linkedin / photon-ml

A scalable machine learning library on Apache Spark

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Refactor computation of means for Game linear/logistic regression models

ferd36 opened this issue · comments

commented

Problem context and statement:

In Game, the "submodels" do not carry their own loss functions: they are trained one by one, but with a global loss function that includes data about the states of all the other submodels.

Currently, the submodels LinearRegressionModel and LogisticRegressionModel have "computeMean" methods (including e.g. link function for logistic regression), which do not make sense in Game (the loss function is global).

This needs to be simplified.

Design: TBD.