Graphviz changes variables in place

[ricardoV94]

Description

import pymc as pm

with pm.Model(coords_mutable={'i': [0]}) as m:
    beta_z = pm.ConstantData("beta_z", 0)
    beta_y = pm.ConstantData("beta_y", 0)
    z = pm.Bernoulli("z", p=pm.invlogit(beta_z), dims="i")
    y = pm.Normal("y", mu=beta_y + z, dims="i")

assert m["beta_y"].name == "beta_y"
assert m["beta_z"].name == "beta_z"
    
pm.model_to_graphviz(m)

assert m["beta_y"].name == "beta_y"
assert m["beta_z"].name == "beta_z"  # raises

Thin only happens with mutable coords, I imagine during evaluation.

[ricardoV94]

Actually this is a PyTensor issue: pymc-devs/pytensor#347

[ricardoV94]

Although possibly, model_graph could used the name in model.named_vars instead of the PyTensor variable name

[ricardoV94]

AFAICT there is no requirement in PyTensor for this behavior. Should be fixed by pymc-devs/pytensor#348