Source code for eval.utils

# from hotpotqa github
import re

import string
from collections import Counter




def normalize_answer(s):

    def remove_articles(text):
        return re.sub(r"\b(a|an|the)\b", " ", text)

    def white_space_fix(text):
        return " ".join(text.split())

    def remove_punc(text):
        exclude = set(string.punctuation)
        return "".join(ch for ch in text if ch not in exclude)

    def lower(text):
        return text.lower()

    return white_space_fix(remove_articles(remove_punc(lower(s))))





def f1_score(y: str, y_gt: str) -> float:
    if not isinstance(y, str) or not isinstance(y_gt, str):
        raise ValueError(f"y: {y},{type(y)}, y_gt: {y_gt},{type(y_gt)} must be string.")
    prediction_tokens = normalize_answer(y).split()
    ground_truth_tokens = normalize_answer(y_gt).split()

    common = Counter(prediction_tokens) & Counter(ground_truth_tokens)
    num_same = sum(common.values())

    if len(prediction_tokens) == len(ground_truth_tokens) == 0:
        # Unlike most tasks, QReCC and SQuAD-2.0 assign 1.0 in this edge case. We don't for uniformity.
        print(
            "\n#> F1 Metric: Rare edge case of len(prediction_tokens) == len(ground_truth_tokens) == 0.\n"
        )

    if num_same == 0:
        return 0

    precision = 1.0 * num_same / len(prediction_tokens)
    recall = 1.0 * num_same / len(ground_truth_tokens)
    f1 = (2 * precision * recall) / (precision + recall)

    return f1