Evaluating models in Numpyro.

[Claudmj]

Is there anyway to evaluate mixture models such as the following Sine-skewed sine (bivariate von Mises) mixture ? I tried using the following:

numpyro.infer.util.log_likelihood(model=ss_model, posterior_samples=posterior_samples, data=data[aa], num_data=len(data[aa]), K=K)

But I get an error in the assertion in the sample method for the key. I have tried to get the "observed" model and pass that through but I get the same error about the key.

[fehiepsi]

Interesting. Could you check which latent variable is missing? I can follow up on https://forum.pyro.ai/ - we use github to track issues and propose new features.

[Claudmj]

May you please do so I'd really like to evaluate the best number of components for this approach.