do operator on evidence

[zzzrbx]

When using the .query method to obtain the effect of Y on X using the do operator, and some evidence Z, is it possible that depending on which nodes are in Z it may not be possible to isolate the causal path from association paths? Would the API return an error?

[ankurankan]

@zzzrbx Unfortunately, this doesn't throw an error at the moment and would return incorrect results.

A minimal reproducible example for reference:

In [29]: model = BayesianNetwork([('X', 'Y'), ('X', 'Z'), ('Y', 'Z')])

In [30]: model_cpds = model.get_random_cpds(n_states=3)

In [31]: from pgmpy.inference import CausalInference

In [32]: infer.query(['Y'], do={'X': 1}, evidence={'Z': 0})
Out[32]: <DiscreteFactor representing phi(Y:3) at 0x7faab2a0e500>