Memory limitations for categorical variables in generalized linear model?

I'm using GeneralizedLinearModel.fit with a large number of non-categorical predictors and one categorical predictor. I find that I run into memory limitations if I either have too many categories or have too many non-categorical predictors. Can anyone explain what GeneralizedLinearModel.fit actually does with a categorical predictor variable? Does it just turn it into a bunch of non-categorical binary predictors behind-the-scenes? What factors lead to memory problems when using GeneralizedLinearModel.fit?

Thanks for any advice that can be offered.

Answer 1

[Shashank Prasanna]

Categorical predictor variable are converted to dummy variables internally. Size depends on number of categorical variables and levels etc. This is very problem dependent since I don't know how many observations you have and how many variables. Memory issues can be attributed to how much RAM or memory you have and also if you are are on a 32 or a 64 bit environment. hth