Been playing around with DoWhy 0.6 for a day
Thought I found a bug, turns out it was that they fixed a bug.
DoWhy 's 0.6 release removed
LINE 80: self._graph = self.add_unobserved_common_cause(observed_node_names)
from its dowhy/causal_graph.py. This means all the wrong "Unobserved Confounders" will be added to the corresponding node categories (i.e., "effect_modifiers") with zero tolerance, which is the expected behaviour.
I know this cuz I've been comparing v0.5.1 with the latest release for a whole day and found that I made a mistake in my DAG...
Anyway.