Why some chains get stuck more than others?

[jelc53]

Thank you for your help previously with installation. We now have a viable model up and running thanks to this repo!

One follow up question I had: why might some NUTS chains get stuck (or just slow down sampler progress) more than others? Is there some reason for example, that the gradient function might become harder to evaluate?

[mfouesneau]

That most likely depends on your model. If the gradient becomes too flat, you need to update the chain parameters.
For example, sometimes the epsilon value become so small after initialization that if becomes an issue