Abstract

Driven by the needs of Flink to expand the offline engine to a hybrid one, a new machine learning (ML) library, called SOLMA is proposed. This library aims to cover online learning algorithms for data streams. In this setting, data streams are processed sequentially example by example. SOLMA, which is under development, currently contains two classes of algorithms: (i) basic streaming routines such as online sampling, online PCA, online statistical moments and (ii) advanced online ML algorithms covering in particular classification, regression and drift/anomaly detection and handling. This paper briefly highlights the concepts underlying SOLMA.