Have you tried fixing the warnings before ignoring them? They can probably be avoided, which is a much better solution to ignoring the warnings.

I have tried fixing the convergence warnings but it was a dead end
The other Runtime error was due to numpy library

It can be tricky to fix those warnings sometimes, but going deeper in the library and finding why you get the warning, you get a better idea of which parameters to change so that the warnings are gone :)

Fixing the ConvergenceWarning can be done by either setting sample_posterior to True, or setting a much higher tol.

You cannot put the tol higher since it will modify the figure inappropriately (and make the take-home message different). However, sample_posterior=True seems to be reasonable.

Looking more in depth. This example reveals a bug in the IterativeImputer. The division by zero should not happen or should be taken handled internally. The fourth feature in boston as an std. dev. of zero leading to an error when computing the correlation coefficient between feature.

This issue is hidden by the cross_val_score which ignore the error.

Edit:
This is a warning in fact. So we should probably catch it.

FYI, the fourth feature is a categorical feature:

- CHAS     Charles River dummy variable (= 1 if tract bounds river; 0 otherwise)

For this PR, I would only apply the fix propose by @NicolasHug (sample_posterior=True).
The other warning should be caught within the code itself.

@0x6f736f646f Thanks for your contribution