EventFilter (EF) is a service that trains classifier models (or formulae) online. Currently it supports only MatrixNet (*) classifier, but it could be extended in the future to support other classifier implementations. EventFilter has a cluster of several machines attached, so it supports parallel training of up to 12 formulas. General ("happy" user) use-case is the following:

• upload dataset
• specify variables and parameters for training & train
• download trained formula

In reality this use-case is just a step in a bigger chain (offline data analysis) that contains such tasks as dataset preparation, quality metric estimation, formula parameter optimization, etc. Since the whole data analysis chain should be automatically executable, EF provides HTTP API as well as Python wrapper that could be easily embedded in data analysis chain. Examples of such wrapper usage, as well as detailed instructions & how-tos that extensively cover data-analysis automation can be found here.

(*) MatrixNet is a custom implementation of Gradient Boosted Decision Trees classification algorithm.

Presentations & tutorials on EventFilter that might give overview of the service are attached to this page.