Renamer transformer improvement suggestions

[damienwojtowicz]

Hi all,

Currently, there is no way to rename dictionary keys within a pipeline without resorting to a custom function included in a FuncTransformer, since Renamer transformer only allows for prefixing or suffixing dictionary keys. I think Renamer should behave more like panda's rename function for DataFrames, so users can define key renaming rules as a Dict[str, str]. It would behave as follows :

>>> x = {'foo': 42, 'bar': 24}
>>> renamer = Renamer({'foo' : 'baz', 'bar': 'dummy'})
>>> renamer.transform_one(x)
{'baz': 42, 'dummy': 24}

Obviously, prefixing and suffixing should be preserved (and perhaps extended by letting users define collections of keys that should not be affected? or maybe picking the one they wish to alter?). I suggest to separate those in a Prefixer and Suffixer transformers :

>>> x = {'foo':42, 'bar':24}
>>> pipeline = Renamer({'foo': 'baz'}) 
...     | Prefixer('x_', exclude={'bar'}) 
...     | Suffixer('_in', only={'baz'})
>>> pipeline.transform_one(x)
{'x_baz_in': 42, 'bar': 24}

What do you think about it?

[MaxHalford]

Thanks @damienwojtowicz, I like this proposal!

I think @AdilZouitine had some thoughts on this. Adil, can I let you own this one?

[AdilZouitine]

Hi @damienwojtowicz,
I had the same idea as you a few weeks ago.
The Renamer API that you propose seems to be very good.

@MaxHalford yes!

[MaxHalford]

I would suggest thinking about what happens when a feature is missing. You might also want to add a unit test for that :)

[damienwojtowicz]

@AdilZouitine I'll take a look and make a PR soon then (I'm thinking about starting to contribute to river), if you're OK with it. Following what Max said, I believe missing features should be ignored (or maybe optionally raise a warning for debugging). It's more flexible IMO.

[AdilZouitine]

Great @damienwojtowicz , I will be happy to review your pull request.
I think you should ignore the missing features.
However, it will be important to write this choice in the documentation.

[MaxHalford]

What about the exclude and only parameters? Usually I would handle that functionality compose.Select and compose.Discard. I suggest because I don't like to have two ways of doing things. But I'll let you make the final decision :)

[damienwojtowicz]

You're right, the parameters are redundant with Select and Discard 🙈 Forget about them!