Marc-Antoine Schmidt

@tbenthompson, if you could look into this I would like to get your opinion on whether we should spend the time fixing the `_split_col_subsets` method or write a new method...

My PR currently contains a fix. My solution was to create a `column_map` attribute to the `SplitMatrix` object. It stores the mapping from index location to actual location in the...

I like the idea, but just want to add a word of caution from my previous experience using patsy. Patsy seems to be focused on non-regularized models. For instance, it's...

> and improves lasso performance with categorical variables. Have you documented the performance improvement with lasso? I would be curious to see the numbers.

No official plans yet, but those are clearly within the scope of this package. The more demand there is for these distributions the more likely they are to be added...

I can take care of this. @tbenthompson, I wanted to see if you have any objection to my solution: - use "family" whenever we talk about an unparametrized distribution. -...

Hi @mathurinm! Unfortunately, glum is not the right package to estimate this type of model. Are you sure your design matrix is correctly specified? You are trying to train a...

We need more details to investigate this issue. - What does the data you are trying to fit the model on look like? - How are you creating the GeneralizedLinearRegressor...

Hi @thobanster! Thanks for raising an issue. I would like to split your question into two parts: 1) estimate a GLM with zero-heavy data, and 2) estimate a GLM on...

> Another thought that occurred to me is that in cases where > 80% of the observations are 0 for the dependent variable, the Tweedie or Poisson GLMs might not...