Add BATS and TBATS model adapters

[iKintosh]

🚀 Feature Request

Add new models to library.

Motivation

Should expand number of models in ETNA

Proposal

Add BATS and TBATS model adapters (use models from tbats library https://github.com/intive-DataScience/tbats)

Create _TBATSAdapter inherit it from BaseAdapter. It should implement methods fit, predict and get_model.

Also create _TBATSPerSegmentModel(PerSegmentPredictionIntervalModel) as a model it should expect class BATS or TBATS from tbats (also you may use Estimator).

After that create classes BATSPerSegmentModel and TBATSPerSegmentModel

To understand the idea look at the

etna/etna/models/sklearn.py

Line 13 in be1b719

class _SklearnAdapter(BaseAdapter):

etna/etna/models/sklearn.py

Line 75 in be1b719

class SklearnPerSegmentModel(PerSegmentModel):

etna/etna/models/linear.py

Line 8 in be1b719

class LinearPerSegmentModel(SklearnPerSegmentModel):

Test cases

• not fitted model (should rise an error)
• test repr

  etna/tests/test_models/test_linear_model.py

  Lines 100 to 107 in be1b719

  	def test_repr_linear(model_class, model_class_repr):
  	"""Check __repr__ method of LinearPerSegmentModel and LinearMultiSegmentModel."""
  	kwargs = {"copy_X": True, "positive": True}
  	kwargs_repr = "copy_X = True, positive = True"
  	model = model_class(fit_intercept=True, **kwargs)
  	model_repr = model.__repr__()
  	true_repr = f"{model_class_repr}(fit_intercept = True, {kwargs_repr}, )"
  	assert model_repr == true_repr

• test on dummy dataset with two seasonality (combine two sinusoids)

Alternatives

No response

Additional context

Look at the sktime for inspiration https://github.com/alan-turing-institute/sktime/blob/2a0df3f52a6fe67f5509f6199cc60f01e3dc0adf/sktime/forecasting/tbats.py#L15

Checklist

• I discussed this issue with ETNA Team