Online Learning

Learning from streams is a hot topic in todays research. The masses of data often require on-time analysis while only a very limited resources with regard to CPU and memory are available.

Several algorithms have been proposed for approximating classifiers, counting elements or clustering items within a data stream.

The stream-analysis Module

The stream-analysis modules of the streams library provides implementations for online methods for analysis such as different approximative counting algorithms or computation of online statistics (e.g. quantile summaries).

In addition, it incorporates the powerful MOA library, a high-end Java library providing implementations for various online learning schemes.

Download and Usage

The stream-analysis module is a standalone module available at:

http://download.jwall.org/streams/stream-analysis-0.9.5-SNAPSHOT.jar

The above JAR archive contains a ready-to-run package including MOA and the rest of the streams library to start experiments by simply running

   # java -jar stream-analysis-0.9.5-SNAPSHOT.jar your-experiment.xml

Examples for experiments can be found e.g. in Integrating MOA.