Upload mdia.py with huggingface_hub
Browse files
mdia.py
ADDED
@@ -0,0 +1,226 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# coding=utf-8
|
2 |
+
# Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor.
|
3 |
+
#
|
4 |
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
+
# you may not use this file except in compliance with the License.
|
6 |
+
# You may obtain a copy of the License at
|
7 |
+
#
|
8 |
+
# http://www.apache.org/licenses/LICENSE-2.0
|
9 |
+
#
|
10 |
+
# Unless required by applicable law or agreed to in writing, software
|
11 |
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
12 |
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
13 |
+
# See the License for the specific language governing permissions and
|
14 |
+
# limitations under the License.
|
15 |
+
|
16 |
+
from pathlib import Path
|
17 |
+
from typing import Dict, List, Tuple
|
18 |
+
|
19 |
+
import datasets
|
20 |
+
import pandas as pd
|
21 |
+
|
22 |
+
from seacrowd.utils.configs import SEACrowdConfig
|
23 |
+
from seacrowd.utils.constants import (SCHEMA_TO_FEATURES, TASK_TO_SCHEMA,
|
24 |
+
Licenses, Tasks)
|
25 |
+
|
26 |
+
_CITATION = """\
|
27 |
+
@misc{zhang2022mdia,
|
28 |
+
title={MDIA: A Benchmark for Multilingual Dialogue Generation in 46 Languages},
|
29 |
+
author={Qingyu Zhang and Xiaoyu Shen and Ernie Chang and Jidong Ge and Pengke Chen},
|
30 |
+
year={2022},
|
31 |
+
eprint={2208.13078},
|
32 |
+
archivePrefix={arXiv},
|
33 |
+
primaryClass={cs.CL}
|
34 |
+
}
|
35 |
+
"""
|
36 |
+
|
37 |
+
_DATASETNAME = "mdia"
|
38 |
+
|
39 |
+
_DESCRIPTION = """\
|
40 |
+
This is a multilingual benchmark for dialogue generation containing real-life
|
41 |
+
Reddit conversations (parent and response comment pairs) in 46 languages,
|
42 |
+
including Indonesian, Tagalog and Vietnamese. English translations are also
|
43 |
+
provided for comments.
|
44 |
+
"""
|
45 |
+
|
46 |
+
_HOMEPAGE = "https://github.com/DoctorDream/mDIA"
|
47 |
+
|
48 |
+
_LANGUAGES = ["ind", "tgl", "vie"]
|
49 |
+
|
50 |
+
_LICENSE = Licenses.CC_BY_4_0.value
|
51 |
+
|
52 |
+
_LOCAL = False
|
53 |
+
|
54 |
+
_URLS = {
|
55 |
+
"raw": "https://github.com/DoctorDream/mDIA/raw/master/datasets/raw.zip",
|
56 |
+
"translated": "https://github.com/DoctorDream/mDIA/raw/master/datasets/translated.zip",
|
57 |
+
}
|
58 |
+
|
59 |
+
_SUPPORTED_TASKS = [Tasks.DIALOGUE_SYSTEM, Tasks.MACHINE_TRANSLATION] # DS, MT
|
60 |
+
_SEACROWD_SCHEMA = {task.value: f"seacrowd_{str(TASK_TO_SCHEMA[task]).lower()}" for task in _SUPPORTED_TASKS} # t2t
|
61 |
+
_SUBSETS = [
|
62 |
+
"ind_dialogue",
|
63 |
+
"ind_eng",
|
64 |
+
"tgl_dialogue",
|
65 |
+
"tgl_eng",
|
66 |
+
"vie_dialogue",
|
67 |
+
"vie_eng",
|
68 |
+
]
|
69 |
+
|
70 |
+
_SOURCE_VERSION = "1.0.0"
|
71 |
+
|
72 |
+
_SEACROWD_VERSION = "2024.06.20"
|
73 |
+
|
74 |
+
|
75 |
+
class MdiaDataset(datasets.GeneratorBasedBuilder):
|
76 |
+
"""Multilingual benchmark for dialogue generation containing real-life Reddit conversations"""
|
77 |
+
|
78 |
+
SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
|
79 |
+
SEACROWD_VERSION = datasets.Version(_SEACROWD_VERSION)
|
80 |
+
|
81 |
+
BUILDER_CONFIGS = []
|
82 |
+
for subset in _SUBSETS:
|
83 |
+
if "dialogue" in subset:
|
84 |
+
BUILDER_CONFIGS += [
|
85 |
+
SEACrowdConfig(
|
86 |
+
name=f"{_DATASETNAME}_{subset}_source",
|
87 |
+
version=SOURCE_VERSION,
|
88 |
+
description=f"{_DATASETNAME} {subset} source schema",
|
89 |
+
schema="source",
|
90 |
+
subset_id=subset,
|
91 |
+
),
|
92 |
+
SEACrowdConfig(
|
93 |
+
name=f"{_DATASETNAME}_{subset}_{_SEACROWD_SCHEMA['DS']}",
|
94 |
+
version=SEACROWD_VERSION,
|
95 |
+
description=f"{_DATASETNAME} {subset} SEACrowd schema",
|
96 |
+
schema=_SEACROWD_SCHEMA["DS"],
|
97 |
+
subset_id=subset,
|
98 |
+
),
|
99 |
+
]
|
100 |
+
else:
|
101 |
+
BUILDER_CONFIGS += [
|
102 |
+
SEACrowdConfig(
|
103 |
+
name=f"{_DATASETNAME}_{subset}_source",
|
104 |
+
version=SOURCE_VERSION,
|
105 |
+
description=f"{_DATASETNAME} {subset} source schema",
|
106 |
+
schema="source",
|
107 |
+
subset_id=subset,
|
108 |
+
),
|
109 |
+
SEACrowdConfig(
|
110 |
+
name=f"{_DATASETNAME}_{subset}_{_SEACROWD_SCHEMA['MT']}",
|
111 |
+
version=SEACROWD_VERSION,
|
112 |
+
description=f"{_DATASETNAME} {subset} SEACrowd schema",
|
113 |
+
schema=_SEACROWD_SCHEMA["MT"],
|
114 |
+
subset_id=subset,
|
115 |
+
),
|
116 |
+
]
|
117 |
+
|
118 |
+
DEFAULT_CONFIG_NAME = f"{_DATASETNAME}_{_SUBSETS[0]}_source"
|
119 |
+
|
120 |
+
def _info(self) -> datasets.DatasetInfo:
|
121 |
+
if self.config.schema == "source":
|
122 |
+
features = datasets.Features(
|
123 |
+
{
|
124 |
+
"lang": datasets.Value("string"),
|
125 |
+
"title": datasets.Value("string"),
|
126 |
+
"source_body": datasets.Value("string"),
|
127 |
+
"target_body": datasets.Value("string"),
|
128 |
+
"link_id": datasets.Value("string"),
|
129 |
+
"source_id": datasets.Value("string"),
|
130 |
+
"target_id": datasets.Value("string"),
|
131 |
+
"translated_source_body": datasets.Value("string"),
|
132 |
+
"translated_target_body": datasets.Value("string"),
|
133 |
+
}
|
134 |
+
)
|
135 |
+
elif self.config.schema == _SEACROWD_SCHEMA["DS"]: # same schema with _SEACROWD_SCHEMA["MT"]
|
136 |
+
features = SCHEMA_TO_FEATURES[TASK_TO_SCHEMA[_SUPPORTED_TASKS[0]]] # text2text_features
|
137 |
+
|
138 |
+
return datasets.DatasetInfo(
|
139 |
+
description=_DESCRIPTION,
|
140 |
+
features=features,
|
141 |
+
homepage=_HOMEPAGE,
|
142 |
+
license=_LICENSE,
|
143 |
+
citation=_CITATION,
|
144 |
+
)
|
145 |
+
|
146 |
+
def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]:
|
147 |
+
"""Returns SplitGenerators."""
|
148 |
+
lang_map = {"ind": "id", "tgl": "tl", "vie": "vi"}
|
149 |
+
lang = lang_map[self.config.subset_id.split("_")[0]]
|
150 |
+
|
151 |
+
data_url = _URLS["translated"]
|
152 |
+
data_dir = Path(dl_manager.download_and_extract(data_url)) / "translated"
|
153 |
+
data_path = "{split}_data/{lang}2en_{split}.csv"
|
154 |
+
|
155 |
+
return [
|
156 |
+
datasets.SplitGenerator(
|
157 |
+
name=datasets.Split.TRAIN,
|
158 |
+
gen_kwargs={
|
159 |
+
"data_path": data_dir / data_path.format(split="train", lang=lang),
|
160 |
+
},
|
161 |
+
),
|
162 |
+
datasets.SplitGenerator(
|
163 |
+
name=datasets.Split.TEST,
|
164 |
+
gen_kwargs={
|
165 |
+
"data_path": data_dir / data_path.format(split="test", lang=lang),
|
166 |
+
},
|
167 |
+
),
|
168 |
+
datasets.SplitGenerator(
|
169 |
+
name=datasets.Split.VALIDATION,
|
170 |
+
gen_kwargs={
|
171 |
+
"data_path": data_dir / data_path.format(split="eval", lang=lang),
|
172 |
+
},
|
173 |
+
),
|
174 |
+
]
|
175 |
+
|
176 |
+
def _generate_examples(self, data_path: Path) -> Tuple[int, Dict]:
|
177 |
+
"""Yields examples as (key, example) tuples."""
|
178 |
+
df = pd.read_csv(data_path)
|
179 |
+
|
180 |
+
# source schema
|
181 |
+
if self.config.schema == "source":
|
182 |
+
for i, row in df.iterrows():
|
183 |
+
yield i, {
|
184 |
+
"lang": row["lang"],
|
185 |
+
"title": row["title"],
|
186 |
+
"source_body": row["source_body"],
|
187 |
+
"target_body": row["target_body"],
|
188 |
+
"link_id": row["link_id"],
|
189 |
+
"source_id": row["source_id"],
|
190 |
+
"target_id": row["target_id"],
|
191 |
+
"translated_source_body": row["translated_source_body"],
|
192 |
+
"translated_target_body": row["translated_target_body"],
|
193 |
+
}
|
194 |
+
|
195 |
+
# t2t schema for dialogue
|
196 |
+
elif "dialogue" in self.config.subset_id:
|
197 |
+
for i, row in df.iterrows():
|
198 |
+
yield i, {
|
199 |
+
"id": str(i),
|
200 |
+
"text_1": row["source_body"],
|
201 |
+
"text_2": row["target_body"],
|
202 |
+
"text_1_name": "source_body",
|
203 |
+
"text_2_name": "target_body",
|
204 |
+
}
|
205 |
+
|
206 |
+
# t2t schema for machine translation
|
207 |
+
elif "eng" in self.config.subset_id:
|
208 |
+
for i, row in df.iterrows():
|
209 |
+
for j in range(2):
|
210 |
+
idx = i * 2 + j
|
211 |
+
if j == 0:
|
212 |
+
yield idx, {
|
213 |
+
"id": str(idx),
|
214 |
+
"text_1": row["source_body"],
|
215 |
+
"text_2": row["translated_source_body"],
|
216 |
+
"text_1_name": "source_body",
|
217 |
+
"text_2_name": "translated_source_body",
|
218 |
+
}
|
219 |
+
else:
|
220 |
+
yield idx, {
|
221 |
+
"id": str(idx),
|
222 |
+
"text_1": row["target_body"],
|
223 |
+
"text_2": row["translated_target_body"],
|
224 |
+
"text_1_name": "target_body",
|
225 |
+
"text_2_name": "translated_target_body",
|
226 |
+
}
|