Fix/model adjustments
This merge request fixes the LSTM and ConvLSTM model training by applying some small refinements on the data preprocessing:
- Removing NaN values from the targets (streamflow) by only considering timespans that doesn't contain NaNs
- Apply Min-Max-Scaling on target (streamflow) data (which in the current implementation is only applied on the forcings)
- Small model architecture adjustments
- Reducing the number of timesteps to reduce the computing time and memory usage
- Fix NSE calculation formular
Every single fix is applied as a separate commit and marked within the code by additional comments.