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
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213 | """All the Hugging Face metrics used in EuroEval."""
import collections.abc as c
import logging
import typing as t
from pathlib import Path
import evaluate
import numpy as np
from datasets import DownloadConfig
from ..utils import HiddenPrints
from .base import Metric
if t.TYPE_CHECKING:
from datasets.arrow_dataset import Dataset
from evaluate import EvaluationModule
from ..data_models import BenchmarkConfig, DatasetConfig
logger: logging.Logger = logging.getLogger("euroeval")
class HuggingFaceMetric(Metric):
"""A metric which is implemented in the `evaluate` package.
Attributes:
name:
The name of the metric in snake_case.
pretty_name:
The pretty name of the metric, used for display purposes.
huggingface_id:
The Hugging Face ID of the metric.
results_key:
The name of the key used to extract the metric scores from the results
dictionary.
compute_kwargs:
Keyword arguments to pass to the metric's compute function. Defaults to
an empty dictionary.
"""
def __init__(
self,
name: str,
pretty_name: str,
huggingface_id: str,
results_key: str,
compute_kwargs: dict[str, t.Any] | None = None,
postprocessing_fn: t.Callable[[float], tuple[float, str]] | None = None,
) -> None:
"""Initialise the Hugging Face metric.
Args:
name:
The name of the metric in snake_case.
pretty_name:
The pretty name of the metric, used for display purposes.
huggingface_id:
The Hugging Face ID of the metric.
results_key:
The name of the key used to extract the metric scores from the results
dictionary.
compute_kwargs:
Keyword arguments to pass to the metric's compute function. Defaults to
an empty dictionary.
postprocessing_fn:
A function to apply to the metric scores after they are computed, taking
the score to the postprocessed score along with its string
representation. Defaults to x -> (100 * x, f"{x:.2%}").
"""
super().__init__(
name=name, pretty_name=pretty_name, postprocessing_fn=postprocessing_fn
)
self.huggingface_id = huggingface_id
self.results_key = results_key
self.compute_kwargs: dict[str, t.Any] = (
dict() if compute_kwargs is None else compute_kwargs
)
self.metric: "EvaluationModule | None" = None
def download(self, cache_dir: str) -> "HuggingFaceMetric":
"""Initiates the download of the metric if needed.
Args:
cache_dir:
The directory where the metric will be downloaded to.
Returns:
The metric object itself.
"""
# Annoying but needed to make the metric download to a different cache dir
download_config = DownloadConfig(cache_dir=Path(cache_dir, "evaluate"))
self.metric = evaluate.load(
path=self.huggingface_id, download_config=download_config
)
return self
def __call__(
self,
predictions: c.Sequence,
references: c.Sequence,
dataset: "Dataset",
dataset_config: "DatasetConfig",
benchmark_config: "BenchmarkConfig",
) -> float | None:
"""Calculate the metric score.
Args:
predictions:
The model predictions.
references:
The ground truth references.
dataset:
The dataset used for evaluation. This is only used in case any
additional metadata is used to compute the metrics.
dataset_config:
The dataset configuration.
benchmark_config:
The benchmark configuration.
Returns:
The calculated metric score, or None if the score should be ignored.
"""
if self.metric is None:
self.download(cache_dir=benchmark_config.cache_dir)
assert self.metric is not None
with HiddenPrints():
results = self.metric.compute(
predictions=predictions, references=references, **self.compute_kwargs
)
# The metric returns None if we are running on multi-GPU and the current
# process is not the main process
if results is None:
return None
# Convert the results to a float score
score = results[self.results_key]
if isinstance(score, list):
score = sum(score) / len(score)
if isinstance(score, np.floating):
score = float(score)
return score
mcc_metric = HuggingFaceMetric(
name="mcc",
pretty_name="Matthew's Correlation Coefficient",
huggingface_id="matthews_correlation",
results_key="matthews_correlation",
)
macro_f1_metric = HuggingFaceMetric(
name="macro_f1",
pretty_name="Macro-average F1-score",
huggingface_id="f1",
results_key="f1",
compute_kwargs=dict(average="macro"),
)
micro_f1_metric = HuggingFaceMetric(
name="micro_f1",
pretty_name="Micro-average F1-score with MISC tags",
huggingface_id="seqeval",
results_key="overall_f1",
)
micro_f1_no_misc_metric = HuggingFaceMetric(
name="micro_f1_no_misc",
pretty_name="Micro-average F1-score without MISC tags",
huggingface_id="seqeval",
results_key="overall_f1",
)
f1_metric = HuggingFaceMetric(
name="f1",
pretty_name="F1-score",
huggingface_id="squad_v2",
results_key="f1",
postprocessing_fn=lambda x: (x, f"{x:.2f}%"),
)
em_metric = HuggingFaceMetric(
name="em",
pretty_name="Exact Match",
huggingface_id="squad_v2",
results_key="exact",
postprocessing_fn=lambda x: (x, f"{x:.2f}%"),
)
bert_score_metric = HuggingFaceMetric(
name="bertscore",
pretty_name="BERTScore",
huggingface_id="bertscore",
results_key="f1",
compute_kwargs=dict(
model_type="microsoft/mdeberta-v3-base", device="cpu", batch_size=16
),
)
rouge_l_metric = HuggingFaceMetric(
name="rouge_l", pretty_name="ROUGE-L", huggingface_id="rouge", results_key="rougeL"
)
accuracy_metric = HuggingFaceMetric(
name="accuracy",
pretty_name="Accuracy",
huggingface_id="accuracy",
results_key="accuracy",
)
|