search

marcoancona / DeepExplain

A unified framework of perturbation and gradient-based attribution methods for Deep Neural Networks interpretability. DeepExplain also includes support for Shapley Values sampling. (ICLR 2018)
https://arxiv.org/abs/1711.06104
MIT License
725 stars 133 forks source link

PyTorch support #1

Open shubhamjain0594 opened 6 years ago

shubhamjain0594 commented 6 years ago

I am interested in developing pytorch support. Any channel where that discussions are going on?

I have background in PyTorch, active contributor to visdom, and other libraries.

rohitghosh commented 6 years ago
marcoancona commented 6 years ago

Hi all, although we have planned support for PyTorch, we have not started implementing it nor gave a lot of thought about possible technical challenges. However, gradient-override in PyTorch should be easier than in Tensorflow. The idea is that PyTorch should be implemented in a module deepexplain.pytorch and should expose the same API. Then it would be used the same way, except from for the import statement:

from deepexplain.tensorflow import DeepExplain
from deepexplain.pytorch import DeepExplain

Pull requests are welcome. If you have in mind a possible discussion channel that could be opened for this let me know.