JPH04348432A - Deciding condition generating method - Google Patents

Abstract

PURPOSE:To enable an expert to easily generate the conditions apporoximate to the unchanged ones and also to easily refine the conditions despite the change of the attribute to be considered in a deciding state by applying the result data on a site to the fault diagnoses, etc., as the instances and at the same time to generate the partial conditions in a fixed time even with a compact computer by designating the attribute to be considered. CONSTITUTION:The least total number of attributes to be secured with the correct deciding result is obtained based on an instance base 10. Then the combination of attributes of specific types less than the least total number is prescrived as the indispensable attributes. The pairs of the least total number and the indispensable attributes are listed in a storage 20, and a processor 30 generates the deciding conditions that secure the attributes of specific types based on the enclosure relation among pairs and the contents of the pair specified by an instance. Based on the deciding conditions, all instances having the correct deciding results are decided. If some instances are not decided correct with the desiding conditions, the deciding condition generation processing and the deciding processing are repeated until those instances are deleted.

Description

[Detailed description of the invention]

0001

[Field of Industrial Application] The present invention relates to a method for generating discriminant conditions, and more specifically, the present invention relates to a method for generating discriminant conditions.
Whether or not certain conditions are met regarding specific numerical values etc. that indicate the attributes) is given, whether the success or failure of these attributes is compatible with the purpose of the equipment (positive example)
The present invention relates to a method of generating a discrimination condition for determining a negative example) based on a case constituted by success/failure of an attribute and a discrimination result, and relates to a discrimination condition generation method applied to, for example, a plant failure diagnosis system.

0002

[Prior Art] Conventionally, this type of discrimination condition has been used to be built into various types of equipment, receive status signals from each part of the equipment, and use these status signals to determine whether or not there is a malfunction in the equipment. has been done. Here, even if the equipment is the same, there are differences in how failures occur depending on the installation environment and operating method. Judgment conditions cannot be set. Therefore, ideally, the cause of the failure is known to the experts maintaining the equipment, and
It is desirable to individually set discrimination conditions for each piece of equipment based on the way the phenomena occur when a failure occurs, and to incorporate them into the equipment.

Conventionally, in order to create such discrimination conditions based on the know-how of experts, attributes to be focused on during discrimination (such as status signals from each part of equipment) are determined, and specific numerical values representing these attributes are determined. Experts were required to sort out the correlation between the success/failure of attributes and the causes of failure in all cases, such as whether certain conditions were met or not. For this reason, there have been cases in which experts have created incorrect discrimination conditions due to a misunderstanding, or there have been cases in which there have been excesses or deficiencies in the types of attributes or failure cases due to insufficient consideration.

[0004] In addition, in order to verify the discrimination conditions created in this way, the success or failure of the attribute is input, the discrimination results based on the discrimination conditions are compared with the knowledge of experts, and if there is an error, the preset discrimination conditions are corrected. It was necessary to do so. In this case, since the expert corrects the discrimination conditions for each faulty case individually, he or she is unaware of the interrelationship of the discrimination conditions, resulting in endless repetition of correction work.

[0005] In order to solve these problems, several methods have been proposed to automate the modification of discriminant conditions (for example, Ginsberg, A., Weiss,
S. and Politakis, P. , “SEEK
2:A Generalized Approac.
hto Automatic Knowledge
BaseRefinement" Proc.o
f IJCAI-85, pp. 367-374,19
85). These methods modify the most effective discrimination condition after examining the correlation of all discrimination conditions for a failure case in which an error occurred. However, in order for these methods to work efficiently, it is necessary that the target discrimination conditions are well-organized in advance. For this reason, the burden on experts who first organize the discrimination conditions remained heavy.

[0006] In order to reduce this burden, several methods have been proposed in the past for automatically generating discriminant conditions based on cases (for example, Quinlan, J.R.
．． , “Induction of Decision
“Trees”, Machine Learning
1(1), Kluwer, pp. 81-106,1
986). These methods input a large number of cases consisting of success/failure of several attributes and judgment results by experts, and generate judgment conditions for judging those cases without error.

[0007]

[Problem to be solved by the invention] These methods assume that each input example has success or failure with respect to all attributes. It is desirable to be able to use the actual data accumulated over the years. However, in general, performance data often records only the data necessary for discrimination, and does not necessarily have information on the success or failure of all attributes. For this reason, the conventional discrimination condition generation method has the disadvantage that it is not possible to use actual data accumulated in the field as a case study.

Furthermore, since the case data does not necessarily cover all discrimination cases, the discrimination conditions generated from this case data cannot be said to be sufficient. Therefore, experts investigate the generated discrimination conditions, add and modify the case data, change the attributes to be considered, generate the discrimination conditions again, and refine the discrimination conditions through repeated trial and error. It turns out. However, when the case data and the attributes to be considered are changed in this way, the generated discrimination conditions become completely different from those before the change. Therefore, in the past, there was a drawback that it was difficult for experts to refine the discrimination conditions step by step and continuously. Further, since the conventional determination condition generation method is a batch processing method, the processing time and the computer resources used cannot be controlled from the outside. For this reason, there have been cases where it has not been possible to generate discrimination conditions within a certain amount of time using a small computer.

[0009] The present invention was made to solve the above problems, and its purpose is to use as an example even actual field performance data that does not necessarily have success/failure information for all attributes. Moreover, even if case data or attributes to be considered during discrimination are changed through trial and error, experts can generate discrimination conditions that are close to the discrimination conditions before the change and proceed with step-by-step and continuous refinement of the discrimination conditions. It is an object of the present invention to provide a method for generating discriminant conditions that is easy to use, and also allows a small computer to generate part of the discriminant conditions within a certain amount of time by specifying attributes to be considered.

0010

[Means for Solving the Problems] In order to achieve the above object, the first invention provides a method for identifying attributes that should hold if the determination result is correct, based on cases consisting of success or failure of attributes and determination results based on the success or failure of the attributes. Find the minimum total number, define combinations of specific types of attributes that are less than or equal to this minimum total number as attributes that must be included, list the sets consisting of the minimum total number and the attributes that must be included, and specify the combination between these sets. From the content of the set specified based on the inclusion relationship of and the above-mentioned cases, a discrimination condition is generated that states that attributes of a specific type are included and the total number of attributes is greater than or equal to the minimum total number, and based on this discrimination condition, among the cases All cases in which the discrimination result was determined to be correct are judged, and as a result, if there remain cases among the cases that are not judged to be correct according to the above discrimination conditions, this remaining case is The above-described determination condition generation process and determination process based on the generated determination conditions are repeatedly executed to generate a required number of determination conditions.

[0011] The second invention calculates the minimum total number of attributes specified by the attribute specifying section that should be satisfied if the judgment result is correct, from the cases constituted by the success or failure of the attributes and the judgment results based on the success or failure of the attributes. and specifying a combination of specific types of attributes that are equal to or less than this minimum total number and specified by the attribute specification unit as attributes that must be included, and enumerating sets consisting of the minimum total number and the attributes that must be included, The above-mentioned discrimination condition is generated from the contents of the set specified based on the inclusion relationship between these sets and the above-mentioned cases, and all the cases for which the discrimination result is correct among the above-mentioned cases are discriminated by this discrimination condition, and the As a result, if there are cases remaining that are not determined as correct according to the above discrimination conditions, the remaining cases are subjected to the above discrimination condition generation processing and the generated discrimination conditions until there are no remaining cases. The required number of discrimination conditions is generated by repeatedly executing the discrimination process.

[0012] The third invention calculates the minimum total number of attributes that should hold true if the determination result is correct from the cases constituted by the success or failure of the attributes and the determination results based on the success or failure of the attributes, and this minimum total number is stored in the attribute number specification section. A combination of specific types of attributes that is less than or equal to the minimum total number when it is smaller than the maximum value specified by is defined as attributes that must be included, and
Enumerate sets consisting of the minimum total number and attributes that must be included, generate the discrimination condition from the content of the set specified based on the inclusion relationship between these sets and the case, and use this discrimination condition to All cases in which the discrimination result was determined to be correct are judged, and as a result, if there remain cases among the cases that are not judged to be correct according to the above discrimination conditions, this process is continued until there are no remaining cases. The above-described discriminant condition generation process and discrimination process using the generated discriminant conditions are repeatedly executed for the remaining cases to generate a required number of discriminant conditions.

[0013] The fourth invention calculates the minimum total number of attributes that should hold true if the determination result is correct from cases consisting of the success or failure of attributes and the determination results based on the success or failure of the attributes, and selects a specific type that is less than or equal to this minimum total number. Among the combinations of attributes, a specific type of combination specified by the attribute combination specification section is specified as an attribute that must be included, and sets consisting of the minimum total number and attributes that must be included are listed, and these sets are From the content of the set specified based on the inclusion relationship between and the above-mentioned case, a judgment condition is generated that the specified type of attribute is included and the above-mentioned minimum total number or more of attributes hold true, and based on this judgment condition, the above-mentioned case As a result, if there are cases remaining among the cases that are not judged to be correct according to the above judgment conditions, this remaining case is In this case, the above-mentioned discriminant condition generation process and the discriminant process based on the generated discriminant conditions are repeatedly executed to generate a required number of discriminant conditions.

[0014]

[Operation] In the first invention, when a case consisting of the success or failure of an attribute and a judgment result by an expert is given, a List all the ``groups'' and their ``
Among the above cases, based on the inclusive relationship between
Find a ``set'' that includes all cases that are determined to be ``negative'' and does not include any cases that are determined to be ``negative.''
According to the content of , the determination condition is that the attributes of a specific type are included and the minimum total number or more of attributes are satisfied. Then, all cases in which the judgment result by the expert is "correct" are judged according to the above judgment conditions, and as a result, if there remain cases among the cases that are not judged as correct according to the above judgment conditions, this Until there are no remaining cases, the above-described discriminant condition generation process and the discriminant process based on the generated discriminant conditions are repeatedly executed for the remaining cases to generate a required number of discriminant conditions.

In the second invention, rather than considering all attributes in a given case, almost the same processing as in the first invention is performed on only the attributes specified by the attribute specification section, and the discrimination condition is generate.

[0016] In the third invention, the maximum value of the minimum total number of attributes that should be satisfied at the time of determination is specified by the attribute number designation section, only "sets" having a minimum total number smaller than this maximum value are determined, and these Processing substantially similar to that of the first invention is performed on the "set" to generate discrimination conditions.

[0017] In the fourth invention, all "sets" consisting of combinations specified by the attribute combination specifying section as the minimum total number of attributes that should be established at the time of discrimination and combinations of attributes that must be included therein are obtained, and these Processing substantially similar to that of the first invention is performed on the "set" to generate a discrimination condition.

[0018]

Embodiments Hereinafter, embodiments of the present invention will be explained with reference to the drawings. FIG. 1 shows a first embodiment of the first invention. In the figure, 10 is the case base, and this case base 1
0 is the attribute column 111, 112, . . . in which the success or failure (satisfaction or failure) of an attribute indicating whether a certain condition is satisfied for specific numerical values etc. regarding multiple types of attributes is stored.
..., 11n, and a determination result column 12 in which the determination results corresponding to the success or failure of the attribute are stored. The result of determining whether the determination result is positive) or not (also negative) is stored as an example. More specifically, the success or failure of each of all or some of the n attributes listed in the attribute columns 111 to 11n is stored in the horizontal direction of FIG. 1, and the determination result column corresponding to these success or failure is stored. Together with the 12 discrimination results, one horizontal row constitutes one case, and a plurality of these cases together constitute the case base 10. Note that each case does not necessarily need to store success/failure information for all n attributes.

Next, the storage device 20 stores a column 21 in which the minimum total number of attributes that must be satisfied at the time of determination, a column 22 in which attributes that must be included among those attributes are stored, and
It consists of a column 23 in which the necessity of a "group" consisting of these columns 21 and 22 is stored. Furthermore, the processing device 30
As explained below, a case is input from the case base 10, a "set" consisting of the minimum total number of attributes that should be satisfied and attributes that must be included in the storage device 20 is generated, and the necessity of the set is determined. However, the determination conditions are generated from the final contents of the storage device 20 and stored in the output device 40.

The processing in the processing device 30 will be explained in detail below. (1) First, the contents of the output device 40 are initialized to be empty. ■ Check the number of established attributes for each case marked as "correct" in the discrimination result column 12 in the case base 10,
Find its minimum value (referred to as mmin). Note that in each of the attribute columns 111 to 11n, "established" is written only when the attribute is established. When these columns 111 to 11n are blank, it indicates that the corresponding attribute has not been established or that the success or failure is unknown.

(2) All "sets" whose minimum total number of attributes to be satisfied is mmin determined in (2) above are generated in the storage device 20, and the necessity column 23 is initialized to "undetermined". For example, there are four attributes (A, B, C, D) in the case base 10.
) is defined, a "set" for which the minimum total number of attributes that should hold is mmin is one in which mmin or more attributes out of the four attributes hold true.

Furthermore, a constraint condition is set that a specific attribute must be included, rather than a combination of mmin or more arbitrary attributes among the four attributes. For example, four attributes (A, B,
FIG. 2 shows the contents of the initialized storage device 20 when mmin=2 for C and D). In this figure, the column for attributes that must be included in the first "set" is blank,
This indicates that at least two of the four attributes need to be satisfied. Also, the second “group” is attribute A
, and at least one arbitrary one of the remaining three attributes (B, C, D) should be satisfied, and two or more of the remaining three attributes (B, C, D) should be satisfied. Furthermore, the sixth "set" always includes attribute A and attribute B, and the remaining two attributes (C,
D) indicates that it may or may not hold true. As is already clear, the number of attributes listed in the column of attributes 22 that must be included in FIG. 2 is always less than or equal to the minimum total number.

■ Among the positive examples in the case base 10 (cases marked as "correct" in the discrimination result column 12), all positive examples in which the number of established attributes is equal to mmin are found and stored in the storage device 20.
Column 2 of attributes that must be included in the "group" expanded in
The column 23 for whether or not the "group" whose content in item 2 completely matches the attribute established in the above positive example is changed from "undetermined" to "required". For example, if mmin=2, in the case base 10,
If only attributes A and B have positive examples marked as “true,” the number of attributes that hold true is two, and this must be included in the sixth “set” in Figure 2. Since it completely matches the contents of the attribute column 22, the sixth “group” is required or not.
3 is changed to "essential".

■ Among the negative examples in the case base 10 (cases marked as "negative" in the discrimination result column 12), find all negative examples in which the number of established attributes is equal to or larger than mmin, The contents of the column 22 of the attributes that must be included in the "set" developed in the storage device 20 are changed to the column 23 of the necessity of the "set" that is completely included in the attributes established in the above negative example.
Change from "Undecided" to "No." For example, when mmin = 2, if there is a negative example in which only attributes C and D are marked as "true" in the case base 10, the number of attributes that are true is two, and Column 22 of attributes that must be included in the 1st, 4th, 5th, and 11th "groups" of
The contents of are completely included (φ⊂{CD}, {C
}⊂{CD}, {D}⊂{CD}, {CD}⊂{CD}
). Therefore, in the column 23 for whether or not these four "groups" are necessary, "
It will be changed to "No".

Furthermore, if there is a negative example in which only attributes B, C, and D are marked as "true" in the case base 10, the number of attributes that are true is three, which is greater than mmin=2,
1st, 3rd, 4th, 5th in Figure 2,
The contents of the columns 22 of attributes that must be included in the 9th, 10th, and 11th "sets" are included in the valid attributes of the negative examples (φ⊂{BCD}, {B}⊂{BCD }, {C
}⊂{BCD}, {D}⊂{BCD}, {BC}⊂{B
CD}, {BD}⊂{BCD}, {CD}⊂{BCD}
). Therefore, the column 23 indicating whether or not these seven "groups" are necessary is changed to "no".

(2) The "set" in the storage device 20 has a partial order relationship determined by the contents of the column 22 of attributes that must be included. For example, the "group" whose content is {A} in the column 22 is included in the "group" of {AB}, but is not included in the "group" of {BC}. Regarding the contents of the storage device 20 after the processing of ■ and ■ above,
If there is a “group” that does not have “No” written in the necessity column 23 and does not include other “groups” that have “required” or “undetermined” written in the necessity column 23, All of them are selected and added to the output device 40 as discrimination conditions.

For example, when mmin=2, FIG.
Shown below. The second “group” in Figure 3 is the necessity column 23
is not marked as “no,” and the sixth item (“required”),
Since it does not include either the 7th ("undetermined") or the 8th ("undetermined") "set," the determination condition is "must include attribute A, and 2 or more attributes are true in total." It is added to the output device 40 as a. In addition, although the necessity column 23 of the 6th, 7th, and 8th "groups" does not indicate "no", they each include the 2nd ("undetermined") "group". , is not selected as a discrimination condition.

(2) Among the positive examples in the case base 10, the determination result column 12 is changed to "completed" for all those that are correctly determined according to the determination conditions output to the output device 40. Thereafter, if a positive example whose determination result column 12 is "correct" remains in the case base 10, the process returns to step (2) above and the same process is performed thereafter. In case base 10, judgment result column 12
The entire process ends when there are no positive examples where is "correct". The above is an explanation of the processing in the processing device 30 of FIG.

Next, as a second embodiment which is a modification of the first embodiment, there is one in which the process (2) is changed as shown in (a) below. ■a Regarding the contents of the storage device 20 after processing the above-mentioned ■ and 23 will be changed from “undecided” to “required”. After completing all the above changes, there will be a "group" that is marked "required" in the necessity column 23 and does not include other "groups" marked as "required" in the necessity column 23. If so, select all of them and use the output device 4 as a judgment condition.
Add each to 0. For example, in Figure 3, the second "
Group" is the 7th and 8th group whose necessity column 23 is "Undetermined".
Since it is included in the th "group", it is not changed to "required". Therefore, the sixth "group" was selected, and "Attribute A and Attribute B
must include, and two or more attributes are established as a whole.”
This determination condition is added to the output device 40.

In the first embodiment described above, a discrimination condition is generated that does not include negative examples in the case base 10 but also includes cases that are not given as positive examples in the case base 10. Therefore, it can be said that this method is effective when the case base 10 stores many negative examples but few positive examples. In contrast, in the second embodiment described above, a discrimination condition is generated such that the case base 10 includes only cases given as positive examples. Therefore, it can be said that this method is effective when the case base 10 stores many positive examples but few negative examples.

Next, a third embodiment of the present invention will be described. In this embodiment, the process (2) in the first embodiment is replaced by the following (1).
This is modified as shown in b. ■b Regarding the contents of the storage device 20 after processing the above-mentioned ■ and "Undecided"
to "required". After completing all of the above changes, if "Required" is written in the Necessity column 23, and if the Necessity column 23
If there is a "group" that does not include other "groups" marked as "required", all of them are selected and added to the output device 40 as determination conditions.

For example, in the example of FIG. 3, the second "set" is included in the sixth "set" whose necessity column 23 is "required", so the second "set" is No column 23 is “required”
will be changed to In this state, the sixth "group" is written as "required" in the necessity column 23;
is not selected because it contains the second "group" which is "key", and conversely, the second "group" does not include the sixth "group", so this second "group" is not selected. group” is selected. Therefore, it is added to the output device 40 as a determination condition that "attribute A must be included and two or more attributes are satisfied in total." According to this embodiment, the case base 10 includes, for example, 1
positive examples, as if there were only two positive or negative examples,
This is effective when there are few negative examples.

Furthermore, as a fourth embodiment of the present invention, it is conceivable that cases of a plurality of purposes are stored in the case base 10. At this time, the case base 10 is composed of one or more attribute columns and discrimination result columns. In the first to third embodiments, the determination result column indicates whether or not the case is compatible with a single purpose, but in this fourth embodiment, the name of the purpose to which the case is compatible is recorded. There is. For example, in a case base related to equipment failure diagnosis, phenomena discovered during failure diagnosis are stored as each attribute, and causes of failure revealed as a result of diagnosis are stored as determination results.

In this fourth embodiment, the following processing is added as pre-processing to the processing ① to ② in the first embodiment. (2) A new case base 10 is generated for each of the purpose names written in the column of case-based discrimination results, and the above-mentioned processes (1) to (4) are performed for each case base 10. For example, given a case-based discrimination result column, three objective names (X, Y
, Z) has appeared. The case base 10 for purpose name For cases marked with Y or Z), change the judgment result column 12 to “
This can be obtained by setting it to "negative". According to this example,
It has the advantage of being able to handle cases where multiple objectives are managed on a single case basis.

Next, FIG. 4 shows an embodiment of the second invention. In this embodiment, the configurations of the case base 10, the storage device 20, the processing device 30, and the output device 40 are the same as in each embodiment of the first invention, but an attribute specifying unit 5
0 is added. This attribute specifying section 50 is provided to instruct that the processing after the above-mentioned (2) be performed in consideration of only some of the attributes 1 to n defined in the case base 10, rather than all of them. The processing in the processing device 30 is similar to each embodiment of the first invention, except that the target attributes are not all the attributes in the case base 10 but are limited to the attributes specified by the attribute specification unit 50. are the same.

As mentioned above, in general, experts often refine the discrimination conditions while adjusting the attributes to be considered. Therefore, according to this embodiment, there is an advantage that the case base 10 does not have to be recreated every time. Furthermore, considering many attributes requires a large amount of processing, and for example, with a small computer, it may not be possible to determine the discrimination conditions within a certain amount of time. According to this embodiment, since the number of attributes to be considered can be adjusted by the attribute specifying section 50, there is an advantage that a part of the discrimination conditions can be obtained within a certain amount of time even with a small computer.

Next, FIG. 5 shows an embodiment of the third invention. In this example, case base 10,
The configurations of the storage device 20, processing device 30, and output device 40 are the same as those in each embodiment of the first invention, but an attribute number designation section 60 is additionally added. The number of attributes specifying unit 60 does not generate a condition for discriminating all positive examples stored in the case base 10, but only determines a discriminating condition in which the minimum total number of attributes that should be satisfied is less than or equal to a certain value. It is provided for the purpose of giving instructions.

In this embodiment, the other processes are the same as those in the first embodiment of the first invention, except that the process (2) in the processing device 30 is changed as shown in (a) below. (a) Check the number of established attributes for each case marked as "correct" in the discrimination result column 12 in the case base, and find the minimum value (referred to as mmin). If, mmin
If the value is larger than the value set in the number of attributes designation unit 60, the entire process ends. Generally, the above-mentioned mmin becomes a large value and the amount of processing becomes large, and for example, in a small computer, the discrimination condition may not be required. Therefore,
According to this embodiment, the size of mmin can be adjusted, so even a small computer can obtain some discrimination conditions within a certain amount of time for cases consisting of many attributes. be.

Next, FIG. 6 shows an embodiment of the fourth invention. In this example, case base 10,
The configurations of the storage device 20 and the output device 40 are the same as in each embodiment of the first invention, and an attribute combination designation section 70 is newly added. The attribute combination designation unit 70 stores combinations of attributes that can be established at the same time or combinations of attributes that cannot be established at the same time, and designates these combinations to the processing device 30. For example, if you assign water temperature data below 40 degrees to the attribute "low water temperature" and 40 degrees or more to the attribute "high water temperature," the combination of "low water temperature" and "high water temperature" that cannot exist at the same time is assigned to the attribute. It is stored in the combination designation section 70.

In this embodiment, processing device 30 inputs cases from case base 10 as described below, and
The minimum total number of attributes that should exist in the storage device 20 and the attribute combination designation unit 7 as a combination of attributes that should definitely be included.
A "group" consisting of the combination specified by 0 is generated, the necessity of the combination is determined, and a determination condition is generated from the final contents of the storage device 20 and stored in the output device 40.

The processing in the processing device 30 will be explained in detail below. (1) First, the contents of the output device 40 are initialized to be empty. ■ Check the number of established attributes for each case marked as "correct" in the discrimination result column 12 in the case base 10,
Find the minimum value mmin. In addition, attribute columns 111-1
1n is marked as "established" only when the attribute is established. When these columns 111 to 11n are blank, it indicates that the corresponding attribute has not been established or that the success or failure is unknown.

■ Generate in the storage device 20 all "sets" of attribute combinations in which the minimum total number of attributes to be established is mmin determined in the above (■) and specified by the attribute combination designation unit 70; The necessity column 23 is initialized to "undetermined". For example, there are four attributes (A, B, C,
D) is defined, and if attributes (A, D) are specified in the attribute combination designation section 70 as being impossible to hold simultaneously, then the minimum total number of attributes that should hold is m
The "set" of attribute combinations that is min and is specified by the attribute combination designation unit 70 is one in which mmin or more attributes that can be established simultaneously among the four attributes are established.

Furthermore, a constraint is set that a specific attribute must be included, rather than a combination of mmin or more of the four attributes that can be simultaneously established. For example, for four attributes (A, B, C, D), when mmin = 2 in a situation where attributes (A, D) cannot be established at the same time,
FIG. 7 shows the contents of the initialized storage device 20.

In this figure, the column for the attributes that must be included in the first "set" is blank, and among the four attributes,
This indicates that any given item only needs to hold at least two of attributes A and D without simultaneously containing them. In addition, the second "set" always includes attribute A, and at least one arbitrary one of the remaining two attributes (B, C) excluding attribute D, which cannot be satisfied at the same time as attribute A. , indicates that it is sufficient if two or more are satisfied together with A. Furthermore, the 6th “group”
indicates that the remaining attribute (C), which necessarily includes attribute A and attribute B and excludes attribute D, which cannot hold true at the same time as attribute A, may or may not hold true. As is already clear, the number of attributes listed in the column of attributes 22 that must be included in FIG. 7 is always less than or equal to the minimum total number.

■ Among the positive examples in the case base 10 (cases marked as "correct" in the discrimination result column 12), all positive examples in which the number of established attributes is equal to mmin are found and stored in the storage device 20.
Column 2 of attributes that must be included in the "group" expanded in
The column 23 for whether or not the "group" whose content in item 2 completely matches the attribute established in the above positive example is changed from "undetermined" to "required". For example, if mmin=2, in the case base 10,
If only attributes A and B have positive examples marked as “true,” the number of attributes that hold true is two, and this must be included in the sixth “set” in Figure 7. Since it completely matches the contents of the attribute column 22, the sixth “group” is required or not.
3 is changed to "essential".

[0046] Among the negative examples in the case base 10 (cases marked as "negative" in the discrimination result column 12), find all negative examples in which the number of established attributes is equal to or larger than mmin; The contents of the column 22 of the attributes that must be included in the "set" developed in the storage device 20 are changed to the column 23 of the necessity of the "set" that is completely included in the attributes established in the above negative example.
Change from "Undecided" to "No." For example, in the case of mmin=2, if there is a negative example in which only attributes C and D are marked as "true" in the case base 10, the number of attributes that are true is two, and as shown in FIG. Column 22 of attributes that must be included in the 1st, 4th, 5th, and 10th "groups" of
The contents of are completely included (φ⊂{CD}, {C
}⊂{CD}, {D}⊂{CD}, {CD}⊂{CD}
). Therefore, in the column 23 for whether or not these four "groups" are necessary, "
It will be changed to "No".

Furthermore, if there is a negative example in which only attribute B, attribute C, and attribute D are marked as "true" in the case base 10, the number of attributes that are true is three, which is greater than mmin=2,
1st, 3rd, 4th, 5th in Figure 7,
The contents of the columns 22 of attributes that must be included in the 8th, 9th, and 10th "sets" are included in the valid attributes of the negative examples (φ⊂{BCD}, {B}⊂{BCD }, {C}
⊂{BCD}, {D}⊂{BCD}, {BC}⊂{BC
D}, {BD}⊂{BCD}, {CD}⊂{BCD})
. Therefore, the column 23 for the necessity of these seven "groups" is "no".
will be changed to

(2) The "set" in the storage device 20 has a partial order relationship determined by the contents of the column 22 of attributes that must be included. For example, the "group" whose content is {A} in the column 22 is included in the "group" of {AB}, but is not included in the "group" of {BC}. Regarding the contents of the storage device 20 after the processing of ■ and ■ above,
If there is a “group” that does not have “No” written in the necessity column 23 and does not include other “groups” that have “required” or “undetermined” written in the necessity column 23, All of them are selected and added to the output device 40 as discrimination conditions.

For example, when mmin=2, FIG.
Shown below. The second “group” in FIG. 8 is the necessity column 23
is not marked as “no,” and the sixth item (“required”),
Since it does not include any of the seventh ("undetermined") "groups," it is added to the output device 40 as a determination condition that "attribute A must be included and two or more attributes are satisfied in total." Also, the 6th and 7th “groups” are in the necessity column 2.
3 is not "no", but the second ("undecided") "
group", so it is not selected as a discrimination condition.

(2) Among the positive examples in the case base 10, the determination result column 12 is changed to "completed" for all those that are correctly determined according to the determination conditions output to the output device 40. Thereafter, if a positive example whose determination result column 12 is "correct" remains in the case base 10, the process returns to step (2) above and the same process is performed thereafter. In case base 10, judgment result column 12
The entire process ends when there are no positive examples where is "correct". The above is an explanation of the processing in the processing device 30 of FIG. 6.

[0051] Also in this fourth invention, the second to fourth embodiments can be considered in the same manner as the first invention. That is, even in the present invention, in the second embodiment, processing
In the third embodiment, the process (2) is changed to the process (2) b, and in the fourth example, the process (2) is changed to the process (2) as a pretreatment for the processes (2) to (2) in the first embodiment. It is an addition. The processing contents and effects of these embodiments are such that the number of columns 22 and 23 (see FIGS. 7 and 8) of the storage device 20 is one more than each of the embodiments of the first invention (see FIGS. 2 and 3). Except for what is caused by the small size, the explanations are omitted because they are the same as the second to fourth embodiments of the first invention.

[0052]

[Effects of the Invention] As described above, according to the first invention, a method of rearranging a given case into a "set" consisting of the minimum total number of attributes that should hold during discrimination and the attributes that must be included therein. As a result, examples only need to be information on attributes that have been established, and the effect is that actual performance data from the field, which does not necessarily have information indicating the success or failure of all attributes, can be used as examples.

[0053] Furthermore, since we have adopted a method of narrowing down the discrimination conditions based on the minimum total number of attributes that should hold during discrimination and the partial order of the "sets" that must be included therein, even if the attributes to be considered are changed, Only related parts of a "set" are affected based on a partial order relationship. Therefore, in the discriminant conditions generated therefrom, relationships between attributes other than the changed attribute are preserved, thereby making it easier for experts to refine the discriminant conditions step by step and continuously.

According to the second invention, some attributes to be considered in the case are specified, so in addition to the effects of the first invention, when refining the discriminant conditions, It has the advantage of not having to re-prepare examples each time according to the attributes to be considered, and since the capacity of the storage device used for processing and the processing time can be controlled externally, even a small computer with low calculation capacity can be used for a certain amount of time. This has the effect of being able to generate part of the discrimination conditions within the system.

According to the third invention, the maximum value of the minimum total number of attributes that should be satisfied at the time of discrimination is specified, and only "sets" having a minimum total number smaller than that are determined.
In addition to the effects of the invention, the capacity of the storage device used in processing and the processing time can be controlled externally, so even a small computer with low computing power can solve the discriminant conditions within a certain amount of time for cases consisting of many attributes. This has the effect of being able to generate a portion of it.

[0056] The fourth invention is to rearrange the given cases into "sets" consisting of the minimum total number of attributes that should be established at the time of discrimination and the combination of specific types of attributes specified by the attribute combination specification section. Thereby, the same effect as the first invention can be obtained.

[Brief explanation of drawings]

FIG. 1 is a configuration diagram of a system to which an embodiment of the first invention is applied.

FIG. 2 is a diagram illustrating an example of a state after initialization of a storage device in the embodiment of FIG. 1;

FIG. 3 is a diagram illustrating an example of a processing state of the storage device in the embodiment of FIG. 1;

FIG. 4 is a configuration diagram of a system to which an embodiment of the second invention is applied.

FIG. 5 is a configuration diagram of a system to which a third embodiment of the invention is applied.

FIG. 6 is a configuration diagram of a system to which an embodiment of the fourth invention is applied.

FIG. 7 is a diagram illustrating an example of the state of the storage device after initialization in the embodiment of FIG. 6;

FIG. 8 is a diagram illustrating an example of a processing state of the storage device in the embodiment of FIG. 6;

[Explanation of symbols]

10 Case bases 111 to 11n Attribute column 12 Discrimination result column 20 Storage device 21 Minimum total number of attributes that should be established column 22 Attribute that must be included column 23 Necessity column 30 Processing device 40 Output device 50 Attribute specification section 60 Attribute Number specification section 70 Attribute combination specification section

Claims (4)

[Claims] [Claim 1] Find the minimum total number of attributes that should hold true if the determination result is correct from cases consisting of success/failure of attributes and determination results based on the success/failure of attributes, and find combinations of specific types of attributes that are less than or equal to this minimum total number. is specified as an attribute that must be included, and lists the sets consisting of the minimum total number and the attributes that must be included, and from the contents of the set specified based on the inclusion relationship between these sets and the example, A discrimination condition is generated that includes attributes of a specific type, and attributes that are equal to or greater than the minimum total number are satisfied, and based on this discrimination condition, all cases for which the discrimination result is correct among the cases are discriminated, and as a result, the relevant If there are cases remaining that are not determined as correct according to the above discrimination conditions, the above discrimination condition generation processing and discrimination processing using the generated discrimination conditions are performed on the remaining cases until there are no remaining cases. A discriminant condition generation method characterized by repeatedly executing the discriminant conditions to generate a required number of discriminant conditions. [Claim 2] Obtain the minimum total number of attributes that should hold true if the determination result is correct among the attributes specified by the attribute specification section from the cases constituted by the success or failure of the attributes and the determination results based on the success or failure of the attributes, and A combination of specific types of attributes specified by the attribute specification section that is less than or equal to the total number is defined as attributes that must be included, and sets of the minimum total number and attributes that must be included are listed, and the combinations of these sets are defined as attributes that must be included. From the content of the set specified based on the inclusion relationship between and the above-mentioned case, a judgment condition is generated that the set contains a specific type of attribute and the above-mentioned minimum total number or more of attributes hold, and this judgment condition is used to determine whether the above-mentioned case is All of the cases in which the discrimination result was determined to be correct are judged, and as a result, if there remain cases among the cases that are not judged to be correct according to the above judgment conditions, this remaining case is A discriminant condition generation method characterized in that the above-described discriminant condition generation process and the discriminant process using the generated discriminant conditions are repeatedly executed for a case to generate a required number of discriminant conditions. [Claim 3] The minimum total number of attributes that should hold true if the determination result is correct is determined from the cases constituted by the success or failure of the attributes and the determination results based on the success or failure of the attributes, and this minimum total number is specified by the attribute number specification section. A combination of attributes of a specific type that is less than or equal to the minimum total number when it is smaller than the maximum value is defined as an attribute that must be included, and sets of the minimum total number and attributes that must be included are listed, and these combinations are defined as attributes that must be included. From the content of the set specified based on the inclusion relationship between and the above-mentioned case, a judgment condition is generated that the specified type of attribute is included and the above-mentioned minimum total number or more of attributes hold true, and based on this judgment condition, the above-mentioned case As a result, if there are cases remaining among the cases that are not judged to be correct according to the above judgment conditions, this remaining case is A method for generating a discriminant condition, characterized in that a required number of discriminant conditions are generated by repeatedly performing the above-described discriminant condition generation process and a discriminating process using the generated discriminant conditions for a case in which the discriminant condition is determined. [Claim 4] Find the minimum total number of attributes that should hold true if the determination result is correct from the cases consisting of the success or failure of attributes and the determination results based on the success or failure of the attributes, and find combinations of specific types of attributes that are less than or equal to this minimum total number. Among them, a specific type of combination specified by the attribute combination specification section is defined as an attribute that must be included, and sets consisting of the minimum total number and attributes that must be included are listed, and inclusions between these sets are defined. From the contents of the set specified based on the relationship and the case, a judgment condition is generated that includes attributes of a specific type and the minimum total number or more of attributes hold true, and based on this judgment condition, a judgment result among the cases is generated. As a result, if there remain cases among the cases that are not judged to be correct according to the above judgment conditions, the above cases will be applied to the remaining cases until there are no remaining cases. 1. A discriminant condition generation method, characterized in that a required number of discriminant conditions are generated by repeatedly performing the discriminant condition generating process and the discriminating process using the generated discriminant conditions.