Sergey Koposov

I think this is related to #13631 (or it is the same issue). I don't think the prompt-toolkit versions mattered there though.

this is a duplicate of #13631

Thanks for the PR @ColmTalbot I think for this to be merged I'd like to see. * an example where this new behaviour performs better than the existing behaviour with...

Thank you for your examples, but i don't think I agree with your interpretations. 1) I am not sure that average jumping length is the best statistic to look at....

> To clarify, the first example above is a 15-dimensional bimodal Gaussian (I forgot to say that above.) I see, that makes more sense. Thanks. Was it with a single...

> To address the first point. If we don't match the correct distribution, we don't have an independent sample. That's the definition of the "correct" distribution. I think the question...

Regarding the ACF, I haven't used it much, so I am not too familiar with its limitations, so you may be right there. My thinking was that we don't need...

Thanks, it looks good to me, I'll merge as soon as the tests are done. Can you please also add a line about the change to CHANGELOG.md ?

The decrease in coverage revealed the issue with this (I saw it before, but didn't have a good solution) The n_effective is defined as one of the arguments of sample_initial...

Yes, the range(1) seems to be a no-op and can be probably removed. And yes the reason why I was a bit reluctant to make the change to sample_initial is...