Experimental optimisations, proposed by ChatGPT.

[neilwalkinshaw]

Prompt: "Can you speed this up?", pasting CausalDag.enumerate_minimal_adjustment_sets

Summary of changes:

moral_graph[t] instead of nx.neighbors(...)
Accesses adjacency dict directly, faster in memory

update(...) instead of union with comprehension
Avoids building intermediate sets

Eliminated intermediate list conversions
Avoids unnecessary copies

Variable renaming (pbd_graph, ancestor_graph, etc.)
Improves clarity and prevents redundant calls

Same applied to list_all_min_sep

Summary of changes:

graph[node] instead of nx.neighbors(graph, node)
Faster adjacency access in memory

treatment_component = ... break
Avoids unnecessary iteration after finding the component

sample(sorted(...), 1)[0]
Returns a value, not a set, avoiding later unpacking

Removed repeated set(...) wrapping
Reduces GC and memory allocations

Same applied to constructive_backdoor_criterion

Avoids repeated set(self.nodes)
self.nodes is probably already a set-like iterable

Combines descendant updates efficiently
Reduces overhead of set.union with unpacking

Avoids constructing logger message unless needed
Significant savings if logging level > INFO

Clearer and faster condition check with & (set intersection)
Cleaner than issubset(difference(...))

[github-actions]

🦙 MegaLinter status: ✅ SUCCESS

Descriptor	Linter	Files	Fixed	Errors	Elapsed time
✅ PYTHON	black	32		0	0.95s
✅ PYTHON	pylint	32		0	5.45s

See detailed report in MegaLinter reports

MegaLinter is graciously provided by

[codecov]

Codecov Report

❌ Patch coverage is 98.75000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 95.67%. Comparing base (ad50cfb) to head (9816b98).
⚠️ Report is 17 commits behind head on main.

Files with missing lines	Patch %	Lines
causal_testing/specification/causal_dag.py	98.63%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #337      +/-   ##
==========================================
- Coverage   95.78%   95.67%   -0.12%     
==========================================
  Files          27       27              
  Lines        1638     1618      -20     
==========================================
- Hits         1569     1548      -21     
- Misses         69       70       +1     

Files with missing lines	Coverage Δ
causal_testing/main.py	96.17% <100.00%> (ø)
...sal_testing/surrogate/causal_surrogate_assisted.py	100.00% <100.00%> (ø)
causal_testing/testing/metamorphic_relation.py	100.00% <100.00%> (ø)
causal_testing/specification/causal_dag.py	98.88% <98.63%> (-0.62%)	⬇️

---

Continue to review full report in Codecov by Sentry.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 756203c...9816b98. Read the comment docs.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[jmafoster1]

I've been through this. Most of the optimisations were fine. Some were a little odd. I also took the opportunity to add some extra ones of my own. Also, most notably, I made this class a proper subclass of nx.DiGraph so we don't have to keep accessing .graph all the time. It doesn't make it any faster, but it does make it much cleaner.

[f-allian]

Suggestions look all reasonable to me. This is exactly what I was looking into some time ago when I tried to speed up the identification for large DAGs (see related issue #259). The use of generators makes sense here, and removing the NetworkX bottlenecks will also likely speed things up. But without doing a detailed profiling analysis into the before/after, we won't know exactly how much this will have optimised the identification of large DAGs.

[jmafoster1]

we won't know exactly how much this will have optimised the identification of large DAGs.

My admittedly limited test runs indicate not much! 🙃 But at least the code is cleaner now.