interpretable-han-for-document-classification-with-keras icon indicating copy to clipboard operation
interpretable-han-for-document-classification-with-keras copied to clipboard

Published 20 hours ago • verified publisher icon charlesdong1991

han.show_word_attention() does add up to 1

Open MalteHB opened this issue 5 years ago • 0 comments

I have tried to implement your functions, and i really do like your work. However, i find that when using the han.show_word_attention() function the attention weights does not add up to 1. How i have understood it, it should have because it is just supposed to be the softmax probabilities for the attention for each word. Do you know how i might fix this? Best Regards Malte

MalteHB avatar Jan 13 '20 09:01 MalteHB