benchmark/tests/test_eval_utils.py

from pathlib import Path
import sys


ROOT = Path(__file__).resolve().parents[1]
if str(ROOT) not in sys.path:
    sys.path.insert(0, str(ROOT))

from eval_utils import grade_response, parse_valid_options  # noqa: E402


def test_mcq_extract_standard_answer_marker():
    question = "A. Red\nB. Blue\nC. Green\nD. Purple\nAnswer in format **Answer:X**."
    response = "**Answer:C**"
    graded = grade_response(question, "c", response)
    assert graded["task_type"] == "mcq"
    assert graded["pred"] == "c"
    assert graded["extract_ok"] is True
    assert graded["correct"] is True


def test_mcq_extract_supports_a_to_f():
    question = "A. Red\nB. Blue\nC. Green\nD. Purple\nE. Pink\nF. Orange\nAnswer in format **Answer:X**."
    response = "Final answer is F"
    graded = grade_response(question, "f", response)
    assert graded["pred"] == "f"
    assert graded["correct"] is True


def test_mcq_extract_from_long_reasoning_tail_option():
    question = "Options are A, B, C, and D."
    response = "I compared all paths. The best match is Option B."
    graded = grade_response(question, "b", response)
    assert graded["pred"] == "b"
    assert graded["correct"] is True


def test_mcq_parse_options_from_list_sentence():
    opts = parse_valid_options("The choices are A, B, C, and D.")
    assert opts == {"A", "B", "C", "D"}


def test_mcq_parse_options_from_slash_sentence():
    opts = parse_valid_options("Pick from A/B/C/D.")
    assert opts == {"A", "B", "C", "D"}


def test_mcq_invalid_option_not_in_question_choices():
    question = "A. First\nB. Second\nC. Third\nD. Fourth"
    response = "**Answer:E**"
    graded = grade_response(question, "d", response)
    assert graded["pred"] == ""
    assert graded["extract_ok"] is False
    assert graded["correct"] is False


def test_fill_extract_standard_marker():
    question = "Please report the full word. Answer in format **Answer:WORD**."
    response = "**Answer:UNPROVERBIAL**"
    graded = grade_response(question, "UNPROVERBIAL", response)
    assert graded["task_type"] == "fill_blank"
    assert graded["pred"] == "UNPROVERBIAL"
    assert graded["pred_norm"] == "unproverbial"
    assert graded["correct"] is True


def test_fill_extract_trailing_punctuation():
    question = "Please report the full word. Answer in format **Answer:WORD**."
    response = "Answer: unproverbial."
    graded = grade_response(question, "UNPROVERBIAL", response)
    assert graded["pred_norm"] == "unproverbial"
    assert graded["correct"] is True


def test_fill_extract_with_quotes():
    question = "Please report the full word. Answer in format **Answer:WORD**."
    response = "Final answer: “UNPROVERBIAL”"
    graded = grade_response(question, "UNPROVERBIAL", response)
    assert graded["pred_norm"] == "unproverbial"
    assert graded["correct"] is True


def test_fill_normalization_case_spaces_fullwidth():
    question = "Please report the full word. Answer in format **Answer:WORD**."
    response = "Answer:  ＵＮＰＲＯＶＥＲＢＩＡＬ  "
    graded = grade_response(question, "unproverbial", response)
    assert graded["pred_norm"] == "unproverbial"
    assert graded["answer_norm"] == "unproverbial"
    assert graded["correct"] is True


def test_fill_wrong_spelling_is_not_correct():
    question = "Please report the full word. Answer in format **Answer:WORD**."
    response = "Answer: UNPROVERBIA"
    graded = grade_response(question, "UNPROVERBIAL", response)
    assert graded["pred_norm"] == "unproverbia"
    assert graded["correct"] is False