Cmdstanpy==1.2.0, arviz and no attribute 'stan_vars_cols'

General
, ,
1

Hello, I’ve upgraded cmdstanpy to 1.2.0. When I try to create an InferenceData object with arviz==0.16.1 I get the below error. It works fine with cmdstanpy=1.1.0.

Get the same with the eight schools example.

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
Cell In[10], line 2
      1 # Create id object
----> 2 idata_stan = az.from_cmdstanpy(
      3     posterior=model_samples
      4 )

File /usr/local/anaconda3/envs/xxx/lib/python3.10/site-packages/arviz/data/io_cmdstanpy.py:829, in from_cmdstanpy(posterior, posterior_predictive, predictions, prior, prior_predictive, observed_data, constant_data, predictions_constant_data, log_likelihood, index_origin, coords, dims, save_warmup, dtypes)
    763 def from_cmdstanpy(
    764     posterior=None,
    765     *,
   (...)
    778     dtypes=None,
    779 ):
    780     """Convert CmdStanPy data into an InferenceData object.
    781 
    782     For a usage example read the
   (...)
    827     InferenceData object
    828     """
--> 829     return CmdStanPyConverter(
    830         posterior=posterior,
    831         posterior_predictive=posterior_predictive,
    832         predictions=predictions,
    833         prior=prior,
    834         prior_predictive=prior_predictive,
    835         observed_data=observed_data,
    836         constant_data=constant_data,
    837         predictions_constant_data=predictions_constant_data,
    838         log_likelihood=log_likelihood,
    839         index_origin=index_origin,
    840         coords=coords,
    841         dims=dims,
    842         save_warmup=save_warmup,
    843         dtypes=dtypes,
    844     ).to_inference_data()

File /usr/local/anaconda3/envs/xxx/lib/python3.10/site-packages/arviz/data/io_cmdstanpy.py:76, in CmdStanPyConverter.__init__(self, posterior, posterior_predictive, predictions, prior, prior_predictive, observed_data, constant_data, predictions_constant_data, log_likelihood, index_origin, coords, dims, save_warmup, dtypes)
     73 self.dtypes = dtypes
     75 if hasattr(self.posterior, "metadata"):
---> 76     if self.log_likelihood is True and "log_lik" in self.posterior.metadata.stan_vars_cols:
     77         self.log_likelihood = ["log_lik"]
     78 elif hasattr(self.posterior, "stan_vars_cols"):

AttributeError: 'InferenceMetadata' object has no attribute 'stan_vars_cols'
2

This is a known Arviz issue

A workaround at the moment seems to be az.InferenceData(posterior=fit.draws_xr())

1 Like
3

Legend, thanks!