BUG: Deterministic variables with dims containing None break model_to_graphviz

[michaelosthege]

Describe the issue:

When a Deterministic variable is created, one may pass a dims=(None, "time") which is a valid dims tuple.

However, when attempting a pm.model_to_graphviz later, a KeyError: None is encountered when accessing Model.dim_lengths.

Reproduceable code example:

import numpy as np
import pymc as pm

with pm.Model(coords=dict(time=np.arange(5))) as pmodel:
    data = at.as_tensor(np.ones((3, 5)))
    # This works:
    # pm.Normal("n", data, dims=(None, "time"))

    # This does not:
    pm.Deterministic("n", data, dims=(None, "time"))
    
pm.model_to_graphviz(pmodel)

Error message:

File pymc\model_graph.py:198, in <genexpr>(.0)
    195 v = self.model[var_name]
    196 if var_name in self.model.RV_dims:
    197     plate_label = " x ".join(
--> 198         f"{d} ({self._eval(self.model.dim_lengths[d])})"
    199         for d in self.model.RV_dims[var_name]
    200     )
    201 else:
    202     plate_label = " x ".join(map(str, self._eval(v.shape)))

KeyError: None

PyMC version information:

4.1.7

Context for the issue:

No response

[michaelosthege]

Using pm.ConstantData instead of pm.Deterministic breaks in a different way:

File pymc\data.py:716, in Data(name, value, dims, coords, export_index_as_coords, mutable, **kwargs)
    706         if not dname in model.dim_lengths:
    707             model.add_coord(
    708                 name=dname,
    709                 # Note: Coordinate values can't be taken from
   (...)
    713                 length=xshape[d],
    714             )
--> 716 model.add_random_variable(x, dims=dims)
    718 return x

File pymc\model.py:1552, in Model.add_random_variable(self, var, dims)
   1550             raise ValueError(f"Dimension {dim} is not specified in `coords`.")
   1551     if any(var.name == dim for dim in dims):
-> 1552         raise ValueError(f"Variable `{var.name}` has the same name as its dimension label.")
   1553     self._RV_dims[var.name] = dims
   1555 self.named_vars[var.name] = var

ValueError: Variable `None` has the same name as its dimension label.

[michaelosthege]

I think this might be a rather simple fix: Checking is None and then printing more useful information if available.