search

spring-epfl / mia

A library for running membership inference attacks against ML models
MIT License
137 stars 27 forks source link

PyTorch support documentation and examples #1

Open amanoel opened 5 years ago

amanoel commented 5 years ago

Hi, you mention in the readme that the package supports PyTorch models, but in ShadowModelBundle._fit you assume the model has fit method (line 116). How exactly have you tested the PyTorch models? I was thinking of maybe using pytorch-fitmodule or SuperModule, but if there's a way you recommend already that would be great. Also it would be nice to include an example of how to load PyTorch modules in the package! (maybe I can do a PR after I'm able to do it myself :-)

bogdan-kulynych commented 5 years ago

Hi, ShadowModelBundle and AttackModelBundle take a scikit-like object, with fit, predict, and predict_proba methods. You can use skorch to wrap your torch model in such an API, or mia's own mia.wrappers.TorchWrapper. You can see example tests that use skorch (shadow, attack, serializers).

bogdan-kulynych commented 5 years ago

If you can add an example, that would be great!

amanoel commented 5 years ago

Ah great I haven't seen the TorchWrapper! In the end I just wrote my own class, but indeed using skorch should be better, I'll do that instead, thanks.