Help navigating implementation of new inference algorithm

[junpenglao]

Nice!

@marius

Is there a way to get the observed value out of a RV, so as to avoid the user needing to do the prob.x = x_obs line?

Yes, you can do x = [model.rvs_to_values[v].data for v in model.observed_RVs]

@marius

As you can see on that page, right now the PyMC version is 30x slower than Numpy and 5x slower than Jax. I’ve done absolute no profiling yet so its maybe too early, but I was wondering if anything glaringly obvious stands out?

Nothing obvious from a quick glance of the implementation (although as you said there is still the concating and transforming the variables for the implementation to work on a general PyMC model). Usually for small functions Aesara should be faster than Jax, so if you could profile the function call and we can find out which is the slow one and try to find solution.