Improve transformer api

[BalzaniEdoardo]

Small PR allowing the chaining behavior of TransformerBasis.

Before this PRs, basis methods returning self could not be chained when the basis was wrapped by the TransformerBasis class.

Old behavior

>>> import nemos as nmo
>>> from nemos.basis._basis import Basis
>>> from nemos.basis._transformer_basis import TransformerBasis
>>> transformer_basis = nmo.basis.BSplineEval(5).to_transformer()
>>> out = transformer_basis.set_input_shape(1)
>>> isinstance(out, TransformerBasis)
False
>>> isinstance(out, Basis)
True

New behavior

>>> import nemos as nmo
>>> from nemos.basis._basis import Basis
>>> from nemos.basis._transformer_basis import TransformerBasis
>>> transformer_basis = nmo.basis.BSplineEval(5).to_transformer()
>>> out = transformer_basis.set_input_shape(1)
>>> isinstance(out, TransformerBasis)
True
>>> isinstance(out, Basis)
False

Key Features

• Decorator that processes the output of chainable basis methods, setting the method output (the updated basis object) to the _basis attribute of the transformer, and returns the transformer.
• New tuple attribute of the TransformerBasis listing all the chainable methods of basis.
• Run-time decoration of chainable methods in the __getattr__ when first called + caching. Run-time decorating is necessary because if set at initialization, it would create a circular reference, and result in infinite loop when pickling.

[sjvenditto]

I committed a fix to a couple typos I found, otherwise it looks good!

[BalzaniEdoardo]

Great! Thanks

…

On Tue, Dec 17, 2024, 4:19 PM Sarah Jo Venditto ***@***.***> wrote: ***@***.**** approved this pull request. I committed a fix to a couple typos I found, otherwise it looks good! — Reply to this email directly, view it on GitHub <#280 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AG2MOSYD5MLCEFD7JFINZR32GCINJAVCNFSM6AAAAABTTVEBUKVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZDKMJQGE2TGMRXHE> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[billbrod]

Don't really have any notes about the chaining, that all looks good to me.

• TransformerBasis objects have a to_transformer() method, which I think just returns self. This is a bit confusing -- is there a reason to have this behavior?
• transform and fit_transform should have the notes about set_input_shape being called first as well (like fit), as well as the Raises section of the docstring