Updated results

Bugfix: correct computation of validation losses

BugFix: Ensure order of the columns is kept in M-PCNNs

Bugfix: normalization of D_inputs for M-PCNNs

Use weighted loss to avoid artificially low errors when sequences are batched

Better handling of unnormalized temperature predictions

Better way of splitting train/validation/test sets without overlap unless really little data provided

Feature: add separate seed for data splitting and randomization

BugFix: move path creation to name factory and only create it if needed

Mixing sequences before splitting, X,Y as tensors

Better tensor and save handling

Bugfix right assessment of None

Remove annoying prints at the end of training

Update PyPI package

BugFix: Ensure heating and cooling are possible at the same time from different power sources