sunner4nwpu / temma Goto Github PK

Thanks for your work. I want to know which lr scheduler do you choose? You did not mention in your paper. Thanks a lot. :)

Hi,

Thanks for the code.

In your paper, you mentioned that the training data is converted into sequence format with sequences of 30s long and 10s overlap.
So, the training data input would be of size (batch_size, seq_length, feature_dimension), where seq_length=750 (i.e, 30/0.04). Now, in your test set, are you doing the same or is the test set size equal to (batch_size, feature_dimension)?

Thanks,
VR

Hi,

I have another question.

In most papers that use RECOLA, they mention about doing some post-processing operations. The parameters of the post-processing filtering, mean removal etc are optimised on the validation set. Did you use a similar post-processing/refinement on the predictions?

Thanks,
VR