|Oracle® Data Mining Application Developer's Guide
11g Release 2 (11.2)
Part Number E12218-05
This section describes new features in the Oracle Data Mining APIs. It includes the following sections:
This section lists the changes that have been introduced in the Oracle Data Mining 11.2 PL/SQL API:
Note:The same changes are implemented in the Java API. Refer to Oracle Data Mining Java API Reference.
In Oracle Data Mining 11g Release 2 (11.2), you can build association rules models without first transforming the transactional data.
See:"Market Basket Data"
SVM Class Weights Specified with CLAS_WEIGHTS_TABLE_NAME
Previously SVM class weights were specified in the priors table (
CLAS_PRIORS_TABLE_NAME setting). Now SVM class weights and GLM class weights are both specified in a class weights table (
CLAS_WEIGHTS_TABLE_NAMEin Oracle Database PL/SQL Packages and Types Reference
FORCE argument to DROP_MODEL
You can now force a drop model operation even if a serious system error has interrupted the model build process.
DBMS_DATA_MINING.DROP_MODELprocedure in Oracle Database PL/SQL Packages and Types Reference
GET_MODEL_DETAILS_SVM has new REVERSE_COEF parameter
To preserve model transparency, the
GET_MODEL_DETAILS functions automatically reverse the transformations generated by ADP during the model build. You can obtain the transformed attribute coefficients used internally by an SVM model by setting the new
reverse_coef parameter to 1. This causes the coefficients and bias to be returned with the normalization shifts and scales applied by ADP.
DBMS_DATA_MINING.GET_MODEL_DETAILS_SVMin Oracle Database PL/SQL Packages and Types Reference
This section describes features introduced in Oracle Data Mining 11g Release 1 (11.1).
Mining Model schema objects
In Oracle 11g, Data Mining models are implemented as data dictionary objects in the
SYS schema. A set of new data dictionary views present mining models and their properties. New system and object privileges control access to mining model objects.
In previous releases, Data Mining models were implemented as a collection of tables and metadata within the
DMSYS schema. In Oracle 11g, the
DMSYS schema no longer exists.
Automatic and Embedded Data Preparation (ADP and EDP)
In most cases, data must be transformed using techniques such as binning, normalization, or missing value treatment before it can be mined. Data for build, test, and apply must undergo the exact same transformations.
In previous releases, data transformation was the responsibility of the user. In Oracle Database 11g, the data preparation process can be automated. Algorithm-appropriate transformation instructions are embedded in the model and automatically applied to the build data and scoring data. The automatic transformations can be complemented by or replaced with user-specified transformations.
Because they contain the instructions for their own data preparation, mining models are known as supermodels.
Oracle Data Mining Concepts for information about automatic, embedded, and custom data transformation for Data Mining
Oracle Database PL/SQL Packages and Types Reference for information about
Scoping of Nested Data
Oracle Data Mining supports nested data types for both categorical and numerical data. Most algorithms require multi-record case data to be presented as columns of nested rows, each containing an attribute name/value pair. Oracle Data Mining processes each nested row as a separate attribute.
Attributes that are not nested are identified by the column name. Since Oracle Database prevents duplicate column names, the names of non-nested attributes are always unique. However, no such guarantee exists for nested attributes. In Oracle Data Mining 10g, it was up to the user to verify the uniqueness of nested attribute names. This name checking was a required step in the preparation of nested data.
In Oracle Data Mining 11g, attribute name duplication is not possible, because nested attribute names are scoped with the column name. The name checking step is no longer required, thus simplifying the task of data preparation for the user.
In Oracle Data Mining 11g, Decision Tree and O-Cluster algorithms do not support nested data.
See Also:"Nested Data"
Standardized Handling of Sparse Data and Missing Values
Handling of sparse data and missing values has been standardized across algorithms in Oracle Data Mining 11g. Data is sparse when a high percentage of the cells are empty but all the values are assumed to be known. This is the case in market basket data. When some cells are empty, and their values are not known, they are assumed to be missing at random. Oracle Data Mining assumes that missing data in a nested column is a sparse representation, and missing data in a non-nested column is assumed to be missing at random.
See Also:"Missing Data"
Generalized Linear Models
A new algorithm, Generalized Linear Models, is introduced in Oracle 11g. It supports two mining functions: classification (logistic regression) and regression (linear regression).
See Also:Oracle Data Mining Concepts for information about Generalized Linear Models
New SQL Data Mining Function
A new SQL Data Mining function,
PREDICTION_BOUNDS, has been introduced for use with Generalized Linear Models.
PREDICTION_BOUNDS returns the confidence bounds on predicted values (regression models) or predicted probabilities (classification).
See Also:Chapter 6, "Scoring and Deployment"
Enhanced Support for Cost-Sensitive Decision Making
Cost matrix support is significantly enhanced in Oracle 11g. A cost matrix can be added or removed from any classification model using the new procedures,
The SQL Data Mining functions support new syntax for specifying an in-line cost matrix. With this new feature, cost-sensitive model results can be returned within a SQL statement even if the model does not have an associated cost matrix for scoring.
Only Decision Tree models can be built with a cost matrix.
Oracle Data Mining Concepts for information about costs
Oracle Database PL/SQL Packages and Types Reference for information about
Oracle Data Mining Scoring Engine
In Oracle 10.2, you could use Database Configuration Assistant (DBCA) to configure the Data Mining option. In Oracle 11g, you do not need to use DBCA to configure the Data Mining option.
Basic Local Alignment Search Tool (BLAST)
Adaptive Bayes Network classification algorithm
DM_USER_MODELS view and functions that provide information about models, model signature, and model settings (for example,
GET_MODEL_SIGNATURE). These are replaced by data dictionary views. See Chapter 5 for information about the data dictionary views.
The Oracle Data Mining Java API (OJDM) fully supports the new features in Oracle Data Mining 11g Release 2 (11.2). This section provides a summary of the new features in the Java API. For details, see Oracle Data Mining Java API Reference (Javadoc).
As described in "Mining Model schema objects", mining models in 11g Release 2 (11.2) are data dictionary objects in the SYS schema. System and object privileges control access to mining models.
In the Oracle Data Mining Java API, a new extension method OraConnection.getObjectNames is added to support listing of mining objects that can be accessed by a user. This method provides various object filtering options that applications can use as needed.
As described in "Automatic and Embedded Data Preparation (ADP and EDP)", Oracle Data Mining 11g Release 2 (11.2) supports automatic and embedded data preparation (supermodels).
In the Oracle Data Mining Java API, a new build setting extension method,
OraBuildSettings.useAutomatedDataPreparations, is added to enable ADP. Using the new
OraBuildTask.setTransformationSequenceName, applications can embed the transformations with the model.
Two new GLM packages are introduced:
oracle.dmt.jdm.modeldetail.glm. These packages have GLM algorithm settings and model details interfaces respectively.
New apply content enumeration values,
probabilityUpperBound, are added to specify probability bounds for classification apply output. The enumeration
oracle.dmt.jdm.supervised.classification.OraClassificationApplyContent specifies these enumerations. Similarly apply contents enumeration values
predictionUpperBound are added to specify prediction bounds for regression model apply output. In this release only GLM models support this feature.
New static methods
removeCostMatrix are added to
OraClassificationModel to support associating a cost matrix with the model. This will greatly ease the deployment of costs along with the model.
Mining task features are enhanced to support the building of mining process workflows. Applications can specify dependent tasks using the new
OraTask.addDependency method. Another notable new task feature is
overwriteOutput, which can be enabled by calling the new
With these new features, applications can easily develop mining process workflows and deploy them to the database server. These task workflows can be monitored from the client side. For usage of these methods refer to the demo programs shipped with the product (See Oracle Data Mining Administrator's Guide for information about the demo programs.)
A new mining object,
oracle.dmt.jdm.transform.OraTransformationSequence supports the specification of user-defined transformation sequences. These can either be embedded in the mining model or managed externally. In addition, the new
OraExpressionTransform object can be used to specify SQL expressions to be included with the model.
oracle.dmt.jdm.OraProfileTask is added to support the new predictive analytics profile functionality.
The Oracle Data Mining Java API can be used with Oracle Database 11g Release 2 (11.2) and with Oracle Database 10.2. When used with a 10.2 database, only the 10.2 features are available.