[Repo Assist] perf(gcm): avoid redundant get_ordered_predecessors calls and pre-allocated DataFrame in fitting_sampling

[github-actions]

🤖 This is an automated pull request from Repo Assist.

Summary

Two performance improvements in dowhy/gcm/fitting_sampling.py:

1. Eliminate redundant get_ordered_predecessors call in fit_causal_model_of_target

Previously, get_ordered_predecessors(graph, node) was called twice per non-root node during fit:

# Call 1 — inside .fit()
X=training_data[get_ordered_predecessors(causal_model.graph, target_node)].to_numpy()
# Call 2 — for PARENTS_DURING_FIT
causal_model.graph.nodes[target_node][PARENTS_DURING_FIT] = get_ordered_predecessors(causal_model.graph, target_node)

Fix: compute once and reuse:

ordered_predecessors = get_ordered_predecessors(causal_model.graph, target_node)
causal_model.causal_mechanism(target_node).fit(X=training_data[ordered_predecessors]..., ...)
causal_model.graph.nodes[target_node][PARENTS_DURING_FIT] = ordered_predecessors

For a graph with N nodes, this halves the number of sorted(graph.predecessors(node)) calls made per gcm.fit().

2. Replace pre-allocated DataFrame with dict of arrays in draw_samples

Previously, a DataFrame was pre-allocated with pd.DataFrame(np.empty(...)) — containing uninitialized data — then filled column-by-column, triggering repeated copy operations in pandas 2.x (copy-on-write).

Fix: accumulate samples in a Dict[node, np.ndarray] and create the DataFrame once at the end:

drawn_samples: Dict[Any, np.ndarray] = {}
for node in sorted_nodes:
    ...
    drawn_samples[node] = causal_mechanism.draw_samples(parent_data).squeeze()
return pd.DataFrame(drawn_samples, columns=sorted_nodes)

This also removes the now-unnecessary _parent_samples_of helper (which was private to fitting_sampling.py — different copies exist in whatif.py and _noise.py and are unchanged).

Root Cause

No bug — these are proactive performance improvements. The pre-allocated DataFrame approach worked but incurred unnecessary overhead.

Trade-offs

• No behaviour change: column order is preserved via columns=sorted_nodes
• No API change: public function signatures are unchanged
• Minor correctness improvement: avoids uninitialized data in the intermediate state

Test Status

Tests could not be run in this environment (no Python runtime with dependencies available). CI will validate. The changes are semantically equivalent transformations with no logic changes.

Relevant test files:

• tests/gcm/test_data_generator.py — exercises draw_samples
• tests/gcm/test_graph.py — exercises fit and draw_samples
• tests/gcm/test_confidence_intervals_cms.py

Generated by 🌈 Repo Assist, see workflow run. Learn more.

To install this agentic workflow, run

gh aw add githubnext/agentics/workflows/repo-assist.md@11c9a2c442e519ff2b427bf58679f5a525353f76