I can‘t understand the Riemman approximation compute in this paper. #11
Description
When calculating the Riemann estimate, it is mentioned in the article that the effect is best when m=20, but in the scaled_input function in the code, batch_ Size=16, num_ Batch=4, I don't quite understand here, and the gradient and attention_weights, Where is the code for multiplying weights to obtain attribution score? As a novice, I still have some questions and hope you can help me solve them. Thank you very much!