How to use free random variables in step method function?

Questions
3

Yes, I have tried it with a pymc3 distribution. I came up with an easy example (similar to this ) where it actually works with a Normal distribution as prior for the mean parameter.

import pymc3 as pm
import numpy as np
import pymc3.backends.ndarray
import matplotlib.pyplot as plt
from pymc3.step_methods.arraystep import ArrayStep
from pymc3 import Continuous

class MyStepMethod(ArrayStep):

    def __init__(self, vars, model=None):
        self.vars = vars
        model = pm.modelcontext(model)
        super(MyStepMethod, self).__init__(vars, [model.fastlogp])
        self.accrate = 0
        self.count = 0

    def astep(self, q0, logp):

        q = q0.copy()
        q[0] = np.random.normal(q[-1], 1) # using current gamma value as mean

        logalt, logneu = logp(q0), logp(q)
        accept = logneu - logalt

        if (accept > 0) or (np.random.uniform(low=0, high=1) < np.exp(accept)):
            self.accrate += 1
            return q

        else: return q0

with pm.Model() as model:
    mean = pm.Normal("test", 5, 0.001)
    I = pm.Normal("I", mean,1)

    updI = MyStepMethod(vars=[I, mean])
    step1 = pm.NUTS(vars=[mean])
    trace = pm.sample(5000, [updI, step1], start={'I':0.5}, njobs=1, chains=1, tune=1000)

pm.traceplot(trace)
plt.show()

However, if I use a different distribution for my gamma parameter and e.g. change the line

mean = pm.Normal(“test”, 5, 0.001)

to

mean = pm.Gamma(“test”, 81, 100)

I get the shape error:

Traceback (most recent call last):
  File "/home/test.py", line 50, in <module>
    updI = MyStepMethod(vars=[I, mean])
  File "/home/test.py", line 26, in __init__
    super(MyStepMethod, self).__init__(vars, [model.fastlogp])
  File "/home/user/.local/lib/python3.6/site-packages/pymc3/step_methods/arraystep.py", line 113, in __init__
    self.ordering = ArrayOrdering(vars)
  File "/home/user/.local/lib/python3.6/site-packages/pymc3/blocking.py", line 36, in __init__
    raise ValueError('Shape of variable not known %s' % name)
ValueError: Shape of variable not known test

Any idea on why it only works for normal distributions or how I can get it to work for a gamma distributed variable?