Fix one input not change when a model have multiple inputs

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

720 stars 133 forks source link

Fix one input not change when a model have multiple inputs #25

Closed SamuelXie closed 5 years ago

SamuelXie commented 5 years ago

Hi! My model have two inputs. I only want one of them to be computed using DeepExplain and the other not change. How can i do this?

SamuelXie commented 5 years ago

Attribution methods are post-hoc interpretability... It didn't need worry about multi-input. I mistake the meaning of attribution.