Show progress bar for long-running log-likelihood evaluations

[michaelosthege]

With most recent ArviZ and PyMC3, the log-likelihoods are automatically computed when return_inferencedata=True.

I just wondered why the notebook cell ran the sampling for 2 hours and then took another hour or so to return.. Turns out I forgot to pass idata_kwargs=dict(log_likelihood=False), causing from_pymc3 to re-evaluate my model on all posterior samples, which for an ODE model is not that fast.

Thoughts on implementation

For long-running log-likelihood evaluations it would be nice to show a progress bar. It would help the user to diagnose what's taking so long.

There may be alternatives to obtain log-likelihoods from sampler stats?