why do some chains complete very quickly while others take forever? is it possible to avoid those slow chains (and have them automatically killed?). if not, what are you supposed to do when say 5 chains complete within a minute (or 10mins) and the rest take an hour or more? im finding myself avoiding using more chains to reduce the chance of having more slow ones dragging the entire experiment down.
That behaviour (dramatic difference in sampling times between chains) suggests that there’s something pathological about the posterior geometry and the chains are not all exploring the same regions. You should therefore be skeptical of the chains that finish quickly and should visualize the posterior from slow vs fast chains to diagnose what’s going on.
Thanks @mike-lawrence, that makes sense. If there was something I could do about it, i’d have looked at the posterior but even if it shows me something ‘pathological’ - i can’t do anything about that information and i already suspect it’s pathological from the poor results