The message below says:

# Also make sure to change `max_iter` default back to 1000 and deprecate None

This removes None straight away and doesn't deprecate it, right?

Yes but this is not how we do it usually. We've decided that when we add None as a valid option for a deprecation, then we can remove it right away at the end of the deprecation. It's not like it's a meaningful option, just a sentinel during deprecation.

The writer of the message must have not been aware of this, but it's alright to do it anyway.

OK my main point was that in scikit-learn 1.6.1 the following code does not give any warning and in 1.7 this is going to break (with this PR merged):

import numpy as np
from sklearn.manifold import TSNE

TSNE(max_iter=None).fit(np.random.randn(1000, 10))

I guess that's OK because it's unlikely that someone has used max_iter=None explicitly?

indeed, we never encouraged to set None explicitely. It's there only to serve as a sentinel while we have the 2 max_iter and n_iter params. We already merged several PRs like this one for 1.7 and and before 1.6 as well.

This make sense @jeremiedbb. So I guess this review comment can be resolved?