umpirix/adapters/python/tests/utils/test_probability_normalization.py
caelum0x 4847e35451 OpenJev: open-source typed decision service
Typed decision/classification service with an in-process gte-small ONNX
embedding engine and lexical fallback, JS/Python SDKs, adapters, and the
otsone model family. Apache-2.0 (sub-packages keep their MIT licenses).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-25 00:12:36 +03:00

75 lines
2.4 KiB
Python

"""Probability normalization tests."""
import pytest
from openjev_adapters._utils.probability_normalization import (
normalize_probabilities_of_all_answers,
probability_debug_data,
)
@pytest.mark.parametrize(
"normalization_enabled,raw_probability,expected_probability,expected_originals,expected_max_error,expected_probability_errors",
[
(
False,
0.2,
0.2,
None,
0.6,
{"stars": 0.6, "genre": 0.6},
),
(
True,
0.2,
0.5,
{
"stars": {"0": 0.2, "1": 0.2},
"genre": {"fiction": 0.2, "nonfiction": 0.2},
},
0.6,
{"stars": 0.6, "genre": 0.6},
),
(False, 0.50000025, 0.50000025, None, 5e-7, {}),
],
)
def test_probability_normalization_and_debug_data(
normalization_enabled: bool,
raw_probability: float,
expected_probability: float,
expected_originals: dict[str, dict[str, float]] | None,
expected_max_error: float,
expected_probability_errors: dict[str, float],
) -> None:
score_probabilities = {"0": raw_probability, "1": raw_probability}
choice_probabilities = {
"fiction": raw_probability,
"nonfiction": raw_probability,
}
probability_normalizations = {
"positive": None,
"stars": normalize_probabilities_of_all_answers(
["0", "1"],
score_probabilities,
"probabilities",
enabled=normalization_enabled,
),
"genre": normalize_probabilities_of_all_answers(
["fiction", "nonfiction"],
choice_probabilities,
"probabilities",
enabled=normalization_enabled,
),
}
assert probability_normalizations["stars"].probabilities == pytest.approx({"0": expected_probability, "1": expected_probability})
assert probability_normalizations["genre"].probabilities == pytest.approx(
{"fiction": expected_probability, "nonfiction": expected_probability}
)
debug_data = probability_debug_data(probability_normalizations)
assert debug_data["max_error"] == pytest.approx(expected_max_error)
assert debug_data["invalid_probs"] == len(expected_probability_errors)
assert debug_data["probability_errors"] == pytest.approx(expected_probability_errors)
assert debug_data.get("original_probabilities") == expected_originals