I would like sklearn to support Poisson, gamma and other Tweedie family loss functions. These loss distributions are widely used in industry for count and other long tailed data. Additionally, they are implemented in other libraries such as R: GLM, GLMNET, GBM ext. Part of implementing these distributions would be to include a way for offsets to be passed to the loss functions. This is a common way to handle exposure when using a log link function with these distributions.

Would the sklearn community be open to adding these loss functions. If so I or (hopefully others) would be willing to research the feasibility of implementing these loss functions and offsets into the sklearn API. Thanks