We are pleased to announce the availability of new algorithms and features for OML4SQL – the SQL interface to Oracle Machine Learning – on Oracle Database 21c (21.3) for Linux.

In Oracle Database 21c, OML4SQL includes two new algorithms: XGBoost and MSET-SPRT. The highly popular and powerful extreme gradient boosting trees, or XGBoost, extends the OML classification and regression suite of algorithms, while also introducing ranking. You can use XGBoost as a stand-alone predictor or incorporate it into real-world production pipelines for a wide range of problems such as ad click-through rate prediction, hazard risk prediction, web text classification. Ranking is a machine learning technique to rank items and is useful for many applications in information retrieval such as e-commerce, social networks, and recommendation systems. OML4SQL XGBoost is a scalable gradient tree boosting system that makes available the open source gradient boosting framework. OML4SQL XGBoost is sparse data-aware with automatic handling of missing values and encoding of categorical values.

To extend our anomaly detection suite of algorithms, OML4SQL includes the multivariate state estimation technique – sequential probability test ratio algorithm from Oracle labs. Also known as MSET-SPRT, this algorithm supports anomaly detection for sensor and IOT data sources allowing the detection of subtle anomalies while producing minimal false alarms.

The 21c release also introduces several algorithm enhancements.

The OML neural network algorithm has a new solver, the Adam solver, that works on mini batches for computationally efficient processing that requires little memory. This makes it well suited to larger data sets. The OML neural network algorithm now also supports the ReLU activation function, or rectified linear activation function. This can result in models that are easier to train and offer better performance.

Another enhancement is to OML prediction details to improve the quality of those results for the range of in-database algorithms.

Oracle Database 21c (21.3) is an Innovation Release and is now available on additional supported platforms (see MOS note 742060.1 for the list). Innovation Releases enable customers to rapidly adopt new and enhanced functionality for use cases and applications that could benefit from them. New features in Oracle Database 21c also include Blockchain Tables, Native JSON Datatype, Oracle Machine Learning for Python, SQL Macros, and Javascript execution inside the Oracle Database. Customers should be aware that Innovation Releases have shorter support windows than Long Term Releases such as Oracle Database 19c. Customers adopting Innovation Releases should be prepared to upgrade to the next database release within one year after the next database release ships.

Learn more about Oracle Machine Learning and the new features and enhancements in Oracle Database 21c. Also, visit the new features learning guide, download the software (or try it in Always Free ADB), and explore the hands-on workshops on Oracle LiveLabs.

Source link

Leave a Reply

Your email address will not be published.