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)
I've been using this fork and just wanted to note that in order to use the DeepExplain library with a model created with TF2 (without having to re-write existing compile and train code), I had to do the following steps first:
Save the model weights
Add the lines:
tf.compat.v1.disable_v2_behavior()
tf.compat.v1.disable_eager_execution()
Build the model using tf.compat.v1
Load the saved model weights
Might be worth adding to documentation as it might be useful to others...
I've been using this fork and just wanted to note that in order to use the DeepExplain library with a model created with TF2 (without having to re-write existing compile and train code), I had to do the following steps first:
Might be worth adding to documentation as it might be useful to others...