5.1 About Scoring and Deployment

Scoring is the application of models to new data. In Oracle Machine Learning for SQL, scoring is performed by SQL language functions.

Predictive functions perform classification, regression, or anomaly detection. Clustering functions assign rows to clusters. Feature extraction functions transform the input data to a set of higher order predictors. A scoring procedure is also available in the DBMS_DATA_MINING PL/SQL package.

Deployment refers to the use of models in a target environment. Once the models have been built, the challenges come in deploying them to obtain the best results, and in maintaining them within a production environment. Deployment can be any of the following:

• Scoring data either for batch or real-time results. Scores can include predictions, probabilities, rules, and other statistics.

• Extracting model details to produce reports. For example: clustering rules, decision tree rules, or attribute rankings from an Attribute Importance model.

• Extending the business intelligence infrastructure of a data warehouse by incorporating machine learning results in applications or operational systems.

• Moving a model from the database where it was built to the database where it used for scoring (export/import)

OML4SQL supports all of these deployment scenarios.

Note:

OML4SQL scoring operations support parallel execution. When parallel execution is enabled, multiple CPU and I/O resources are applied to the execution of a single database operation.

Parallel execution offers significant performance improvements, especially for operations that involve complex queries and large databases typically associated with decision support systems (DSS) and data warehouses.