I’ve been using this example to implement a “blackbox” likelihood function, but I am having trouble trying to extend it in the following way.
In the linked example, two parameters (m and c) are defined with priors and then converted to a theano tensor. This is then given as the “observed” argument to DensityDist():
pm.DensityDist('likelihood', lambda v: logl(v), observed={'v': theta})
Suppose I have a variable called l whose prior I do not want to update during sampling (I think this is called an unobserved variable, but I could be wrong here). The variable l is a normally distributed RV and I know the mean and the standard deviation, and I want to sample from this distribution in the likelihood blackbox function, but I don’t want its posterior updated as the sampler runs. How can I expand the example in the link to include this?
Please let me know if my question is unclear and thanks for your help!