🚀 Feature Request

The idea is to create a transform with custom transformation/inverse transformation.

Motivation

It can help users to write less boilerplate code for custom transforms.

Proposal

class LambdaTransform(Transform):

    def __init__(self, in_column, inplace, out_column, transform_func, inverse_transform_func):
        pass

Constructor parameters:

• in_column: name of column to transform
• inplace: should transform be applied inplace or create a new column
• out_column: name of column to create if inplace=False
• transform_func: function that takes wide dataframe with selected in_column and transforms it so the shape and columns remains the same. It can be
  • some function on dataframe itself
  • calling apply with any user-defined function in lambda
  • type: Callable[[pd.DataFrame], pd.DataFrame]
• inverse_transform_func: is necessary if inplace=True, signature is the same as signature of transform_func

Reference transforms (multi-segment):

• LagTransform
• AddConstTransform
• LogTransform

After creation of this transform some existing transforms can be rewritten using it. We can inherit them from LambdaTransform and set transform_func, inverse_transform_func to some methods of the class.

• LagTransform -- add some checks during the init an
• AddConstTransform
• LogTransform -- add some checks during the transform

Test cases

1. Make sure that rewritten transforms pass the tests.
2. Test that if inplace=True inverse_transform_func is necessary.
3. Test interface if inplace=True.
4. Test interface if inplace=False: column name should be generated by out_column if it is present.
5. Test transform method for some function examples.
6. Test inverse_transform for some function examples in inplace=True and inplace=False.

Alternatives

No response

Additional context

No response

Checklist

• I discussed this issue with ETNA Team