Fix/regr model multi series

[dennisbader]

Summary

• coverts single timeseries to list of timeseries at beginning of fit/predict in regression models
• general reordering of things in RegressionModel & TorchForecastingModels methods fit() and predicT()
  • i.e. compare predict now between the two models. they are now almost identical. we could think of further factoring out the code duplication
• added component names to encoders to keep track of the encoded components inside the covariates
• abstracted away encode_train and encode_inference behavior inside models into GlobalForecastingModel methods generate_fit_encodings() and generate_predict_encodings()
  • opens up encoding generation to public
  • allows to generate the encodings in a simple/unified way through the models
• refactored parts of encoders to overwrite encoded components if they are in the input (overwriting the output not the input)
  • this simplifies how we can use encoders in forecasting models (i.e. when predicting without series, and covariates, the model loads the covariates which were saved at fitting time in the single series case. The encodings were saved with these covariates and we don't want to generate duplicate encodings for inference, ...)
  • added fit before predict logic for encoders that require it
  • some additional comments/docs where needed

Additional info:

• fixes CatboosModel always writing catboost_info files. By default models now do not write this output

[codecov-commenter]

Codecov Report

Base: 93.88% // Head: 94.01% // Increases project coverage by +0.12% :tada:

Coverage data is based on head (27bd552) compared to base (06824b2). Patch coverage: 97.86% of modified lines in pull request are covered.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1221      +/-   ##
==========================================
+ Coverage   93.88%   94.01%   +0.12%     
==========================================
  Files          73       73              
  Lines        8097     8203     +106     
==========================================
+ Hits         7602     7712     +110     
+ Misses        495      491       -4     

Impacted Files	Coverage Δ
...arts/models/forecasting/torch_forecasting_model.py	87.42% <96.00%> (-0.08%)	:arrow_down:
darts/utils/data/encoder_base.py	96.22% <96.36%> (+2.53%)	:arrow_up:
darts/utils/data/encoders.py	98.88% <98.41%> (+1.67%)	:arrow_up:
darts/models/forecasting/catboost_model.py	100.00% <100.00%> (ø)
darts/models/forecasting/forecasting_model.py	96.65% <100.00%> (+0.08%)	:arrow_up:
darts/models/forecasting/regression_model.py	98.32% <100.00%> (-0.02%)	:arrow_down:
darts/utils/timeseries_generation.py	96.15% <100.00%> (+0.28%)	:arrow_up:
darts/timeseries.py	92.34% <0.00%> (-0.06%)	:arrow_down:
darts/models/forecasting/block_rnn_model.py	98.24% <0.00%> (-0.04%)	:arrow_down:
... and 3 more

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

:umbrella: View full report at Codecov.
:loudspeaker: Do you have feedback about the report comment? Let us know in this issue.