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)
Support for a new perturbation method: approximate Shapley Values using the sampling-based algorithm by Castro et al. 2009
Support for batch processing (using the batch_size param on the explain() method). Batch processing is disabled by default. #31
The weights or mask for the target tensor can now be passed explicitly with the ys parameter. While the previous manual masking technique is still supported, this does not work with batch processing. #31
Upgrade to v 0.2
batch_sizeparam on theexplain()method). Batch processing is disabled by default. #31ysparameter. While the previous manual masking technique is still supported, this does not work with batch processing. #31