an eighth embodiment of the first invention.

FIG. 17 is an explanatory diagram of a display example according to the eighth embodiment of the first invention.

FIG. 18 is a configuration diagram showing a ninth embodiment of the first invention.

FIG. 19 is a configuration diagram showing a ninth embodiment of the first invention.

FIG. 20 is an explanatory diagram of a display example in the ninth embodiment of the first invention.

FIG. 21 is a configuration diagram showing a tenth embodiment of the first invention.

FIG. 22 is a configuration diagram showing a tenth embodiment of the first invention.

FIG. 23 is an explanatory diagram of a display example according to the tenth embodiment of the first invention.

FIG. 24 is a configuration diagram showing an eleventh embodiment of the first invention.

FIG. 25 is a configuration diagram showing an eleventh embodiment of the first invention.

FIG. 26 is an explanatory diagram of a display example in the eleventh embodiment of the first invention.

FIG. 27 is a configuration diagram showing a twelfth embodiment of the first invention.

FIG. 28 is a configuration diagram showing a twelfth embodiment of the first invention.

FIG. 29 is an explanatory diagram of a display example in the twelfth embodiment of the first invention.

FIG. 30 is a configuration diagram showing a first embodiment of the second invention.

FIG. 31 is an explanatory diagram of a storage device according to the first embodiment of the second invention.

FIG. 32 is an explanatory diagram of a storage device according to the first embodiment of the second invention.

FIG. 33 is a configuration diagram showing a second embodiment of the second invention.

FIG. 34 is a configuration diagram showing a third embodiment of the second invention.

FIG. 35 is an explanatory diagram of a storage device according to a third embodiment of the second invention.

FIG. 36 is an explanatory diagram of a knowledge base in the third embodiment of the second invention.

FIG. 37 is an explanatory diagram of a storage device according to a third embodiment of the second invention.

FIG. 38 is an explanatory diagram of a knowledge base according to the third embodiment of the second invention.

FIG. 39 is a configuration diagram showing a fourth embodiment of the second invention.

FIG. 40 is an explanatory diagram of a storage device according to the fourth embodiment of the second invention.

FIG. 41 is an explanatory diagram of an attribute value defining unit in the fourth embodiment of the second invention.

FIG. 42 is a configuration diagram showing a fifth embodiment of the second invention.

FIG. 43 is a configuration diagram showing a sixth embodiment of the second invention.

[Explanation of symbols]

 100 Storage Device 200 Processing Device 300 Case Data 350 Use Attribute Designation Unit 400 Attribute Value Definition Unit 500 Knowledge Base 600 Output Device 700 Input Device 800 Analysis Result Storage Unit

Claims (18)

[Claims] 1. Attribute values for a plurality of types of attributes,
There are multiple cases that consist of discrimination results for these attribute values, and at least one case has a positive discrimination result, case data that combines the same attributes so that they do not overlap, and these attributes Corresponding to the combination of values, the number of positive examples as a value for determining whether the case is a positive determination result (hereinafter referred to as positive example) or the combination of the attributes of the positive case, and The number of negative examples as the number of negative examples (hereinafter referred to as negative examples), the storage device in which the necessity of necessity of the combination of the attribute values is set, and the initial stage of the storage device based on the case data. Processing, which sets the number of positive cases and negative cases, and the setting of necessity, and sets the combination of attribute values and necessity in the storage device to “blank” by the processing apparatus, and sets the number of positive cases and Negative number After initializing the storage device by setting "0", all combinations of a plurality of attribute values in the positive example in the case data are set in the storage device and the number of positive examples corresponding to the positive example in the case data is set to " 1 ”is set, a combination of a plurality of attribute values in the negative example in the case data is set in the storage device, and for each combination of the attribute values in the storage device, each combination is set as a negative example. If it exists, the number of negative examples is set by accumulating the number of negative examples by one, and it corresponds to the combination having the smallest number of attributes among the combinations of the attribute values for which the number of negative examples in the storage device is “0”. “Necessity” is set for the necessity, and a combination of attribute values for which necessity is set as “necessary” by the above processing is included among combinations of attribute values for which the number of negative examples in the storage device is “0”. Supports all attribute value combinations By setting "No" to the necessity, the combination of attribute values included in the positive example and not included in the negative example is generated as the determination condition for the positive example and the negative example. Method. 2. The discrimination condition generating method according to claim 1, wherein the case data has a plurality of positive examples. 3. The discrimination condition generating method according to claim 1, wherein the attribute value of the case data has a defect. 4. The determination condition generating method according to claim 1, wherein the processing device determines whether or not the content is required in the storage device.
A discriminant condition generating method characterized in that a rule having a conclusion part as "correct" is generated as a knowledge base for a combination of attribute values. 5. The method according to claim 1, wherein the processing device generates a combination of attribute values in the storage device using only the attributes designated by the use attribute designating section. Condition generation method. 6. The discrimination condition generating method according to claim 1, wherein the processing device replaces data having continuous values in the case data with an attribute value according to a satisfaction condition of an attribute value definition unit, and uses this attribute value as a case. A discrimination condition generation method characterized by being set in data. 7. The discrimination condition generating method according to claim 1, wherein the processing device causes the output device to display the combination of attribute values and the number of positive cases in which the necessity is “necessary” among the contents of the storage device. A method for generating a distinction condition. 8. The discrimination condition generating method according to claim 1, wherein the case data has a case number of a serial number attached to each case, and the processing device sets a combination of attribute values in the storage device. , A case number corresponding to the discrimination result in the case data is set in the storage device as a positive example number and a negative example number, and a contradiction included in the case data detected from the positive example number and the negative example number is output to the output device. A method for generating a discrimination condition, characterized by displaying by. 9. The determination condition generating method according to claim 1, wherein the storage device has a serial number combination number for each combination of attribute values, and the processing device determines whether or not the contents of the storage device are required. And the combination number corresponding to the combination of the attribute values having the smallest number of attribute values and the part having the attribute value number which is one less than the number of the attribute values of the combination of the attribute values, which is required, is common. The combination number corresponding to the combination of the attribute values is set in the analysis result storage unit as a positive combination number, and the combination of the attribute values common to all the combination of the attribute values related to these positive combination numbers corresponds to the combination. The combination number is set as the target combination number in the analysis result storage unit, and based on the contents of the analysis result storage unit, the rule having only the attribute value common to the condition unit and the attribute value common to the condition unit are included. The output device A method for generating a discrimination condition, which is characterized by displaying it on a display. 10. The discrimination condition generating method according to claim 9, wherein the case data has a case number of a serial number given to each case, and the storage device has a discrimination result of “negative” in the case data. The processing device has a case number as a negative example number, and the processing device includes a combination of the attribute values of the storage device indicated by the target combination number, and the combination number of which the negative example number of the storage device is not “0” is a negative combination number. Based on the contents of the analysis result storage unit, a rule having only attribute values common to the condition unit, a rule group including attribute values common to the condition unit, and a common rule A method for generating a discrimination condition, characterized in that the output device displays the case number that inhibits the conversion. 11. The discrimination condition generating method according to claim 1, wherein the case data has a test identification for identifying whether or not to use for a test for each case, and the processing device stores the case data. The contents of the storage device are generated by using the combination of attribute values whose test identification is not “necessary”, and the combination of the attribute values whose test identification of the case data is “necessary” and the necessity in the storage device are “necessary”. The number of successes, the number of false wins and the number of false wins determined in accordance with the inclusion relationship between the two and the determination result of the case data are set in the storage device. Based on the number and the number of unsatisfied cases, the number of cases of success, failure, and failure is displayed on the output device when a combination of attribute values whose necessity in the storage device is “necessary” is used as a determination condition. A method for generating a judgment condition. 12. The discrimination condition generation method according to claim 11, wherein the case data has a case number of a serial number attached to each case, and the processing device corrects the non-establishment when the non-establishment number and the false establishment number are detected. The case number of the example and the case number of the false negative example are set in the storage device as the positive positive example number and the false negative example number, respectively, and the positive positive example number and the false negative example are established together with the number of cases of the valid, false, and false cases. A method for generating a discrimination condition, wherein the number is displayed on an output device. 13. A storage device generated by the discriminating condition generation method according to claim 1, an input device to which an attribute value is input, an output device to which an inference result is output, and an attribute value input from the input device. And a processor for correcting the contents of the storage device according to the evaluation of the inference result and performing the inference by comparing the contents of the storage device with the contents of the storage device. It is compared with a combination of attribute values whose necessity is "necessary", and if the attribute values input from the input device include at least this combination of attribute values, it is called "positive", and if not, it is called "negative". The inference result is output to the output device, and when the above inference result is evaluated as correct, no processing is performed, and the above inference result is evaluated as incorrect, and another conclusion “negative” is obtained from the input device. When entered, The attribute value stored in the input device is set in the attribute value combination field added in the storage device, and the number of positive cases is set to "0", the number of negative cases is set to "1", and "blank" is set for necessity. Among the combinations of the attribute values that have been previously set in the storage device, the number of negative examples corresponding to those included in the content of the newly added attribute value combination is increased by “1”, and whether or not it is necessary If "necessary", this is set to "blank", and then the storage device is searched, and the content of the combination of attribute values whose necessity is "No" is any other attribute whose necessity is "Necessary". When the content of the value combination is not included, the necessity is changed from “No” to “Necessary”, and the inference result is evaluated as incorrect, and when another conclusion “correct” is input from the input device, Set the attribute value saved in the input device in the attribute value combination field added in the storage device. At the same time, the number of positive cases is set to “1”, the number of negative cases is set to “0”, and the necessity is set to “Necessary”, and the judgment conditions of the positive example and the negative example are generated by correcting the contents of the storage device. Discriminating condition generating method. 14. The method according to claim 13, wherein the processing device determines the attribute value stored in the input device and the inference result output to the output device or the inference result input from the input device. A method for generating a discrimination condition, characterized in that a plurality of cases consisting of attribute values for a plurality of types of attributes and discrimination results for these attribute values are stored in case data that can be stored. 15. The discrimination condition generating method according to claim 13, wherein the processing device infers the attribute value input from the input device according to the contents of the knowledge base according to claim 4, and inputs an evaluation of the inference result. A discriminating condition generating method characterized by changing the contents of a storage device and a knowledge base while inputting from a device. 16. The method according to claim 13, wherein the processing device converts a continuous value into an attribute value for the attribute value input from the input device according to the contents of the attribute value definition unit according to claim 6. A determination condition generating method characterized in that inference is performed according to the contents of a storage device, an inference result is output to an output device, and an evaluation for this inference result is input from an input device to change the contents of the storage device. 17. The discrimination condition generating method according to claim 16, wherein the processing device sets the attribute value stored in the input device and the inference result output to the output device or the inference result input by the input device. A method for generating a discrimination condition, characterized in that a plurality of cases consisting of attribute values for a plurality of types of attributes and discrimination results for these attribute values are stored in case data that can be stored. 18. The discrimination condition generating method according to claim 16, wherein the processing device infers the attribute value input from the input device according to the contents of the knowledge base according to claim 4, and inputs an evaluation of the inference result. A discriminating condition generating method characterized by changing the contents of a storage device and a knowledge base while inputting from a device.