Arviz plot_energy doesn't like PyMC custom step method?

Yeah I was doing it hierarchically in that fashion since my original model that isn’t the working example I provided was quite degenerate/non-identifiable meaning there were many different optimal configurations which was screwing up the HMC sampling and so I tried to reduce the configuration space in a manner of speaking by first sampling one free variable, then three free variables and then finally all free variables. At least that is my understanding of that specific problem which was why I tried that approach which appeared to work apart from this hitch of screwing with arviz diagnostic plots.

I got the same error that I think you are mentioning which reminded me as to why I rolled forward to 5.11.0 in the first place is the truth value error:

ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

which following some investigation was discovered and resolved in:

So in summary it may be a bug then that custom step_methods expressed as [step1,step2…etc] throw this error and it may be misapplication of the argument or a bug.

1 Like