yuvalkirstain
commited on
Commit
•
5f30172
1
Parent(s):
0592565
add other datasets
Browse files
mrqa.py
CHANGED
@@ -95,6 +95,18 @@ class MRQA(datasets.GeneratorBasedBuilder):
|
|
95 |
VERSION = datasets.Version("1.1.0")
|
96 |
|
97 |
BUILDER_CONFIGS = [
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
98 |
MRQAConfig(
|
99 |
name="newsqa",
|
100 |
data_url={"validation": _URLs["validation+NewsQA"],
|
@@ -113,6 +125,12 @@ class MRQA(datasets.GeneratorBasedBuilder):
|
|
113 |
"train": _URLs["train+HotpotQA"],
|
114 |
"test": _URLs["validation+HotpotQA"]}
|
115 |
),
|
|
|
|
|
|
|
|
|
|
|
|
|
116 |
]
|
117 |
|
118 |
def _info(self):
|
@@ -123,38 +141,9 @@ class MRQA(datasets.GeneratorBasedBuilder):
|
|
123 |
{
|
124 |
"subset": datasets.Value("string"),
|
125 |
"context": datasets.Value("string"),
|
126 |
-
# "context_tokens": datasets.Sequence(
|
127 |
-
# {
|
128 |
-
# "tokens": datasets.Value("string"),
|
129 |
-
# "offsets": datasets.Value("int32"),
|
130 |
-
# }
|
131 |
-
# ),
|
132 |
"qid": datasets.Value("string"),
|
133 |
"idx": datasets.Value("int32"),
|
134 |
"question": datasets.Value("string"),
|
135 |
-
# "question_tokens": datasets.Sequence(
|
136 |
-
# {
|
137 |
-
# "tokens": datasets.Value("string"),
|
138 |
-
# "offsets": datasets.Value("int32"),
|
139 |
-
# }
|
140 |
-
# ),
|
141 |
-
# "detected_answers": datasets.Sequence(
|
142 |
-
# {
|
143 |
-
# "text": datasets.Value("string"),
|
144 |
-
# "char_spans": datasets.Sequence(
|
145 |
-
# {
|
146 |
-
# "start": datasets.Value("int32"),
|
147 |
-
# "end": datasets.Value("int32"),
|
148 |
-
# }
|
149 |
-
# ),
|
150 |
-
# "token_spans": datasets.Sequence(
|
151 |
-
# {
|
152 |
-
# "start": datasets.Value("int32"),
|
153 |
-
# "end": datasets.Value("int32"),
|
154 |
-
# }
|
155 |
-
# ),
|
156 |
-
# }
|
157 |
-
# ),
|
158 |
"answers": datasets.Sequence(datasets.Value("string")),
|
159 |
"answer": datasets.Value("string"),
|
160 |
}
|
@@ -204,35 +193,19 @@ class MRQA(datasets.GeneratorBasedBuilder):
|
|
204 |
idx = 0
|
205 |
for row in f:
|
206 |
paragraph = json.loads(row)
|
207 |
-
context = paragraph["context"]
|
208 |
-
if subset == "HotpotQA":
|
209 |
-
context = context.replace("[PAR] ", "\n\n")
|
210 |
-
context = context.replace("[TLE]", "Title:")
|
211 |
-
context = context.replace("[SEP]", "\nPassage:").strip()
|
212 |
-
# context_tokens = [{"tokens": t[0], "offsets": t[1]} for t in paragraph["context_tokens"]]
|
213 |
for qa in paragraph["qas"]:
|
214 |
qid = qa["qid"]
|
215 |
question = qa["question"].strip()
|
216 |
-
|
217 |
-
|
218 |
-
|
219 |
-
# detected_answers.append(
|
220 |
-
# {
|
221 |
-
# "text": detect_ans["text"].strip(),
|
222 |
-
# "char_spans": [{"start": t[0], "end": t[1]} for t in detect_ans["char_spans"]],
|
223 |
-
# "token_spans": [{"start": t[0], "end": t[1]} for t in detect_ans["token_spans"]],
|
224 |
-
# }
|
225 |
-
# )
|
226 |
-
answers = qa["answers"]
|
227 |
final_row = {
|
228 |
"subset": subset,
|
229 |
-
"context": context,
|
230 |
-
# "context_tokens": context_tokens,
|
231 |
"qid": qid,
|
232 |
"idx": idx,
|
233 |
-
"question": question,
|
234 |
-
# "question_tokens": question_tokens,
|
235 |
-
# "detected_answers": detected_answers,
|
236 |
"answers": answers,
|
237 |
"answer": answers[0]
|
238 |
}
|
@@ -240,7 +213,53 @@ class MRQA(datasets.GeneratorBasedBuilder):
|
|
240 |
yield f"{source}_{qid}", final_row
|
241 |
|
242 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
243 |
if __name__ == '__main__':
|
244 |
from datasets import load_dataset
|
245 |
-
|
|
|
246 |
x = 5
|
|
|
95 |
VERSION = datasets.Version("1.1.0")
|
96 |
|
97 |
BUILDER_CONFIGS = [
|
98 |
+
MRQAConfig(
|
99 |
+
name="searchqa",
|
100 |
+
data_url={"validation": _URLs["validation+SearchQA"],
|
101 |
+
"train": _URLs["train+SearchQA"],
|
102 |
+
"test": _URLs["validation+SearchQA"]}
|
103 |
+
),
|
104 |
+
MRQAConfig(
|
105 |
+
name="squad",
|
106 |
+
data_url={"validation": _URLs["validation+SQuAD"],
|
107 |
+
"train": _URLs["train+SQuAD"],
|
108 |
+
"test": _URLs["validation+SQuAD"]}
|
109 |
+
),
|
110 |
MRQAConfig(
|
111 |
name="newsqa",
|
112 |
data_url={"validation": _URLs["validation+NewsQA"],
|
|
|
125 |
"train": _URLs["train+HotpotQA"],
|
126 |
"test": _URLs["validation+HotpotQA"]}
|
127 |
),
|
128 |
+
MRQAConfig(
|
129 |
+
name="triviaqa",
|
130 |
+
data_url={"validation": _URLs["validation+TriviaQA"],
|
131 |
+
"train": _URLs["train+TriviaQA"],
|
132 |
+
"test": _URLs["validation+TriviaQA"]}
|
133 |
+
),
|
134 |
]
|
135 |
|
136 |
def _info(self):
|
|
|
141 |
{
|
142 |
"subset": datasets.Value("string"),
|
143 |
"context": datasets.Value("string"),
|
|
|
|
|
|
|
|
|
|
|
|
|
144 |
"qid": datasets.Value("string"),
|
145 |
"idx": datasets.Value("int32"),
|
146 |
"question": datasets.Value("string"),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
147 |
"answers": datasets.Sequence(datasets.Value("string")),
|
148 |
"answer": datasets.Value("string"),
|
149 |
}
|
|
|
193 |
idx = 0
|
194 |
for row in f:
|
195 |
paragraph = json.loads(row)
|
196 |
+
context = clean_context(paragraph["context"])
|
|
|
|
|
|
|
|
|
|
|
197 |
for qa in paragraph["qas"]:
|
198 |
qid = qa["qid"]
|
199 |
question = qa["question"].strip()
|
200 |
+
if question[-1] != "?":
|
201 |
+
question += "?"
|
202 |
+
answers = [clean_up_spaces(a) for a in qa["answers"]]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
203 |
final_row = {
|
204 |
"subset": subset,
|
205 |
+
"context": clean_up_spaces(context),
|
|
|
206 |
"qid": qid,
|
207 |
"idx": idx,
|
208 |
+
"question": clean_up_spaces(question),
|
|
|
|
|
209 |
"answers": answers,
|
210 |
"answer": answers[0]
|
211 |
}
|
|
|
213 |
yield f"{source}_{qid}", final_row
|
214 |
|
215 |
|
216 |
+
def clean_context(context):
|
217 |
+
return (
|
218 |
+
context.replace("[PAR] ", "\n\n")
|
219 |
+
.replace("[TLE]", "Title:")
|
220 |
+
.replace("[SEP]", "\nPassage:").strip()
|
221 |
+
.replace("<Li>", "")
|
222 |
+
.replace("</Li>", "")
|
223 |
+
.replace("<OI>", "")
|
224 |
+
.replace("</OI>", "")
|
225 |
+
.replace("<Ol>", "")
|
226 |
+
.replace("</Ol>", "")
|
227 |
+
.replace("<Dd>", "")
|
228 |
+
.replace("</Dd>", "")
|
229 |
+
.replace("<UI>", "")
|
230 |
+
.replace("</UI>", "")
|
231 |
+
.replace("<Ul>", "")
|
232 |
+
.replace("</Ul>", "")
|
233 |
+
.replace("<P>", "")
|
234 |
+
.replace("</P>", "")
|
235 |
+
.replace("[DOC]", "")
|
236 |
+
).strip()
|
237 |
+
|
238 |
+
|
239 |
+
def clean_up_spaces(s):
|
240 |
+
out_string = s
|
241 |
+
return (
|
242 |
+
out_string.replace(" .", ".")
|
243 |
+
.replace(" ?", "?")
|
244 |
+
.replace(" !", "!")
|
245 |
+
.replace(" ,", ",")
|
246 |
+
.replace(" ' ", "'")
|
247 |
+
.replace(" n't", "n't")
|
248 |
+
.replace(" 'm", "'m")
|
249 |
+
.replace(" 's", "'s")
|
250 |
+
.replace(" 've", "'ve")
|
251 |
+
.replace(" 're", "'re")
|
252 |
+
.replace("( ", "(")
|
253 |
+
.replace(" )", ")")
|
254 |
+
.replace(" %", "%")
|
255 |
+
.replace("`` ", "\"")
|
256 |
+
.replace(" ''", "\"")
|
257 |
+
.replace(" :", ":")
|
258 |
+
)
|
259 |
+
|
260 |
+
|
261 |
if __name__ == '__main__':
|
262 |
from datasets import load_dataset
|
263 |
+
|
264 |
+
ssfd_debug = load_dataset("mrqa.py", name="squad")
|
265 |
x = 5
|