A function for enumerating d-separators in causal graphs about pywhy-graphs HOT 13 CLOSED

That's great to hear @jaron-lee . Adding you.

from pywhy-graphs.

Ah yeah, that pattern should be sufficient. Note: networkx/networkx#6216

I'm going to submit a PR to patch this edge-case behavior. We should also follow a similar design.

from pywhy-graphs.

Ah I see. Yeah that is a bug forsure. Let's raise the issue w/ them and see what they say. For now, on our end we can return None in the edge case that they are d-connected.

from pywhy-graphs.

I will take a stab at this next! BTW is there a way I can self assign an issue?

from pywhy-graphs.

Will look into it!

from pywhy-graphs.

Turns out GH only has this issue if you're a collaborator. No idea why.

Would you be interested in becoming an official collaborator on the repo?

Would just want your input and code reviews whenever relevant.

from pywhy-graphs.

Yeah I'd be happy to become an official collaborator, could probably learn some things on reviewing code.

from pywhy-graphs.

@adam2392 so it seems that I'll need to implement an anterior function on the way to implementing the m-separator function. Is there a strategy more generally for genealogical functions on MixedEdgeGraphs, in terms of where to implement them? I could just put it in m_separation.py but was not sure.

from pywhy-graphs.

I found the answer to my question in networkx.algorithms.d_separated - the template there is clear to follow (minimal d-separator functions implemented in the same file).

from pywhy-graphs.

Regarding that last comment Adam, I notice that minimal_d_separator doesn't seem to handle the edge case when you have e.g. A -> B and there is in fact no separator at all between A and B. Is this intended?

Van der Zander handle this by returning a special value (maybe in Python we could return None), which is probably what I'll do in minimal_m_separator.

from pywhy-graphs.

I think it may be fixed in networkx/networkx#6427, where we first check if in fact {z} is a d-separator for {x} and {y} at all. If it is, then we proceed to check its minimality. Can we do the same thing for m-separation?

from pywhy-graphs.

networkx/networkx#6427 only addresses this in the verification function is_minimal_d_separated but not the constructor function minimal_d_separator, as I understand.

In fact,

>>> import networkx as nx
>>> nx.__version__
'3.0'
>>> G = nx.DiGraph()
>>> G.add_edge("A", "B")
>>> nx.minimal_d_separator(G, "A", "B")
set()

but clearly the empty set does not d-separate A and B in that graph. Might this be a bug in networkx?

from pywhy-graphs.

Xref: networkx/networkx#6430

from pywhy-graphs.