NameError: unbound axis name: _provenance

1

I am trying to produce a mixture model of StudentT distribution, but I am getting an error message that is very obscure to me about a NameError: unbound axis name: _provenance.

The very simplified model below still generates the error.

def model_at_given_mag(x=None, y=None, peak=None):

    # Adjust peak location
    loc = numpyro.sample('peak', dist.Normal(peak, 0.1))

    # stable component
    stable_logshape = numpyro.sample('s_logshape', dist.Uniform(1, 3))
    stable_logscale = numpyro.sample('s_logscale', dist.Uniform(-3, 0))

    stable_shape = 10 ** stable_logshape
    stable_scale = 10 ** stable_logscale

    stable = dist.StudentT(stable_shape, loc=loc, scale=stable_scale)
    
    with numpyro.plate('obs', y.shape[0]):
        # yobs represents the median at the pdf x location
        obs = numpyro.sample("yobs", stable, obs=y)

Can anyone spot the issue?
It is very hard to find where the issue is from the traceback and error messages.

2

Hi @mfouesneau, the issue has been fixed upstream. We will make a new release next week. For now, could you use the dev version instead?

pip install git+https://github.com/pyro-ppl/numpyro.git
3

The small example does not pass with the dev.

4

Interesting. Could you create a github issue with small reproducible code? We’ll try to resolve it before the release.

5

No need, sorry I tested too quickly. The dev version works on that example.
Thanks.