'model_to_graphviz' throws TypeError when using shared variables

[sidravi1]

pm.model_to_graphviz gives a TypeError with sharedVariable

Please provide a minimal, self-contained, and reproducible example.

import pymc3 as pm
import numpy as np
import theano as th

# some data
x = np.linspace(0, 10, 100)
y = x * 5 + np.random.normal(0, 0.2)

x_shared = th.shared(x)
y_shared = th.shared(y)

# simple model
with pm.Model() as test:
    
    beta = pm.Normal('beta', 0, 10)
    _y = beta * x_shared 
    
    sigma = pm.HalfNormal('sigma', 2)
    obs = pm.Normal('obs', _y, sigma, observed=y_shared)

# generate DAG
pm.model_to_graphviz(test)

Please provide the full traceback.

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-13-1e5201be2a87> in <module>
----> 1 pm.model_to_graphviz(test)

~/anaconda3/envs/cotopaxi/lib/python3.6/site-packages/pymc3/model_graph.py in model_to_graphviz(model)
    184     """
    185     model = pm.modelcontext(model)
--> 186     return ModelGraph(model).make_graph()

~/anaconda3/envs/cotopaxi/lib/python3.6/site-packages/pymc3/model_graph.py in make_graph(self)
    154         graph = graphviz.Digraph(self.model.name)
    155         for shape, var_names in self.get_plates().items():
--> 156             label = ' x '.join(map('{:,d}'.format, shape))
    157             if label:
    158                 # must be preceded by 'cluster' to get a box around it

TypeError: unsupported format string passed to TensorVariable.__format__

Please provide any additional information below.

The fix seems straightforward. Just need an extra line:

if isinstance(shape, th.tensor.sharedvar.SharedVariable):
    shape = shape.eval()

before line 156 in model_graph.py

Versions and main components

• PyMC3 Version: 3.6
• Theano Version: 1.0.4
• Python Version: 3.6
• Operating system: Ubuntu
• How did you install PyMC3: (conda/pip) conda

[ColCarroll]

Thanks for report and fix (which looks right to me). Do you have any interest/time in submitting it?

[sidravi1]

Would love to! :)

[ColCarroll]

Fixed by #3490