Academia.eduAcademia.edu

Figure 5 – uploaded by Dora Cai

See full PDFdownloadDownload figure
TABLE 7. A generalized relation We now describe the second database learning algorithm, LCLR, learning classification rules from databases (Cai et al. 1990a). In a manner similar to the LCHR algorithm, the LCLR algorithm applies the attribute-oriented induction technique as well. The difference is that in the extraction of classification rules, the facts that support the target class serve as positive examples, while those that support the other classes serve as ‘‘negative’’ examples. Since the learning task is to discover the concepts that have discriminant proper- ties, the portions of the facts in the target class that overlap with other classes should be detected and removed from the description of classification rules. We analyze such a learning process using a similar example.

Table 7 A generalized relation We now describe the second database learning algorithm, LCLR, learning classification rules from databases (Cai et al. 1990a). In a manner similar to the LCHR algorithm, the LCLR algorithm applies the attribute-oriented induction technique as well. The difference is that in the extraction of classification rules, the facts that support the target class serve as positive examples, while those that support the other classes serve as ‘‘negative’’ examples. Since the learning task is to discover the concepts that have discriminant proper- ties, the portions of the facts in the target class that overlap with other classes should be detected and removed from the description of classification rules. We analyze such a learning process using a similar example.

Related Figures (9)

To add a single entry mathematics as a sublevel entry o! Science, the key word is_a can be used as below.
To add a single entry mathematics as a sublevel entry o! Science, the key word is_a can be used as below.
TABLE 1. A tuple in a relation is essentially a logic formula The intermediate and final learning results can also be rep- resented using relational tables. Such a relation is called a generalized relation. The final generalized relation may con- tain several tuples which represent a disjunction of several conjunctions. Therefore, our logic bias (Genesereth and Nilsson 1987) for learned rules is not limited to conjunctive rules. A learned rule can also be a disjunctive rule with a small number of disjuncts. Such a relaxation makes learning more effective because it is often necessary to represent a Jearned rule in a disjunctive form. A maximum number of disjuncts of the resulting formula, that is, the maximum number of tuples in the final generalized relation, can be specified by users as a threshold value in the learning pro- cess. For example, if a threshold value is three, the final generalized rule will consist of at most three disjuncts with each being a sequence of conjuncts.
TABLE 1. A tuple in a relation is essentially a logic formula The intermediate and final learning results can also be rep- resented using relational tables. Such a relation is called a generalized relation. The final generalized relation may con- tain several tuples which represent a disjunction of several conjunctions. Therefore, our logic bias (Genesereth and Nilsson 1987) for learned rules is not limited to conjunctive rules. A learned rule can also be a disjunctive rule with a small number of disjuncts. Such a relaxation makes learning more effective because it is often necessary to represent a Jearned rule in a disjunctive form. A maximum number of disjuncts of the resulting formula, that is, the maximum number of tuples in the final generalized relation, can be specified by users as a threshold value in the learning pro- cess. For example, if a threshold value is three, the final generalized rule will consist of at most three disjuncts with each being a sequence of conjuncts.
TABLE 2. A relation Student in a sample university database In Fig. 2, {...] denotes a set of zero or more occurrences, while [...] denotes one or zero occurrences. The syntax of (Where_Clause) is the same as SQL. However, the constants in (Where_Clause) and in (Class_Name) can be any node in the concept hierarchies. For example, undergraduate is not a constant in the Student relation table but a node in the concept hierarchy of the attribute status. It can be used as a class name or inside an SQL (Where_Clause).
TABLE 2. A relation Student in a sample university database In Fig. 2, {...] denotes a set of zero or more occurrences, while [...] denotes one or zero occurrences. The syntax of (Where_Clause) is the same as SQL. However, the constants in (Where_Clause) and in (Class_Name) can be any node in the concept hierarchies. For example, undergraduate is not a constant in the Student relation table but a node in the concept hierarchy of the attribute status. It can be used as a class name or inside an SQL (Where_Clause).
(Michalski 1983), A tuple is a conjunctive normal form. Each attribute and its value form one of the conjuncts. When a conjunct is removed, a conjunct constraint is elim- inated, and thus the rule is generalized. If there is a large set of distinct values for an attribute, the large set of values
(Michalski 1983), A tuple is a conjunctive normal form. Each attribute and its value form one of the conjuncts. When a conjunct is removed, a conjunct constraint is elim- inated, and thus the rule is generalized. If there is a large set of distinct values for an attribute, the large set of values
TABLE 3. The set of data relevant to ‘‘graduates’’ must be generalized in order to derive a generalized concept. However, if there is no higher-level concept provided for the attribute, it cannot be generalized by ascending concept trees. Therefore, the attribute should be removed. “
TABLE 3. The set of data relevant to ‘‘graduates’’ must be generalized in order to derive a generalized concept. However, if there is no higher-level concept provided for the attribute, it cannot be generalized by ascending concept trees. Therefore, the attribute should be removed. “
TABLE 6. Simplification of the generalized relation no larger than the specified threshold value. However, the total number of tuples in the resulting relation may still be above the threshold value. In this case, further generalization on some attribute(s) should be performed. The choice of the generalized attribute may depend on the tuple reduction ratio, the simplicity of the final learned rules, etc.
TABLE 6. Simplification of the generalized relation no larger than the specified threshold value. However, the total number of tuples in the resulting relation may still be above the threshold value. In this case, further generalization on some attribute(s) should be performed. The choice of the generalized attribute may depend on the tuple reduction ratio, the simplicity of the final learned rules, etc.
TABLE 8. A generalized relation or may not be a graduate student. Therefore, in order to get an effective classification rule, care must be taken to handle the overlapping tuples. We consider strategy 5.
TABLE 8. A generalized relation or may not be a graduate student. Therefore, in order to get an effective classification rule, care must be taken to handle the overlapping tuples. We consider strategy 5.
Fic. 5. Problem space for learning classification rules.
Fic. 5. Problem space for learning classification rules.
FIG. 7. (a) The entire and (b) the factored version spaces for the same concepts.
FIG. 7. (a) The entire and (b) the factored version spaces for the same concepts.
Related topics:
Cognitive ScienceComputer ScienceRelational DatabaseArtificial Intelligence
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved