Prediction per timestamp

[joniali]

Thanks a lot for the great code! It's nicely written.

If I understand it correctly, given a sequence of events and timings your code only predicts the last event and the corresponding timing. As opposed to https://github.com/musically-ut/tf_rmtpp/blob/ea4ab25b12422d3b0657082c90bc4beb957c0e83/src/tf_rmtpp/rmtpp_core.py#L575 which I believe predicts every event and timing and computes its corresponding losses.

It would be great if you could clarify this!

[ritvik06]

I also believe what you have stated is true (Reference lines 69 and 70 in model.py). For a sequence of length L, the model should train (and predict) the next timestamp and category for every subset sequence of length K (<L). I will raise a pull request for this soon.

[ritvik06]

If you go through utils.py, you will notice that the dataloader consists of length seq_len. If there is a sequence of length L, the author utilizes all contiguous seq_len length sequences (possible only if seq_len <= L) in its training and testing, so each sequence in the data contributes (L-seq_len+1) sequences in the training and test data. What most other papers do is what I talked about above (predict next event for every subset sequence of length K (<L)).