How to save/serialize model: WeibulAFTFitter

[torresxavier]

Hi there! Nice library this one! I have trained a model using WeibullAFTFitter and would like to reuse it without training it again... I've tried with the suggestions given in the documentation (pickle, dill, and also joblib...) but it keeps on throwing the same mistake:

" Sorry, pickling not yet supported. See pydata/patsy#26 if you want to help."

If I try with the CoxPH model it works, but not under the WeibullAFTFitter. And the last one is the one showing best performance so I need to save it to disk.

Any thoughts on how to save/serialize a trained WeibullAFTFitter model for future usage?

Thanks

[CamDavidsonPilon]

Hi @torresxavier, there was a recent regression in the latest lifelines: in order to support formulas (which, I am guessing you are using), we had to sacrifice some serialization. However, a solution is coming soon, in #1131 - should only be a few weeks.

If you are comfortable with pip, you can do pip install https://github.com/CamDavidsonPilon/lifelines/archive/try-formulaic.zip to install a working lifelines with full support for serialization.

[torresxavier]

I'll check that. Thanks @CamDavidsonPilon !!!

[CamDavidsonPilon]

This is now native in lifelines