[pemalama12]

Contributor: Pema Sangpo Lama (@pemalama12)

Contribution Type: Bug fix with regression tests

Description: Integer arrays containing zeros produce NaN for KL divergence: assigning the 1e-6 floor into an integer copy truncates it to zero. Use np.clip to create copies that can represent the floor. This also fixes the time-series wrapper while preserving the inputs and MSE/MAE values.

Files to Review:

• pyhealth/metrics/regression.py: clipping fix.
• tests/core/test_regression_integer_kl.py: signed/unsigned integers, mixed dtypes, all-zero arrays, float compatibility, metric order, unchanged inputs, and the time-series wrapper.

Validation: 15 integer cases fail on master; all 5 new tests and 10 existing time-series tests pass with the fix. Ruff passes for the new tests, and tools/check_pr_rules.py passes.

Manual reproduction:

import numpy as np
from pyhealth.metrics import regression_metrics_fn

scores = regression_metrics_fn(np.array([0, 1]), np.array([1, 0]))
print(scores)

Expected: finite KL divergence ≈ 13.81548, MSE 1.0, MAE 1.0. Master returns NaN for KL.

Separate from #1324's docstring and shape-check changes.

[github-actions]

Test Coverage

Total: 73.30% (19606 / 26747 lines)

Modified file	Coverage	Lines
pyhealth/metrics/regression.py	85.29%	29 / 34

Workflow run · 10696dc

[solarsys]

Reviewed the diff and ran the full core suite locally with this PR applied; no new failures. Thanks!