Mu of pm.Normal() depends on output of RandomVariable on previous time step. How do I implement this?

[ricardoV94]

@MaxT

My problem is this: I don’t want CustomDist or model.register_rv to compute any logp function. I just want them to run the scan which stores mu and sigma values in tensors (e.g., T.set_subtensor(sigmaList[idx], sig,inplace=True), which I can then use after the scan to compute the likelihood of observations in pm.Normal(“likelihood”, mu=mus, sigma=sigmas, observed=observations).

The thing is you can’t create PyMC variables inside a Scan. Your other alternative is to work purely on the logp side of your model. Assign Flat priors to all your variables outside of the scan, pass them as sequences and then at the end define a Potential with the expression that represents the whole model joint-probability.