Adrian Seyboldt

That's a good question, initially I was using the square of the gradient. From some experiments I'm pretty sure that that just doesn't work in practice though (but verification of...

@sethaxen Someone also asked on the stan slack, and I made a small notebook for comparing those with a diagonal normal posterior: https://gist.github.com/aseyboldt/46ce8a0967026fd3b7ae11a53c884975

@nsiccha I still think looking at a standard normal is a bit misleading. It is I think quite normal that the initial mass matrix gets worse with increasing dimension, as...

> Sounds like a job for [posteriordb](https://github.com/stan-dev/posteriordb) @mike-lawrence That's where the plot in the description comes from :-)

@aarcher07 Thank you for reporting this, and sorry for the very late reply... I think the problem you are seeing is due to a small mistake in the arguments to...

I'm closing this because I think it was a problem in the example code, but feel free to reopen or comment if you don't agree or have questions.

This is a fun and interesting idea, and I really enjoy the inventiveness of the implementation! However, I'm not sure I really would want this to be in pymc. The...

> This won't prevent users from using names, so I don't see how this is an argument against. Of course users didn't bother to give unique assignment names in the...

Well, it seems like I'm the only one to who this looks like too much magic, if that's the case then I guess it's fine. :-) I'll just argue a...

From a _very_ quick look: - I think you need a more challenging example problem. With that one everything will converge really quickly, and you won't see much of a...