The question is similar to the one asked in SO, but I think it is more appropriate to be here. It remains anyway unanswered.

Assuming multivariate time series, how to evaluate the importance of all inputs of an already trained LSTM-RNN?

Thus, which are the inputs that affect the output the most?

I am aware of Pearson correlation (for linear relations) and Information Gain based on Entropy reduction to explore input/output relations in general, but this is not what I need. How can you infer from an already trained LSTM network, which inputs affect the output the most? Can you tell by the weights?

Any suggestion or literature would be welcome.



Your Answer

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

Browse other questions tagged or ask your own question.