Updated _output_2d_ attributes documentation for DummyClassifier and…

[nitya]

… DummyRegressor

Reference Issues/PRs

Partially addresses #14312

What does this implement/fix? Explain your changes.

Added documentation for output_2d attribute in DummyRegressor and DummyClassifier

Any other comments?

[NicolasHug]

Thanks for the PR @nitya

I'm pretty sure that we always have

self.output_2d_ == (self.n_outputs_ != 1)

and that output_2d_ is redundant. I'd be happy to see it deprecated and removed. Do you want to give it a try @nitya ?

[nitya]

Will give it a try tomorrow. Thank you!

[nitya]

@NicolasHug
Looking for clarifications here:

(Line 123) - self.output_2d_ is defined as:
self.output_2d_ = y.ndim == 2 and y.shape[1] > 1

(Line 125-128) - subsequently, self.n_outputs_ is defined:

if y.ndim == 1:
     y = np.reshape(y, (-1, 1))

self.n_outputs_ = y.shape[1]

My original PR was for the docstring (documentation of attribute) for output_2d_ and not for any code changes.

So clarification requested:

1. Are you suggesting I remove the self.output_2d_ attribute completely and just use (self.n_outputs_ != 1) as the inline replacement? OR

2. Are you suggesting we keep the attribute but simply update its definition to this simple check vs. the ones above?

If #1, then need to make only code changes (original docstring-related PR is irrelevant)
If #2, then I can make that change but can you verify the docstring PR (language used) is correct?

[NicolasHug]

I'm suggesting that we deprecate the attribute, meaning that you would have to:

• use n_outputs instead (like you mentioned in 1.)
• create a property that raises a deprecation warning when outputs_2d_ is used.

You can look at our deprecation guidelines in the contributing guide.

[NicolasHug]

Do you need any help @nitya ? Or should somelse take it up?

[NicolasHug]

@nitya FYI I have opened a PR for deprecating the attribute #14933, which would close this one once/if merged

[reshamas]

@kellycarmody
Can you drop Nitya an email? Thanks.

[kellycarmody]

Thanks for your help @nitya! Really appreciate it!