holylovenia
commited on
Commit
•
de28af9
1
Parent(s):
65f42c0
Upload indosmd.py with huggingface_hub
Browse files- indosmd.py +273 -0
indosmd.py
ADDED
@@ -0,0 +1,273 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
from pathlib import Path
|
3 |
+
from typing import Dict, List, Tuple
|
4 |
+
|
5 |
+
import datasets
|
6 |
+
|
7 |
+
from seacrowd.utils import schemas
|
8 |
+
from seacrowd.utils.configs import SEACrowdConfig
|
9 |
+
from seacrowd.utils.constants import Tasks, Licenses
|
10 |
+
|
11 |
+
_CITATION = """\
|
12 |
+
@article{kautsar2023indotod,
|
13 |
+
author={Kautsar, Muhammad Dehan Al and Nurdini, Rahmah Khoirussyifa' and Cahyawijaya, Samuel and Winata, Genta Indra and Purwarianti, Ayu},
|
14 |
+
title={IndoToD: A Multi-Domain Indonesian Benchmark For End-to-End Task-Oriented Dialogue Systems},
|
15 |
+
journal={arXiv preprint arXiv:2311.00958},
|
16 |
+
year={2023},
|
17 |
+
}
|
18 |
+
"""
|
19 |
+
|
20 |
+
_LANGUAGES = ["ind"]
|
21 |
+
_LOCAL = False
|
22 |
+
|
23 |
+
_DATASETNAME = "indosmd"
|
24 |
+
|
25 |
+
_DESCRIPTION = """\
|
26 |
+
IndoSMD is a synthetic task-oriented dialogue system dataset that was translated from the In-Car Assistant (SMD) dataset (Eric et al., 2017) into the new Indonesian dataset using the translation pipeline method
|
27 |
+
including delexicalization, translation, and delexicalization. The dataset consists of 323 dialogues in the POI Navigation, Calendar Scheduling, and Weather Information Retrieval domain, with a user and an agent talking to each other.
|
28 |
+
It also consists of slots and dialogue acts from the user and the agent.
|
29 |
+
"""
|
30 |
+
|
31 |
+
_HOMEPAGE = "https://github.com/dehanalkautsar/IndoToD/tree/main/IndoSMD"
|
32 |
+
|
33 |
+
_LICENSE = Licenses.CC_BY_SA_4_0.value
|
34 |
+
|
35 |
+
_URLS = {
|
36 |
+
_DATASETNAME: {
|
37 |
+
"train": "https://raw.githubusercontent.com/dehanalkautsar/IndoToD/main/IndoSMD/IndoSMD_split/IndoSMD_train.json",
|
38 |
+
"validation": "https://raw.githubusercontent.com/dehanalkautsar/IndoToD/main/IndoSMD/IndoSMD_split/IndoSMD_dev.json",
|
39 |
+
"test": "https://raw.githubusercontent.com/dehanalkautsar/IndoToD/main/IndoSMD/IndoSMD_split/IndoSMD_test.json",
|
40 |
+
},
|
41 |
+
}
|
42 |
+
|
43 |
+
_SUPPORTED_TASKS = [Tasks.E2E_TASK_ORIENTED_DIALOGUE]
|
44 |
+
|
45 |
+
_SOURCE_VERSION = "1.0.0"
|
46 |
+
|
47 |
+
_SEACROWD_VERSION = "2024.06.20"
|
48 |
+
|
49 |
+
|
50 |
+
class IndoSMDDataset(datasets.GeneratorBasedBuilder):
|
51 |
+
"""IndoToD: A Multi-Domain Indonesian Benchmark For End-to-End Task-Oriented Dialogue Systems"""
|
52 |
+
|
53 |
+
SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
|
54 |
+
SEACROWD_VERSION = datasets.Version(_SEACROWD_VERSION)
|
55 |
+
|
56 |
+
BUILDER_CONFIGS = [
|
57 |
+
SEACrowdConfig(
|
58 |
+
name=f"{_DATASETNAME}_source",
|
59 |
+
version=SOURCE_VERSION,
|
60 |
+
description="IndoToD: IndoSMD source schema",
|
61 |
+
schema="source",
|
62 |
+
subset_id=f"{_DATASETNAME}",
|
63 |
+
),
|
64 |
+
SEACrowdConfig(
|
65 |
+
name=f"{_DATASETNAME}_seacrowd_tod",
|
66 |
+
version=SEACROWD_VERSION,
|
67 |
+
description="IndoToD: IndoSMD SEACrowd End-to-end Task Oriented Dialogue schema",
|
68 |
+
schema="seacrowd_tod",
|
69 |
+
subset_id=f"{_DATASETNAME}",
|
70 |
+
),
|
71 |
+
]
|
72 |
+
|
73 |
+
DEFAULT_CONFIG_NAME = "indosmd_source"
|
74 |
+
|
75 |
+
def _info(self) -> datasets.DatasetInfo:
|
76 |
+
if self.config.schema == "source":
|
77 |
+
features = datasets.Features(
|
78 |
+
{
|
79 |
+
"index": datasets.Value("string"),
|
80 |
+
"dialogue": [
|
81 |
+
{
|
82 |
+
"turn": datasets.Value("string"),
|
83 |
+
"data": {
|
84 |
+
"end_dialogue": datasets.Value("string"),
|
85 |
+
"utterance": datasets.Value("string"),
|
86 |
+
"delex_utterance": datasets.Value("string"),
|
87 |
+
"requested": {
|
88 |
+
"distance": datasets.Value("string"),
|
89 |
+
"traffic_info": datasets.Value("string"),
|
90 |
+
"poi_type": datasets.Value("string"),
|
91 |
+
"address": datasets.Value("string"),
|
92 |
+
"poi": datasets.Value("string"),
|
93 |
+
"room": datasets.Value("string"),
|
94 |
+
"agenda": datasets.Value("string"),
|
95 |
+
"time": datasets.Value("string"),
|
96 |
+
"date": datasets.Value("string"),
|
97 |
+
"party": datasets.Value("string"),
|
98 |
+
"event": datasets.Value("string"),
|
99 |
+
"weather_attribute": datasets.Value("string"),
|
100 |
+
"location": datasets.Value("string"),
|
101 |
+
},
|
102 |
+
"slots": {
|
103 |
+
"distance": datasets.Value("string"),
|
104 |
+
"traffic_info": datasets.Value("string"),
|
105 |
+
"poi_type": datasets.Value("string"),
|
106 |
+
"address": datasets.Value("string"),
|
107 |
+
"poi": datasets.Value("string"),
|
108 |
+
"room": datasets.Value("string"),
|
109 |
+
"agenda": datasets.Value("string"),
|
110 |
+
"time": datasets.Value("string"),
|
111 |
+
"date": datasets.Value("string"),
|
112 |
+
"party": datasets.Value("string"),
|
113 |
+
"event": datasets.Value("string"),
|
114 |
+
"weather_attribute": datasets.Value("string"),
|
115 |
+
"location": datasets.Value("string"),
|
116 |
+
},
|
117 |
+
},
|
118 |
+
}
|
119 |
+
],
|
120 |
+
"scenario": {
|
121 |
+
"kb": {
|
122 |
+
"items": [
|
123 |
+
{
|
124 |
+
"distance": datasets.Value("string"),
|
125 |
+
"traffic_info": datasets.Value("string"),
|
126 |
+
"poi_type": datasets.Value("string"),
|
127 |
+
"address": datasets.Value("string"),
|
128 |
+
"poi": datasets.Value("string"),
|
129 |
+
"room": datasets.Value("string"),
|
130 |
+
"agenda": datasets.Value("string"),
|
131 |
+
"time": datasets.Value("string"),
|
132 |
+
"date": datasets.Value("string"),
|
133 |
+
"party": datasets.Value("string"),
|
134 |
+
"event": datasets.Value("string"),
|
135 |
+
"monday": datasets.Value("string"),
|
136 |
+
"tuesday": datasets.Value("string"),
|
137 |
+
"wednesday": datasets.Value("string"),
|
138 |
+
"thursday": datasets.Value("string"),
|
139 |
+
"friday": datasets.Value("string"),
|
140 |
+
"saturday": datasets.Value("string"),
|
141 |
+
"sunday": datasets.Value("string"),
|
142 |
+
"today": datasets.Value("string"),
|
143 |
+
"location": datasets.Value("string"),
|
144 |
+
}
|
145 |
+
],
|
146 |
+
"column_names": [datasets.Value("string")],
|
147 |
+
"kb_title": datasets.Value("string"),
|
148 |
+
},
|
149 |
+
"task": {"intent": datasets.Value("string")},
|
150 |
+
"uuid": datasets.Value("string"),
|
151 |
+
},
|
152 |
+
}
|
153 |
+
)
|
154 |
+
elif self.config.schema == "seacrowd_tod":
|
155 |
+
features = schemas.tod_features
|
156 |
+
else:
|
157 |
+
raise NotImplementedError(f"Schema {self.config.schema} has not been implemented")
|
158 |
+
|
159 |
+
return datasets.DatasetInfo(
|
160 |
+
description=_DESCRIPTION,
|
161 |
+
features=features,
|
162 |
+
homepage=_HOMEPAGE,
|
163 |
+
license=_LICENSE,
|
164 |
+
citation=_CITATION,
|
165 |
+
)
|
166 |
+
|
167 |
+
def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]:
|
168 |
+
"""Returns SplitGenerators."""
|
169 |
+
|
170 |
+
urls = _URLS[_DATASETNAME]
|
171 |
+
data_dir = dl_manager.download_and_extract(urls)
|
172 |
+
|
173 |
+
return [
|
174 |
+
datasets.SplitGenerator(
|
175 |
+
name=datasets.Split.TRAIN,
|
176 |
+
gen_kwargs={
|
177 |
+
"filepath": data_dir["train"],
|
178 |
+
"split": "train",
|
179 |
+
},
|
180 |
+
),
|
181 |
+
datasets.SplitGenerator(
|
182 |
+
name=datasets.Split.VALIDATION,
|
183 |
+
gen_kwargs={
|
184 |
+
"filepath": data_dir["validation"],
|
185 |
+
"split": "validation",
|
186 |
+
},
|
187 |
+
),
|
188 |
+
datasets.SplitGenerator(
|
189 |
+
name=datasets.Split.TEST,
|
190 |
+
gen_kwargs={
|
191 |
+
"filepath": data_dir["test"],
|
192 |
+
"split": "test",
|
193 |
+
},
|
194 |
+
),
|
195 |
+
]
|
196 |
+
|
197 |
+
def _generate_examples(self, filepath: Path, split: str) -> Tuple[int, Dict]:
|
198 |
+
"""Yields examples as (key, example) tuples."""
|
199 |
+
|
200 |
+
key_slot_constant = ["distance", "traffic_info", "poi_type", "address", "poi", "room", "agenda", "time", "date", "party", "event", "weather_attribute", "location"]
|
201 |
+
key_kb_constant = ["distance", "traffic_info", "poi_type", "address", "poi", "room", "agenda", "time", "date", "party", "event", "monday", "tuesday", "wednesday", "thursday", "friday", "saturday", "sunday", "today", "location"]
|
202 |
+
|
203 |
+
with open(filepath, "r+") as fw:
|
204 |
+
data = json.loads(fw.read())
|
205 |
+
|
206 |
+
if self.config.schema == "source":
|
207 |
+
for idx, example in enumerate(data):
|
208 |
+
example["index"] = str(idx)
|
209 |
+
for i in range(len(example["dialogue"])):
|
210 |
+
if "requested" not in example["dialogue"][i]["data"]: # the difference between user and system utterance (user and system utterance is divided into each dict in the origin dataset)
|
211 |
+
example["dialogue"][i]["data"]["requested"] = {}
|
212 |
+
example["dialogue"][i]["data"]["slots"] = {}
|
213 |
+
for key in key_slot_constant:
|
214 |
+
example["dialogue"][i]["data"]["requested"][key] = ""
|
215 |
+
example["dialogue"][i]["data"]["slots"][key] = ""
|
216 |
+
else:
|
217 |
+
for key in key_slot_constant:
|
218 |
+
if key not in example["dialogue"][i]["data"]["requested"]:
|
219 |
+
example["dialogue"][i]["data"]["requested"][key] = ""
|
220 |
+
if key not in example["dialogue"][i]["data"]["slots"]:
|
221 |
+
example["dialogue"][i]["data"]["slots"][key] = ""
|
222 |
+
|
223 |
+
if not example["scenario"]["kb"].get("items"):
|
224 |
+
example["scenario"]["kb"]["items"] = []
|
225 |
+
|
226 |
+
for i in range(len(example["scenario"]["kb"]["items"])):
|
227 |
+
for key in key_kb_constant:
|
228 |
+
if key not in example["scenario"]["kb"]["items"][i]:
|
229 |
+
example["scenario"]["kb"]["items"][i][key] = ""
|
230 |
+
|
231 |
+
yield str(idx), example
|
232 |
+
|
233 |
+
elif self.config.schema == "seacrowd_tod":
|
234 |
+
for idx, tod_dialogue in enumerate(data):
|
235 |
+
example = {}
|
236 |
+
example["dialogue_idx"] = idx
|
237 |
+
|
238 |
+
dialogue = []
|
239 |
+
# NOTE: the dialogue always started with `driver` as first utterance
|
240 |
+
for turn, i in enumerate(range(0, len(tod_dialogue["dialogue"]) + 2, 2)):
|
241 |
+
dial = {}
|
242 |
+
dial["turn_idx"] = turn
|
243 |
+
|
244 |
+
# system_utterance properties
|
245 |
+
dial["system_utterance"] = ""
|
246 |
+
dial["system_acts"] = []
|
247 |
+
if turn != 0:
|
248 |
+
dial["system_utterance"] = tod_dialogue["dialogue"][i - 1]["data"]["utterance"]
|
249 |
+
if i < len(tod_dialogue["dialogue"]):
|
250 |
+
# NOTE: system_acts will be filled with every slot that has 'True' value on the origin dataset (on the requested field)
|
251 |
+
for act in tod_dialogue["dialogue"][i + 1]["data"]["requested"]:
|
252 |
+
if tod_dialogue["dialogue"][i + 1]["data"]["requested"][act]:
|
253 |
+
dial["system_acts"].append([act])
|
254 |
+
|
255 |
+
# user_utterance properties
|
256 |
+
dial["turn_label"] = []
|
257 |
+
dial["belief_state"] = []
|
258 |
+
if i == len(tod_dialogue["dialogue"]):
|
259 |
+
# case if turn_idx > len(dialogue) --> add dummy user_utterance
|
260 |
+
dial["user_utterance"] = ""
|
261 |
+
else:
|
262 |
+
dial["user_utterance"] = tod_dialogue["dialogue"][i]["data"]["utterance"]
|
263 |
+
# NOTE: belief_state will be filled with request act from `requested` field & inform act from `slots` field in the origin dataset
|
264 |
+
for act in tod_dialogue["dialogue"][i + 1]["data"]["requested"]:
|
265 |
+
if tod_dialogue["dialogue"][i + 1]["data"]["requested"][act]:
|
266 |
+
dial["belief_state"].append({"slots": [["slot", act]], "act": "request"})
|
267 |
+
for slot, slot_value in tod_dialogue["dialogue"][i + 1]["data"]["slots"].items():
|
268 |
+
dial["belief_state"].append({"slots": [[slot, slot_value]], "act": "inform"})
|
269 |
+
|
270 |
+
# append to dialogue
|
271 |
+
dialogue.append(dial)
|
272 |
+
example["dialogue"] = dialogue
|
273 |
+
yield str(idx), example
|