Thanks for the quick feedback! I’d certainly say it’s challenging, particularly since i’m new to Theano in general.
So I’ve made a small amount of progress and the theano.scan it works when I run the test outside the PyMC3 model. To accomplish this, instead of explicitly specifying my constants I replaced them with tt.fscalar() values. But when I test it within a pm.Model() it fails with error:
AttributeError: stim_n has no test value
It seems PyMC always sets theano.config.compute_test_value to ‘raise’ when i instantiate the model. I’ve tried running “tn.config.compute_test_value = ‘off’” at the start of the model, but this causes the error:
AttributeError: ‘scratchpad’ object has no attribute ‘test_value’
So I’m guessing that’s a bad idea? Through searching I’ve found several variants of this issue, and they all seem to happen sometimes to some people and not others. Maybe due to broken packages? As per this issue, I just ran theano.test() and had 261 errors 
. I guess it’s time to try a fresh install!
I’ll update with my progress here in case someone finds it helpful in the future. Again, if you know of anything obvious I’m missing please let me know!