Custom likelihood from a python function

[BaptisteDE]

@BrynWalton

custon

Hello !
I have a similar problem, I would like to use the result of an FMU (Fuctional Mockup Unit) calculation as the mean of my likelyhood function. I already wrapped the FMU in a python function, but it doesn’t accept Pytensors.

I think this is similar to this discussion:

Using PyMC3 with external functions

I am trying to use PyMC3 to replace emcee as my MCMC sampler of choice. However, I depend on external functions to draw the model curves, which I import from existing packages that are not compatible with Theano’s style. How can I still use PyMC3 as a sampler? For example: #::: Option 1: How the external fct currently looks def external_fct_1(x, xmid, depth): y = np.zeros_like(x) y[ (x>xmid-0.1) & (x<xmid+0.1) ] = -depth return y #::: Option 2: Rewriting the external fct in Theano…

I tried to find the tutorials listed here :
“Specifically, the tutorial on using a blackbox likelihood function, the tutorial on custom distributions, the tutorial on wrapping jax functions into PyTensor Ops.”
from Likelihood Approximations with Neural Networks in PyMC - PyMC Labs

The one you mention @jessegrabowski is among them, but all the links seem to be broken.

Do you have any idea where I should look ?

Thank you !