Handling NaN Values in char_with_label Dataset Leading to NaN Training Loss

[liuquant]

Hello,

I followed the steps in the readme.md within the data folder and ran make_dataset.py to fully reproduce the results. However, I found that the dataset contains NaN values, which may cause the training loss to become NaN as well. Could you please advise on how you handled NaN values in the dataset? Should I simply drop all rows with NaNs in char_with_label, or is there a more appropriate way to handle this?

Thank you for your help!

[x7jeon8gi]

Hello,

Could you please let us know where the NaN values are occurring? Even if NaN values are present, they are handled during the model training process by the following code, which fills them in:

dataset = TSDatasetH(df, step_len=step_len, start=start, end=end, fillna_type='ffill+bfill')

If you continue to experience issues with NaN values affecting the training loss, please provide more details on where the NaNs are appearing so we can assist you further.

Thank you!