Function to transform etna objects to dict

[martins0n]

🚀 Feature Request

We need function to serialize and deserialize etna objects to dict for experiments, cli and platform.
For now we can generate etna objects with hydra_slayer.get_from_params but we can't reverse that transformation.

Proposal

• Add to_dict method to BaseMixin, which will save etna object in hydra-core, hydra-slayer format.
• In some cases for example ChangePointsTrendTransform we have external objects as input parameters:
  • For sklearn BaseEstimator we can get parameters with BaseEstimator.get_parameters
  • For other cases: we should leave knowledge of _target_ class only and raise warning with the name of field.
• For inspiration you can look at __repr__ implementation or at to_json:

    def to_json(self):
        """Library objects to json representation.

        Returns
        -------
        Dict with object allenai-common representation.
        """
        class_to_name = {}
        self.reversed_index(Registrable._registry, class_to_name)
        json_config = {}
        init_args = inspect.signature(self.__init__).parameters
        json_config["type"] = class_to_name[self.__class__]
        for arg, param in init_args.items():
            if param.kind == param.VAR_POSITIONAL:
                continue
            value = self.__dict__[arg]
            if param.kind == param.VAR_KEYWORD:
                json_config.update(value)
            elif getattr(value, "to_json", None) is not None:
                json_config[arg] = value.to_json()
            elif value.__class__ in class_to_name:
                json_config[arg] = {"type": class_to_name[value.__class__]}
            elif isinstance(value, tuple) or isinstance(value, list):
                json_config[arg] = [item if not isinstance(item, BaseMixin) else item.to_json() for item in value]
            else:
                json_config[arg] = value
        return json_config

Test cases

• Pipeline, model, metrics, transforms (choose some not trivial cases: ChangePointsTrendTransform for example) with to_dict and hydra_slayer.get_from_params initialize the same object
• check that ensembles are working with to_dict correctly
• check that warning is arised in case of non etna-sklearn models.

Additional context

Example of Pipeline:

_target_: etna.pipeline.Pipeline
horizon: 14
model:
  _target_: etna.models.CatBoostPerSegmentModel
transforms:
- _target_: etna.transforms.TimeSeriesImputerTransform
  in_column: target
  strategy: mean
- _target_: etna.transforms.AddConstTransform
  in_column: target
  value: 100
- _target_: etna.transforms.LogTransform
  in_column: target
- _target_: etna.transforms.LagTransform
  in_column: target
  lags: ${shift:${pipeline.horizon},[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16,
    17, 18, 19, 20]}
- _target_: etna.transforms.DateFlagsTransform
  day_number_in_month: true
  day_number_in_week: true
  is_weekend: true
  month_number_in_year: true
  week_number_in_month: true
  week_number_in_year: true
  year_number: true