Implementation to calculate conditional independencies

[cthoyt]

Closes #24

This pull request implements an algorithm to calculate all conditional independencies in a given ADMG.

• Add outline of implementation, signature, and tests (@cthoyt; 1140eca)
• Add additional unit tests (@y0-causal-inference/y0-team, following Collect named ADMGs for use as examples #26)
• Implementation (@djinnome or @JosephCottam, please assign yourself)
• Remove references to the dsl in ConditionalIndependency
• Merge DSeparationJudgement and ConditionalIndependency

[cthoyt]

@JosephCottam I think it's quite mature - can you please take care of all of the linting warnings then I will give a proper review?

[JosephCottam]

@cthoyt I have addressed most of the formatting issues.

@cthoyt @djinnome I move that we ignore error N400 (backslash for line continuation). The PEP8 description centers on expressions with operator chaining. However, it is used for function chaining here. Though the warning can be cleared by adding parenthesis around the whole chain (see below), I think the line continuation version is more clear in what is going on.

    evidence = (pd.DataFrame(rows, columns=["left", "right", "given", "chi^2", "p", "dof"])
        .sort_values("p")
        .assign(**{"Holm–Bonferroni level": significance_level / pd.Series(range(len(rows) + 1, 0, -1))})
        .pipe(lambda df: df.assign(flagged=(df["p"] < df["Holm–Bonferroni level"])))
        .sort_values(["flagged", "dof"], ascending=False))```

[cthoyt]

@JosephCottam it looks like some of the tests started failing, but not related to code I changed. Are these non-deterministic, or is there something I missed?

[JosephCottam]

@cthoyt To the best of my knowledge, all of the tests are deterministic.

[JosephCottam]

@cthoyt Fixed the test case that started failing. The changes made to loading the asia example data broke that test case. It was just putting the file path into the dataframe, rather than loading the data from the file. Now it loads the data again!

[cthoyt]

@cthoyt Fixed the test case that started failing. The changes made to loading the asia example data broke that test case. It was just putting the file path into the dataframe, rather than loading the data from the file. Now it loads the data again!

Nice catch, sorry for the trouble

[cthoyt]

Thank you so much @JosephCottam for all of the effort! Looking forward to what's next