Fail to run example on "Automatic marginalization of discrete variables"

Questions v5
1

Hi, I found a nice example of a technique I’d like to apply in practice:
Automatic marginalization of discrete variables — PyMC example gallery

Also, I found an update to the example still residing in a pull request ( Pull Request #764):
Automatic marginalization of discrete variables — PyMC example gallery

I try to run the latter, which succeeds until the second example, “Coal mining model” raises a NotImplementedError: Cannot convert models with non-default initial_values, during execution of the line

disaster_model_marginalized = pmx.marginalize(disaster_model, ["switchpoint"])

Debugging info refers to the switchpoint variable in particular.
I have used the environment of the watermark info.
Anyone any idea what is the problem here?

2

The model has a custom initval that is not supported by marginalize

3

Thanks, Ricardo, indeed I overlooked that in the updated example the initvals supplied in the original example have been removed. Cheers :smile:.