Thanks! Is there a difference between doing that and passing it through all in one go?:
pm.Binomial('obs', n=Ntrial, p=p, observed=data_raw)
I tried your way which was slower but but it looked like it worked slightly better so I was wondering if there is a difference.
I also get the same errors as before for both (max tree depth reached / number of effective samples is smaller than… for some parameters). I also get posterior values quite different to the observed success rates but I’m not sure if it’s because 9 parameters is too little to express the (4 x 4) grid