Closed halvgaard closed 3 years ago
Hi Rasmus,
Absolutely, you caught a bug I oversaw. I have updated the code and am curious to run it again with the day and month encodings, and see if the model can catch temporal dependencies over longer time periods.
Thanks for the feedback !
Stumbled upon these identical lines in your preprocessing: https://github.com/nklingen/Transformer-Time-Series-Forecasting/blob/d85a5fff2af508d3f3a7b64994d28dbe90818597/Preprocessing.py#L50-L51
I assume they should have .day and .month instead of .hour ?