Yohan Chalier Projects Portfolio Blog

Implementation of the Hoeffding AnyTime Tree Algorithm

Supervised by Albert Bifet, for the Data & Knowledge master.

Hoeffding AnyTime Tree makes a simple change to the current de facto standard for incremental tree learning. The current stateof-the-art Hoeffding Tree aims to only split at a node when it has identified the best possible split and then to never revisit that decision. In contrast HATT aims to split as soon as a useful split is identified, and then to replace that split as soon as a better alternative is identified.

The project aims for an implementation of this algorithm for it to be included in Scikit-Multiflow.