✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: c9f6351. Link to the linter CI: here}

@betatim thanks for the review! Do you think it is worthwhile changing n_iter_without_progress to be max_iter_without_progress for consistency (and semantically, it makes more sense, but another deprecation). Happy to do this in this PR.

So the common tests check_estimators_overwrite_params and check_dont_overwrite_parameters fail because I am updating max_iter during fit. It is used too late to do the parameter check then. We could:

• add xfail (I've done this)
• add self.max_iter_ attribute and use this in the subsequent functions. This would be more annoying to remove when the time comes. (happy to do this instead, or another option)

Do you think it is worthwhile changing n_iter_without_progress to be max_iter_without_progress for consistency (and semantically, it makes more sense, but another deprecation). Happy to do this in this PR.

Probably yes. Like you I think it would be nice for consistency, but it would mean more deprecation. However, I think a large fraction of people who use n_iter_without_progress are also using n_iter. So they are already "there" in the code to make a change, this means the cost of changing an additional parameter name is not so high?

Not sure what to recommend. I think I'd implement the change and see what others say.

Do you think it is worthwhile changing n_iter_without_progress to be max_iter_without_progress for consistency (and semantically, it makes more sense, but another deprecation). Happy to do this in this PR.

In other estimators it's usually named n_iter_no_change so I wouldn't change to max_iter_without_progress. I think that if we want to change it we should make it consistent with other estimators. I doesn't need to be part of this PR though.

Will leave n_iter_without_progress here and it can be changed to match other estimators later. I think in general 'max'/'max_iter_no_change' makes more sense semantically but its better that the estimators are consistent and probably not worth changing.

I've amended the OP to 'closes' because I don't think BernoulliRBM needs amending to max_iter as there is no convergence criterion. So this PR would close the issue.