search

pytorch / captum

Model interpretability and understanding for PyTorch
https://captum.ai
BSD 3-Clause "New" or "Revised" License
4.92k stars 497 forks source link

Issue With Reference Provided for Input x Gradient Method #451

Closed Europium248 closed 4 years ago

Europium248 commented 4 years ago

Thank you very much for building Captum! I noticed that both in the README.md and in the method documentation at https://captum.ai/api/input_x_gradient.html, the reference provided for the "input x gradient" method is a paper that appeared in November 2016 (https://arxiv.org/abs/1611.07270). However, the "input x gradient" method (as well as the equivalence to epsilon-LRP for ReLU networks) actually appears six months prior in May 2016 in https://arxiv.org/abs/1605.01713. I know this may seem like a small thing, but I think it's important to give credit where credit is due especially in academia!

NarineK commented 4 years ago

@Europium248, Thank you for pointing out to it. We can certainly fix it. Feel free to send a PR if you'd like too :)

Europium248 commented 4 years ago

Thanks, made the PR!

NarineK commented 4 years ago

Thank you very much. Do you mind fixing it here too? https://github.com/pytorch/captum/blob/master/captum/attr/_core/input_x_gradient.py#L16

Europium248 commented 4 years ago

The current PR says the change is already present in both locations - does look different to you?

Screenshot 2020-08-20 at 12 38 36 PM
NarineK commented 4 years ago

I see the change, thank you :)

NarineK commented 4 years ago

Merged the PR. Closing the issue. Thank you!