I have 60,000 samples with each having 1,800 features. I have made a multilayer perceptron in Keras and I want to use SHAP values to arrive at global feature importance. Is the matrix too big for SHAP? Is there any way to run SHAP in parallel or make it faster? because currently, it takes a very long time. What other methods do you recommend in case SHAP is impossible (LIME or Deepexplain or DeepLIFT?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.