| Component Name | Version |
| Oracle Data Mining | 9.2.0.4.0 |
Bug Fixes by Category
Bug Category
| Category | Fixed | BugNo | Description |
|---|---|---|---|
| Oracle Data Mining | 9204 | 2702947 | Enhance ABN rules output |
| Oracle Data Mining | 9204 | 2770756 | TX binning utility does not clean up temp tables |
| 9204 | 2837853 | Compute Lift invalid error message generated when no positive target values | |
| 9204 | 2904658 | Attribute Importance results are incorrect for special cases | |
| 9204 | 2932601 | Build ABN Model: Error message should be maximumbuildtime>=0 (not >0) | |
| 9204 | 2941824 | Validate restricted data types | |
| 9204 | 2941831 | Single value target error message | |
| 9204 | 2944622 | Get rules by content fails when rule length is the selection criterion | |
| 9204 | 2956602 | Need GetRules API similar to 10i | |
| 9204 | 2967476 | Inadequate binning can cause ABN model to fail | |
| 9204 | 2989926 | Clustering Apply probabilities are biased for numeric attributes with outliers | |
| 9204 | 3010046 | Model Seeker sample code gets exception in mining function settings creation phase |
ABN rules should reflect three modes: multifeature, singlefeature, and NaiveBayes, whereas rules exist only in singlefeature mode.
This behavior was introduced by a fix to bug 2156957 related to internal TX binning that was hanging with many attributes. The problem was related to the size of the query as well as the number of expressions.
Invalid error message when there was no positive target values in the test data. Error message should state that when there is no positive target in the data it does not compute lift.
Attribute Importance results are incorrect in some special cases, such as single bin values. This has been corrected.
Error message should say that maximum build time is equal to or greater than 0 (instead of greater than 0).
ODM should check whether customer is trying to use an unsupported data type for a data mining operation.
Data mining algorithms that require binary or multi-class targets should generate an error message explaining that target does not have different values.
The new getRulesByContent fails when the rule selection criterion is RuleSortOrder.ruleID. Instead of using the same query as well as the rules themselves in rule cursor, a much simpler counting query is used to get the number of rules.
Currently only a PL/SQL API exists and a sample Java program to get rule using a rule ID. Product management wants a Java API in the current product.
High cardinality in a predictor can cause other predictors to have a low prior probability of selection, while simultaneously causing itself to be eliminated because of high description length. The result is no model found. The user should be alerted to the cause of the problem and be able to override the validation.
Numeric attributes with outliers may create an extreme condition where most of the cases (after internal binning) fall into a single bin. Probabilities computed on such datasets are likely to be biased and lead to incorrect assignment of clusters.
In Mining Function Settings Creation Phase: Received an exception in main: Algorithm "NULL" not supported by ModelSeeker. This has been corrected.