Current non-stationary normalization may result in NaN values

[WenjieDu]

1. System Info

The current implementation of non-stationary normalization from the official TimesNet code will result in NaN values when missing rate is high, e.g. >0.8. We need to add a small number to avoid dividing by 0.

https://github.com/WenjieDu/PyPOTS/blob/main/pypots/imputation/timesnet/modules/core.py#L55-L64

2. Information

• The official example scripts
• My own created scripts

3. Reproduction

pass

4. Expected behavior

Nan values in model output.