Target multi-input examples to specific neutrons for correct/incorrect outputs

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

729 stars 133 forks source link

Target multi-input examples to specific neutrons for correct/incorrect outputs #40

Open cap76 opened 5 years ago

cap76 commented 5 years ago

Hi, I'm wondering if the multi-input attribution models can be targeted to specific neutrons for e.g., correct/incorrect class. In the single input models you can specify a ys=ys, but in the examples for the multi-input models, this doesn't appear to be the case, and throws an error. Am I missing something obvious :S

marcoancona commented 5 years ago

Could you please report the error and the code that you are using? Notice that instead of passing ys you can mask the target tensor directly (eg. target = original_target * ys)... maybe this help in your case.