Add comprehensive precomputed input pytest coverage

[MattScicluna]

No description provided.

[github-actions]

Pull Request Test Coverage Report for Build 22677665907

Details

• 102 of 103 (99.03%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.6%) to 87.504%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
Python/test/test_precomputed.py	102	103	99.03%

Totals
Change from base Build 21373206547:	0.6%
Covered Lines:	2437
Relevant Lines:	2785

---

💛 - Coveralls

[Copilot]

Pull request overview

Adds a new pytest module to validate PHATE’s handling of precomputed distance/affinity inputs, including alias behavior, sparse inputs, and expected error cases.

Changes:

• Introduces regression tests for knn_dist="precomputed" vs explicit precomputed_distance / precomputed_affinity.
• Adds coverage for sparse COO precomputed inputs.
• Adds error-path tests for invalid matrices and unsupported out-of-sample transforms.

Comments suppressed due to low confidence (7)

Python/test/test_precomputed.py:11

• Directly mutating sys.path inside tests is brittle and can mask packaging/import issues (e.g., importing the repo checkout instead of the installed distribution). Prefer configuring the test environment via proper packaging (editable install in CI), PYTHONPATH, or conftest.py/pytest configuration so imports resolve naturally.

import sys
import os

# Add parent directory to path
sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))

Python/test/test_precomputed.py:46

• These tests emit a lot of print() output. In pytest suites this adds noise and makes failures harder to scan in CI logs. Prefer relying on assertions alone, or use pytest’s reporting mechanisms (e.g., -vv test names) and only print diagnostic context when asserting/raising (or use caplog/logging if needed).

    print("\n" + "=" * 70)
    print("TEST 1: precomputed alias with random valid distance matrix")
    print("=" * 70)

Python/test/test_precomputed.py:34

• Setting n_jobs=-1 in unit tests can unnecessarily saturate CI runners and can introduce nondeterminism depending on internal parallel reductions. Consider using n_jobs=1 (or parametrizing) for test stability and predictable runtime.

    return dict(knn=7, t=10, n_jobs=-1, verbose=False, random_state=42)

Python/test/test_precomputed.py:81

• The tolerance atol=1e-10 is extremely strict for algorithms involving graph construction/eigendecompositions and can lead to flaky tests across BLAS/LAPACK implementations, OSes, or minor dependency changes. Use a more realistic tolerance (e.g., 1e-6/1e-7) and/or add rtol, and consider comparing a more stable invariant if exact embedding coordinates can vary.

    assert np.allclose(
        emb_alias, emb_distance, atol=1e-10
    ), "Alias and explicit distance mode diverged"

Python/test/test_precomputed.py:150

• The match=... patterns are tightly coupled to exact error-message wording (and could break on minor copy edits). Consider using a more flexible regex (e.g., matching key terms with case-insensitivity) or validating a more stable property (exception type + minimal distinctive substring) to reduce brittleness.

    with pytest.raises(ValueError, match="square matrix"):
        phate.PHATE(knn_dist="precomputed", **create_precomputed_kwargs()).fit_transform(
            D_non_square
        )

Python/test/test_precomputed.py:1

• For pytest test modules, a shebang and __main__ runner are typically unnecessary and can be undesirable in some test discovery/packaging contexts. It’s cleaner to rely on pytest invocation from the command line/CI and keep test files import-only.

#!/usr/bin/env python

Python/test/test_precomputed.py:198

• For pytest test modules, a shebang and __main__ runner are typically unnecessary and can be undesirable in some test discovery/packaging contexts. It’s cleaner to rely on pytest invocation from the command line/CI and keep test files import-only.

if __name__ == "__main__":
    pytest.main([__file__, "-v"])

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.