[BUG] Recurrent models do not update scales in cross_validation and predict_isample

[cchallu]

Recurrent models compute scaling parameters one for the whole training data. Scales are not updated during cross-validation or predict_insample with the lastest information, degrading the performance.

[janyaswaller]

Hello, I am using the insample prediction for recurrent models. In my case the insample predictions are significantly worse then the values I get from the crossvalidation(e.g MAE 2590 -> 3494 for the same daterange). Do you have an idea how I could improve the performance? I tried training without local_scaling_type, but it did not change the performance difference between crossvalidation and insample predictions. Another different idea I have is to use the crossvalidation function again a second time, while increasing the number of windows for crossvalidation and try to stop new learning somehow. Can you give me a tip how to solve this, please?