canho's picture
Add files using upload-large-folder tool
3c332d2 verified
Raw History Blame Contribute Delete
1.3 kB
from researchbench.clients import MockClient
from researchbench.ranking import parse_selection, run_ranking_record, score_ranking
def test_parse_selection_handles_strict_and_loose_outputs() -> None:
assert parse_selection("**Selection of research hypothesis candidate**: candidate 2") == 2
assert parse_selection("I choose candidate 1 because it is clearer.") == 1
assert parse_selection("no valid selection") is None
def test_ranking_order_reversal_rank_threshold_and_position_bias_stats() -> None:
record = {
"sample_id": "sample/1",
"research_question": "question",
"gold_hypothesis": "gold",
"fake_negative_hypotheses": [f"fake {idx}" for idx in range(5)],
"model_negative_hypotheses": [f"model {idx}" for idx in range(10)],
}
pred = run_ranking_record(record, MockClient(), order="both")
assert pred["orders"]["res"]["rank"] == 16
assert pred["orders"]["res"]["gold_wins"] is True
assert pred["orders"]["fan_1_res"]["rank"] == 1
assert pred["orders"]["fan_1_res"]["gold_wins"] is False
result = score_ranking([pred])
assert result["summary"]["overall_accuracy"] == 0.5
assert result["summary"]["sum_counts"] == {1: 1}
assert result["summary"]["pair_counts"] == {"self_contradictory": 15}