Skip navigation links

oracle.dmt.jdm.task
Interface OraPredictiveAnalyticsTaskFactory

All Superinterfaces:

javax.datamining.Factory

public interface OraPredictiveAnalyticsTaskFactory

extends javax.datamining.Factory

OraPredictiveAnalyticsFactory is the factory class to create predictive analytics tasks, such as predict, explain, and profile.

Predictive analytics tasks simplify developing of data mining applications by automating data mining operations for most common use cases. For example, OraPredictTask generates predictions and its accuracy for the specified target column by using input table details, case identifier, and the target column name as input. It automates the data preparation, model building, testing, and scoring to produce the prediction results that applications can readily consume.

Since:

OJDM 11.1.0

See Also:

OraPredictTask, OraExplainTask, OraSegmentTask

Method Summary

OraExplainTask	createExplainTask(java.lang.String inputDataURI, java.lang.String explainColumn, java.lang.String explainResultTableName) Creates an instance of OraExplainTask that is used to produce importance and ranks for the columns with respect to the explain column in the input dataset.
OraPredictTask	createPredictTask(java.lang.String inputDataURI, java.lang.String caseID, java.lang.String targetColumn, java.lang.String predictionResultTableName) Creates an instance of OraPredictTask that is used to produce predictions for the specified target column in the input dataset as predictions output table.
OraProfileTask	createProfileTask(java.lang.String inputDataURI, java.lang.String targetAttributeName, java.lang.String profileResultTableName) Creates an instance of OraProfileTask allows users to segment data based on some target attribute and value.

Method Detail

createPredictTask

public OraPredictTask createPredictTask(java.lang.String inputDataURI,
                                        java.lang.String caseID,
                                        java.lang.String targetColumn,
                                        java.lang.String predictionResultTableName)
                                 throws javax.datamining.JDMException

Creates an instance of OraPredictTask that is used to produce predictions for the specified target column in the input dataset as predictions output table. The input consists of a table that has case id column, other predictor columns and a target column. The target column containing the values to predict. The input data must contain some cases in which the target value is known (that is, is not NULL). Cases where the target values are known are used to train models.

Predictions output table has the following fixed columns:

• case_id VARCHAR2 or NUMBER
• • Each of the cases identified in the case_id column. This is the same as the name that was passed in. The data type is the same as input case_id type.
• prediction VARCHAR2 or NUMBER
• • The predicted value of the target column for the given case. The data type is the same as the input target_column_name type.
• probability NUMBER
• • For classification (categorical target), the probability of the prediction. For regression problems (numerical target), this column contains NULL.

Predictions are returned for all cases in the input data. Predicted values for known cases may be interesting in some situations, for example, to perform deviation analysis, that is, to compare predicted values and actual values.

Parameters:

inputDataURI - input data uri ([schemaName].tableName)

caseID - column with unique entries (like customer id)

targetColumn - column name for which predictions has to be made

predictionResultTableName - prediction results table name

Throws:

javax.datamining.JDMException

createExplainTask

public OraExplainTask createExplainTask(java.lang.String inputDataURI,
                                        java.lang.String explainColumn,
                                        java.lang.String explainResultTableName)
                                 throws javax.datamining.JDMException

Creates an instance of OraExplainTask that is used to produce importance and ranks for the columns with respect to the explain column in the input dataset.

Explain task output table has the following fixed columns:

• attribute_name VARCHAR2
• • Name of a column in the input data; all columns except the explained column are listed in the result table.
• explanatory_value NUMBER
• • Value indicating how useful the column is for determining the value of the explained column. Higher values indicate greater explanatory power. Value can range from 0 to 1. An individual column's explanatory value is independent of other columns in the input table. Instead, the values are based on how strong each individual column correlates with the explained column. The value is affected by the number of records in the input table, and the relations of the values of the column to the values of the explain column. An explanatory power value of 0 implies there is no useful correlation between the column's values and the explain column's values. An explanatory power of 1 implies perfect correlation; such columns should be eliminated from consideration for PREDICT. In practice, an explanatory power equal to 1 is rarely returned.
• rank NUMBER
• • Ranking of explanatory power. Rows with equal values for explanatory_power have the same rank. Rank values are not skipped in the event of ties.

Parameters:

inputDataURI - input data uri ([schemaName].tableName)

explainColumn - column name with respect to which other columns are ranked

explainResultTableName - explain results table name

Throws:

javax.datamining.JDMException

createProfileTask

public OraProfileTask createProfileTask(java.lang.String inputDataURI,
                                        java.lang.String targetAttributeName,
                                        java.lang.String profileResultTableName)
                                 throws javax.datamining.JDMException

Creates an instance of OraProfileTask allows users to segment data based on some target attribute and value. It will create profiles or rules for records where the specific attribute and value exist, in some sense it can be seen a directed segmentation. E.g. a user may want to understand what attributes customers have in common that responded to the latest promotion offer.

• PROFILE_ID NUMBER
• • ID of the profile
• RECORD_COUNT NUMBER
• • Number of cases included in the specific profile id
• DESCRIPTION XMLTYPE
• • A rule indicating the specific attributes and their values for this profile (e.g. AGE>30 and MARITAL_STATUS = `DEVORCED')

Parameters:

inputDataURI - input data uri ([schemaName].tableName)

targetAttributeName - name of the target column

profileResultTableName - name of the table where the profile results are persisted

Returns:

Throws:

javax.datamining.JDMException

Skip navigation links