Feat/sample weights

[dennisbader]

Checklist before merging this PR:

• Mentioned all issues that this PR fixes or addresses.
• Summarized the updates of this PR under Summary.
• Added an entry under Unreleased in the Changelog.

Fixes #1175, fixes #2107.

(this is a continuation of #2362)

Summary

• adds support for training regression models with sample weights.
• fit() now has two new parameters sample_weight and val_sample_weight which allow to pass sample weights for training and validation (val_sample_weights only for models that support validation sets)

sample_weight
            Optionally, some sample weights to apply to the target `series` labels.
            They are applied per observation, per label (each step in `output_chunk_length`), and per component.
            If a string, then the weights are generated using built-in weighting functions. The available options are
            `"linear_decay"` or `"exponential_decay"`. The weights are only computed the longest series in `series`,
            and then applied globally to all `series` to have a common time weighting.
            If a `TimeSeries` or `Sequence[TimeSeries]`, then those weights are used. The number of series must
            match the number of target `series` and each series must contain at least all time steps from the
            corresponding target `series`. If the weight series only have a single component / column, then the weights
            are applied globally to all components in `series`. Otherwise, for component-specific weights, the number
            of components must match those of `series`.
val_sample_weight
            Same as for `sample_weight` but for the evaluation dataset.

[VascoSch92]

why this is an error if in the docstring you are saying that If a TimeSeries is passed, then those weights are used.?

[madtoinou]

This is a placeholder, we are trying to think if it's worth it to check that the time index of the passed series matches the index of the training series for example.

[codecov]

Codecov Report

Attention: Patch coverage is 96.29630% with 7 lines in your changes missing coverage. Please review.

Project coverage is 93.78%. Comparing base (05f6ddf) to head (ae06f47).
Report is 1 commits behind head on master.

❗ Current head ae06f47 differs from pull request most recent head 0b9b9cc

Please upload reports for the commit 0b9b9cc to get more accurate results.

Files	Patch %	Lines
darts/utils/multioutput.py	77.77%	4 Missing ⚠️
darts/utils/data/tabularization.py	97.05%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2404      +/-   ##
==========================================
+ Coverage   93.77%   93.78%   +0.01%     
==========================================
  Files         138      138              
  Lines       14384    14492     +108     
==========================================
+ Hits        13488    13592     +104     
- Misses        896      900       +4     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[madtoinou]

Great work!

One minor comment about the index for the generated weights; maybe we should consider the two most extremes index values rather than the length of the longest series? WDYT?