fix(safaa): warn when threshold argument cannot be honored

[Valyrian-Code]

Description

SafaaAgent.predict documents a configurable threshold parameter, but the shipped classifier (SGDClassifier(loss='hinge')) does not implement predict_proba. At runtime:

hasattr(self.false_positive_detector, "predict_proba")

returns False, execution falls through to the binary predict() branch, and threshold is silently ignored.

This PR emits a UserWarning when a non-default threshold is passed to a model that cannot honor it, and documents the limitation in the docstring. Default usage remains warning-free, so existing callers are unaffected.

Changes

• [Safaa.py](https://github.com/fossology/safaa/blob/main/Safaa/src/safaa/Safaa.py?utm_source=chatgpt.com)

  • add import warnings
  • emit UserWarning when threshold != 0.5 but the loaded model lacks predict_proba
  • update the predict() docstring to document the constraint and workaround (loss='log_loss' or 'modified_huber')

• tests/__init__.py, tests/test_safaa.py

  • add 9 pytest tests covering warning behavior, threshold semantics, and probabilistic-model behavior

• [pyproject.toml](https://github.com/fossology/safaa/blob/main/pyproject.toml?utm_source=chatgpt.com)

  • add pytest to dev dependencies

Test coverage

TestPredictThreshold verifies:

• no warning at the default threshold
• warnings for non-default thresholds (0.0, 1.0, 0.99, 0.01)
• warning text mentions predict_proba
• explicitly passing threshold=0.5 remains silent
• predictions still succeed when warnings are emitted
• threshold behavior works correctly when predict_proba is available
• the >= threshold boundary is inclusive
• no warning is emitted for probabilistic classifiers

How to test

poetry install
poetry run pytest tests/test_safaa.py -v

Manual reproduction

import warnings
from safaa.Safaa import SafaaAgent

agent = SafaaAgent()

with warnings.catch_warnings(record=True) as ws:
    warnings.simplefilter("always")
    agent.predict(["Copyright 2024 Foo"], threshold=0.3)

    print(ws[0].message)

Output:

UserWarning: The loaded false positive detector does not support
probability estimates (predict_proba); the 'threshold' argument is being ignored...

This closes #51.

Note: this PR shares the tests/ scaffolding introduced in #48 and #50. Depending on merge order, a small rebase may be needed for tests/__init__.py and the test module header.

[Valyrian-Code]

Hi @GMishx & @Kaushl2208

Hi! I opened a small PR with bug fixe and regression test. I’d appreciate a review whenever you have time thanks for maintaining the project!

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.