from_pymc3 fails in certain circumstances

[canyon289]

Describe the bug
A clear and concise description of what the bug is.

To Reproduce

import arviz as az
import pymc3 as pm
print(az.__version__)
print(pm.__version__)

with pm.Model() as planetmodel:
    p_water = pm.Uniform("p", 0 ,1)
    w = pm.Binomial("w", p=p_water, n=2, observed=1)
    trace = pm.sample(5000, chains=2)

water_data = az.from_pymc3(trace=trace)

0.4.0
3.7
Auto-assigning NUTS sampler...
Initializing NUTS using jitter+adapt_diag...
Multiprocess sampling (2 chains in 2 jobs)
NUTS: [p]
Sampling 2 chains: 100%|███████████████████████████████████████████████████████████████| 11000/11000 [00:03<00:00, 2882.36draws/s]
/Users/canyon/.miniconda3/envs/bayestutorial/lib/python3.6/site-packages/arviz/data/io_pymc3.py:56: FutureWarning: arrays to stack must be passed as a "sequence" type such as list or tuple. Support for non-sequence iterables such as generators is deprecated as of NumPy 1.16 and will raise an error in the future.
chain_likelihoods.append(np.stack(log_like))
Traceback (most recent call last):
File "trace_error.py", line 11, in
water_data = az.from_pymc3(trace=trace)
File "/Users/canyon/.miniconda3/envs/bayestutorial/lib/python3.6/site-packages/arviz/data/io_pymc3.py", line 152, in from_pymc3
dims=dims,
File "/Users/canyon/.miniconda3/envs/bayestutorial/lib/python3.6/site-packages/arviz/data/io_pymc3.py", line 137, in to_inference_data
"sample_stats": self.sample_stats_to_xarray(),
File "/Users/canyon/.miniconda3/envs/bayestutorial/lib/python3.6/site-packages/arviz/data/base.py", line 25, in wrapped
return func(cls, *args, **kwargs)
File "/Users/canyon/.miniconda3/envs/bayestutorial/lib/python3.6/site-packages/arviz/data/io_pymc3.py", line 78, in sample_stats_to_xarray
log_likelihood, dims = self._extract_log_likelihood()
File "/Users/canyon/.miniconda3/envs/bayestutorial/lib/python3.6/site-packages/arviz/data/base.py", line 25, in wrapped
return func(cls, *args, **kwargs)
File "/Users/canyon/.miniconda3/envs/bayestutorial/lib/python3.6/site-packages/arviz/data/io_pymc3.py", line 56, in _extract_log_likelihood
chain_likelihoods.append(np.stack(log_like))
File "/Users/canyon/.miniconda3/envs/bayestutorial/lib/python3.6/site-packages/numpy/core/shape_base.py", line 410, in stack
arrays = [asanyarray(arr) for arr in arrays]
File "/Users/canyon/.miniconda3/envs/bayestutorial/lib/python3.6/site-packages/numpy/core/shape_base.py", line 410, in
arrays = [asanyarray(arr) for arr in arrays]
File "/Users/canyon/.miniconda3/envs/bayestutorial/lib/python3.6/site-packages/arviz/data/io_pymc3.py", line 55, in
log_like = (log_likelihood_vals_point(point) for point in self.trace.points([chain]))
File "/Users/canyon/.miniconda3/envs/bayestutorial/lib/python3.6/site-packages/arviz/data/io_pymc3.py", line 51, in log_likelihood_vals_point
return np.concatenate(log_like_vals)
ValueError: zero-dimensional arrays cannot be concatenated

Expected behavior
An Inferencedata object should be created
Additional context

[ahartikainen]

In this func

    def log_likelihood_vals_point(point):
            """Compute log likelihood for each observed point."""
            log_like_vals = []
            for var, log_like in cached:
                log_like_val = log_like(point)
                if var.missing_values:
                    log_like_val = log_like_val[~var.observations.mask]
                log_like_vals.append(log_like_val)
            return np.concatenate(log_like_vals)

log_like(point) returns a 0-dim array (np.array(num)), which can't be concatenated.

One option is to wrap that to np.atleast_1d(log_like(point))

[canyon289]

Fix makes sense, but moreso how this started appearing in the scipy tutorial when the env was frozen. Maybe it'll just be a mystery :)

[ahartikainen]

This change comes from pymc, not arviz (?)

[canyon289]

Tutorial has pinned version? No matter, I'll just fix and move on with life. Thanks @ahartikainen

[canyon289]

Came from ArviZ. I'll wrap like you suggested

95c9888

[ahartikainen]

True. Ravel makes 0d -> 1d?

[ColCarroll]

Closing since it seems fixed.