Thanks for this PR @bhaskarb

I do not see a reference to the this in the case of BernoulliNB. I was looking at the naive_bayes.rst file in the source.Thanks in advance

also would need to be changed (and same for BernoulliNB) as setting to 0 will raise a warning following #9131 on this line.

But then I'm not sure what should be said there... "set to a small positive number to disable smoothing" doesn't sound right.

I'm not sure if raising a warning for alpha=0 is such a good thing in the first place. On one side, it's understandable to warn the user but it's also unreasobable to expect him/her to enter a small value that is close enough to zero to disable smoothing but large enough to avoid the warning.

I could be worth waiting for a second opinion on this.

I would allow alpha=0 and warn / error if we divide by zero later?

Closing as superseded by #22269.