Changes in This Release for Oracle Machine Learning for SQL Concepts

Changes in this release for Oracle Machine Learning for SQL Concepts.

Changes in Oracle Machine Learning for SQL 21c

Changes in Oracle Machine Learning for SQL Concepts for Oracle Database 21c.

New Features in 21c

Oracle Machine Learning for SQL features new in Oracle Database 21c.

New Algorithms

• MSET-SPRT

  The Multivariate State Estimation Technique - Sequential Probability Ratio Test (MSET-SPRT) algorithm is a nonlinear, nonparametric anomaly detection machine learning technique designed for monitoring critical processes. It detects subtle anomalies while also producing minimal false alarms.

  The algorithm calibrates an expected behavior from available, historical data from the normal operational sequence of monitored signals. The learned behavior of the system is then incorporated into a persistent Oracle Machine Learning for SQL MSET-SPRT model that captures expected normal behavior and can be applied to new records to detect anomalous behaviors.

• XGBoost

  XGBoost is machine learning algorithm for regression and classification that makes available the XGBoost open source package. Oracle Machine Learning for SQL XGBoost prepares training data, invokes XGBoost, builds and persists a model, and applies the model for prediction.

New Algorithm Setting

Adam Optimization Solver

Adam is an extension to stochastic gradient descent that uses mini-batch optimization. The Adam solver can make progress faster by seeing less data than the L-BFGS solver. Adam is computationally efficient, with little memory requirements, and is well-suited for problems that are large in terms of data or parameters or both.

Enhancements

Neural Network Algorithm Settings

The Neural Network algorithm setting NNET_ACTIVATIONS now accepts the value NNET_ACTIVATIONS_RELU. Rectified Linear Units is a commonly used activation function for deep learning models that addresses the vanishing gradient problem in large neural networks.

The algorithm has a new setting, NNET_SOLVER, that specifies the method of optimization, either L-BFGS or Adam.

For the NNET_NODES_PER_LAYER and NNET_ACTIVATIONS settings, you can now specify a single value that is then applied to each hidden layer.

The NNET_ITERATIONS setting has a default value for the LBFGS solver and now has a default value for the Adam solver. The default values are different for each solver.

Other Changes

The following are additional changes in Oracle Machine Learning for SQL Concepts for 21c:

• Updated Tuning the O-Cluster Algorithm.

• Updated About Oracle Machine Learning for SQL.

• Added "Anomaly Detection for Time Series Data" topic in the Anomaly Detection chapter.

• Updated machine learning process illustration and added process descriptions. See The Oracle Machine Learning Process.

• Added machine learning functions overview. See Machine Learning Techniques.