evaluate/metrics/squad_v2/squad_v2.py

# Copyright 2020 The HuggingFace Evaluate Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" SQuAD v2 metric. """

import datasets

import evaluate

from .compute_score import (
    apply_no_ans_threshold,
    find_all_best_thresh,
    get_raw_scores,
    make_eval_dict,
    make_qid_to_has_ans,
    merge_eval,
)


_CITATION = """\
@inproceedings{Rajpurkar2016SQuAD10,
  title={SQuAD: 100, 000+ Questions for Machine Comprehension of Text},
  author={Pranav Rajpurkar and Jian Zhang and Konstantin Lopyrev and Percy Liang},
  booktitle={EMNLP},
  year={2016}
}
"""

_DESCRIPTION = """
This metric wrap the official scoring script for version 2 of the Stanford Question
Answering Dataset (SQuAD).

Stanford Question Answering Dataset (SQuAD) is a reading comprehension dataset, consisting of questions posed by
crowdworkers on a set of Wikipedia articles, where the answer to every question is a segment of text, or span,
from the corresponding reading passage, or the question might be unanswerable.

SQuAD2.0 combines the 100,000 questions in SQuAD1.1 with over 50,000 unanswerable questions
written adversarially by crowdworkers to look similar to answerable ones.
To do well on SQuAD2.0, systems must not only answer questions when possible, but also
determine when no answer is supported by the paragraph and abstain from answering.
"""

_KWARGS_DESCRIPTION = """
Computes SQuAD v2 scores (F1 and EM).
Args:
    predictions: List of triple for question-answers to score with the following elements:
        - the question-answer 'id' field as given in the references (see below)
        - the text of the answer
        - the probability that the question has no answer
    references: List of question-answers dictionaries with the following key-values:
            - 'id': id of the question-answer pair (see above),
            - 'answers': a list of Dict {'text': text of the answer as a string}
    no_answer_threshold: float
        Probability threshold to decide that a question has no answer.
Returns:
    'exact': Exact match (the normalized answer exactly match the gold answer)
    'f1': The F-score of predicted tokens versus the gold answer
    'total': Number of score considered
    'HasAns_exact': Exact match (the normalized answer exactly match the gold answer)
    'HasAns_f1': The F-score of predicted tokens versus the gold answer
    'HasAns_total': Number of score considered
    'NoAns_exact': Exact match (the normalized answer exactly match the gold answer)
    'NoAns_f1': The F-score of predicted tokens versus the gold answer
    'NoAns_total': Number of score considered
    'best_exact': Best exact match (with varying threshold)
    'best_exact_thresh': No-answer probability threshold associated to the best exact match
    'best_f1': Best F1 (with varying threshold)
    'best_f1_thresh': No-answer probability threshold associated to the best F1
Examples:

    >>> predictions = [{'prediction_text': '1976', 'id': '56e10a3be3433e1400422b22', 'no_answer_probability': 0.}]
    >>> references = [{'answers': {'answer_start': [97], 'text': ['1976']}, 'id': '56e10a3be3433e1400422b22'}]
    >>> squad_v2_metric = evaluate.load("squad_v2")
    >>> results = squad_v2_metric.compute(predictions=predictions, references=references)
    >>> print(results)
    {'exact': 100.0, 'f1': 100.0, 'total': 1, 'HasAns_exact': 100.0, 'HasAns_f1': 100.0, 'HasAns_total': 1, 'best_exact': 100.0, 'best_exact_thresh': 0.0, 'best_f1': 100.0, 'best_f1_thresh': 0.0}
"""


@evaluate.utils.file_utils.add_start_docstrings(_DESCRIPTION, _KWARGS_DESCRIPTION)
class SquadV2(evaluate.Metric):
    def _info(self):
        return evaluate.MetricInfo(
            description=_DESCRIPTION,
            citation=_CITATION,
            inputs_description=_KWARGS_DESCRIPTION,
            features=datasets.Features(
                {
                    "predictions": {
                        "id": datasets.Value("string"),
                        "prediction_text": datasets.Value("string"),
                        "no_answer_probability": datasets.Value("float32"),
                    },
                    "references": {
                        "id": datasets.Value("string"),
                        "answers": datasets.features.Sequence(
                            {"text": datasets.Value("string"), "answer_start": datasets.Value("int32")}
                        ),
                    },
                }
            ),
            codebase_urls=["https://rajpurkar.github.io/SQuAD-explorer/"],
            reference_urls=["https://rajpurkar.github.io/SQuAD-explorer/"],
        )

    def _compute(self, predictions, references, no_answer_threshold=1.0):
        no_answer_probabilities = {p["id"]: p["no_answer_probability"] for p in predictions}
        dataset = [{"paragraphs": [{"qas": references}]}]
        predictions = {p["id"]: p["prediction_text"] for p in predictions}

        qid_to_has_ans = make_qid_to_has_ans(dataset)  # maps qid to True/False
        has_ans_qids = [k for k, v in qid_to_has_ans.items() if v]
        no_ans_qids = [k for k, v in qid_to_has_ans.items() if not v]

        exact_raw, f1_raw = get_raw_scores(dataset, predictions)
        exact_thresh = apply_no_ans_threshold(exact_raw, no_answer_probabilities, qid_to_has_ans, no_answer_threshold)
        f1_thresh = apply_no_ans_threshold(f1_raw, no_answer_probabilities, qid_to_has_ans, no_answer_threshold)
        out_eval = make_eval_dict(exact_thresh, f1_thresh)

        if has_ans_qids:
            has_ans_eval = make_eval_dict(exact_thresh, f1_thresh, qid_list=has_ans_qids)
            merge_eval(out_eval, has_ans_eval, "HasAns")
        if no_ans_qids:
            no_ans_eval = make_eval_dict(exact_thresh, f1_thresh, qid_list=no_ans_qids)
            merge_eval(out_eval, no_ans_eval, "NoAns")
        find_all_best_thresh(out_eval, predictions, exact_raw, f1_raw, no_answer_probabilities, qid_to_has_ans)
        return dict(out_eval)