File size: 5,094 Bytes
b965a6e
 
 
 
 
 
 
 
 
 
 
 
 
29a0e42
b965a6e
29a0e42
 
 
b965a6e
29a0e42
 
b965a6e
 
 
29a0e42
 
 
b965a6e
 
 
 
29a0e42
b965a6e
 
 
 
 
29a0e42
b965a6e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ddf1ba7
29a0e42
 
 
b965a6e
 
 
 
 
 
 
 
29a0e42
 
 
 
 
 
b965a6e
29a0e42
b965a6e
29a0e42
 
b965a6e
 
29a0e42
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b965a6e
29a0e42
 
 
 
 
 
 
b965a6e
29a0e42
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
#
# 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.
"""Accuracy metric for the Test of Time benchmark by Bahar et al. (2025)."""

import ast
import json
from typing import Literal

import datasets
import evaluate

_CITATION = """\
@InProceedings{huggingface:module,
title = {Test of Time Accuracy},
authors={Auss Abbood},
year={2025}
}
"""

_DESCRIPTION = """\
The Test of Time (ToT) benchmarks expects models format their answers as a JSON with an explanation field and an answer field that follows a predefined format. The metrics extracts JSONs objects from the model's output, retains only the first JSON, drops the explanation field and compares it with the reference answer.
"""


# TODO: Add description of the arguments of the module here
_KWARGS_DESCRIPTION = """
Compares the extracted answer from the model's output with the reference answer.
Args:
    predictions: list of predictions to score. Each predictions
        should be a string with tokens separated by spaces.
    references: list of reference for each prediction. Each
        reference should be a string with tokens separated by spaces.
Returns:
    accuracy: description of the first score,
    another_score: description of the second score,
Examples:
    Examples should be written in doctest format, and should illustrate how
    to use the function.

    >>> my_new_module = evaluate.load("my_new_module")
    >>> results = my_new_module.compute(references=[0, 1], predictions=[0, 1])
    >>> print(results)
    {'accuracy': 1.0}
"""


@evaluate.utils.file_utils.add_start_docstrings(_DESCRIPTION, _KWARGS_DESCRIPTION)
class TestOfTimeAccuracy(evaluate.Metric):
    """Accuracy metric for the Test of Time benchmark by Bahar et al. (2025)."""

    __test__ = False

    def _info(self):
        return evaluate.MetricInfo(
            module_type="metric",
            description=_DESCRIPTION,
            citation=_CITATION,
            inputs_description=_KWARGS_DESCRIPTION,
            # This defines the format of each prediction and reference
            features=datasets.Features(
                {
                    "predictions": datasets.Value("string"),
                    "references": datasets.Value("string"),
                }
            ),
            # Homepage of the module for documentation
            # homepage="http://module.homepage",
            # Additional links to the codebase or references
            # codebase_urls=["http://github.com/path/to/codebase/of/new_module"],
            # reference_urls=["http://path.to.reference.url/new_module"],
        )

    @staticmethod
    def _extract_first_json_object(s: str) -> dict | None:
        decoder = json.JSONDecoder()
        idx, end = 0, len(s)
        while idx < end:
            try:
                obj, next_idx = decoder.raw_decode(s, idx)
                idx = next_idx
                if isinstance(obj, dict):
                    return obj
            except ValueError:
                idx += 1
        return None

    @staticmethod
    def _pop_explanation(d):
        if isinstance(d, dict):
            d.pop("explanation", None)
        return d

    @staticmethod
    def _get_answer(d):
        if isinstance(d, dict):
            return d.get("answer", None)
        return d

    @staticmethod
    def _parse_label(s):
        """Parses a string that could be a JSON object or a Python dict."""
        try:
            return json.loads(s)
        except json.JSONDecodeError:
            try:
                # Safe: only parses literals, does not execute code
                return ast.literal_eval(s)
            except (ValueError, SyntaxError):
                return None

    def _compute(
        self,
        predictions,
        references,
        subset: Literal["arithmetic", "semantic"],
        return_average: bool = True,
    ):
        """Returns the scores"""
        predictions = [self._extract_first_json_object(p) for p in predictions]
        if subset == "semantic":
            predictions = [self._get_answer(p) for p in predictions]
        elif subset == "arithmetic":
            predictions = [self._pop_explanation(p) for p in predictions]
            references = [self._parse_label(r) for r in references]
        else:
            raise ValueError(f"Invalid subset: {subset}")
        accuracy = [i == j for i, j in zip(predictions, references)]
        if return_average:
            return {"accuracy": sum(accuracy) / len(accuracy)}
        return {"accuracy": accuracy}