Stochastic node from a trace - or sharing params across models

Questions
2

I was trying to come up with a way to sample from another distribution using the following code but I ran into an error which I’m a little perplexed by. See below if it helps it all. The general idea is to make a step function which just makes a draw from a preexisting array.

import pymc3 as pm
import numpy as np

from pymc3.step_methods.arraystep import BlockedStep
external_vals = np.random.randn(10,2) * 5 + 2

class ExternalSample(BlockedStep):
    def __init__(self, external, var):
        self.vars = [var]
        self.name = var.name
        self.external = external

    def step(self, point):
        row_chosen = np.random.choice(self.external.shape[0])
        point[self.name] = self.external[row_chosen]       
        return point
    
with pm.Model() as model:
    external_pm = pm.Normal('external', shape=2)
    y = pm.Normal('y', mu=external_pm, shape=2)
    step = [ExternalSample(external_vals, external_pm)]
    trace = pm.sample(step=step)