forked from elena-luo/SODE
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpattern_extraction.py
506 lines (486 loc) · 27.5 KB
/
pattern_extraction.py
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
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
from tqdm import tqdm
import json
import os
import difflib
from prompts import pattern_split_instruction_1, Solution_Explore_split_instruction, Verification_split_instruction
class AnswerSplitOperator():
def __init__(self, partition, part, stat=""):
self.input_partition = partition
self.part = part
self.stat = stat
def part1_split(self, split_result):
result = {}
current_key = None
current_value = []
split_result = split_result.split("\n# Structure")[0]
lines = split_result.split("\n")
for line in lines:
if line.startswith('## '):
if current_key is not None:
result[current_key] = '\n'.join(current_value).strip()
current_value = []
# a new title
current_key = line[3:].strip()
else:
if current_key is not None:
# content under current title
current_value.append(line)
# last one
if current_key is not None:
result[current_key] = '\n'.join(current_value).strip()
return result
def solution_explore_split(self, split_result):
result = {}
current_key = None
current_value = []
split_result = split_result.split("\n# Structure")[0]
split_result = split_result.split("\n# structure")[0]
lines = split_result.split("\n")
for line in lines:
# print(line)
if line.startswith('## '):
if current_key is not None:
if sub_current_key is not None and sub_current_value != []:
sub_result[sub_current_key] = '\n'.join(sub_current_value).strip()
result[current_key] = sub_result
current_key = line[3:].strip()
sub_result = {}
sub_current_key = None
sub_current_value = []
else:
if current_key is not None:
if line.startswith('### '):
if sub_current_key is not None:
sub_result[sub_current_key] = '\n'.join(sub_current_value).strip()
sub_current_value = []
sub_current_key = line[4:].strip()
else:
if sub_current_key is not None:
sub_current_value.append(line)
if current_key is not None:
if sub_current_key is not None and sub_current_value != []:
sub_result[sub_current_key] = '\n'.join(sub_current_value).strip()
result[current_key] = sub_result
return result
def find_str_in_list(self, A, B):
for index, element in enumerate(B):
if A in element:
return index
return -1
def jaccard_similarity(self, A, B):
A = set(A.replace("\n", "").split(" "))
B = set(B.replace("\n", "").split(" "))
intersection = A.intersection(B)
union = A.union(B)
return len(intersection) / len(union)
def split_refine(self, data):
Solution_Explore_save = []
Verify_Explore_save = []
split_dic_save = []
for d in tqdm(data):
try:
# dict_keys(['id', 'prompt', 'task_l1', 'task_l2', 'ground_truth', 'result', 'split_result'])
# get_response(d)
result = d['result']
result_list = result.split(".")
split_result = d["split_result"]
if self.part == "basic_structure":
split_dic = self.part1_split(split_result)
pre_key = None
for k, v in split_dic.items():
# dict_keys(['Question_Repeat', 'Problem_Understand', 'Solution_Explore', 'Verify', 'Conclusion'])
v_list = v.split(".")
if ((v_list[0] in result_list) or ("\n" + v_list[0] in result_list) or (
"\n\n" + v_list[0] in result_list)) and pre_key is not None:
if (v_list[0] in result_list):
start = result_list.index(v_list[0])
elif "\n" + v_list[0] in result_list:
start = result_list.index("\n" + v_list[0])
else:
start = result_list.index("\n\n" + v_list[0])
if start > 0:
v_real = split_dic[pre_key] + ".".join(result_list[: start]) + "."
split_dic[pre_key] = v_real
result_list = result_list[start:]
else:
a1 = self.find_str_in_list(v_list[0], result_list)
a2 = self.find_str_in_list("\n\n" + v_list[0], result_list)
if a2 >= 0 and pre_key is not None:
if a2 > 0 or (a2 == 0 and not result_list[0].startswith("\n\n" + v_list[0])):
v_real = split_dic[pre_key] + ".".join(result_list[: a2]) + result_list[
a2].replace("\n\n" + v_list[0], "")
split_dic[pre_key] = v_real
result_list[a2] = "\n\n" + v_list[0]
result_list = result_list[a2:]
elif a1 >= 0 and pre_key is not None:
if a1 > 0 or (a1 == 0 and not result_list[0].startswith(v_list[0])):
v_real = split_dic[pre_key] + ".".join(result_list[: a1]) + result_list[
a1].replace(v_list[0], "")
split_dic[pre_key] = v_real
result_list[a1] = v_list[0]
result_list = result_list[a1:]
if v_list[-1] == "":
v_list = v_list[:-1]
for i in range(len(v_list)):
flag = -i - 1
if v_list[flag] in result_list:
end = result_list.index(v_list[flag])
v_real = ".".join(result_list[: end + 1]) + "."
split_dic[k] = v_real
result_list = result_list[end + 1:]
break
elif "\n\n" + v_list[flag] in result_list:
end = result_list.index("\n\n" + v_list[flag])
v_real = ".".join(result_list[: end + 1]) + "."
split_dic[k] = v_real
result_list = result_list[end + 1:]
break
elif " " + v_list[flag] in result_list:
end = result_list.index(" " + v_list[flag])
v_real = ".".join(result_list[: end + 1]) + "."
split_dic[k] = v_real
result_list = result_list[end + 1:]
break
else:
# print(flag)
pass
# print("after refine: ")
# print(split_dic)
pre_key = k
if result_list:
if split_dic[pre_key] in result:
v_real = split_dic[pre_key] + ".".join(result_list)
split_dic[pre_key] = v_real
else:
split_dic[pre_key] = ".".join(result_list)
Solution_Explore = split_dic.get("Solution_Explore")
Verify = split_dic.get("Verify")
if Solution_Explore:
if Verify:
Solution_Explore_split_input = "<|im_start|>user\n{Solution_Explore_split_instruction}<|im_end|>\n<|im_start|>assistant\n".format(
Solution_Explore_split_instruction=Solution_Explore_split_instruction.replace(
"{solution}",
Solution_Explore))
Verify_Explore_split_input = "<|im_start|>user\n{Verification_split_instruction}<|im_end|>\n<|im_start|>assistant\n".format(
Verification_split_instruction=Verification_split_instruction.replace("{solution}",
Solution_Explore).replace(
"{reflection}", Verify))
solution = {"id": d['id'], "prompt": Solution_Explore_split_input, "answer": "",
"tag": d["task_l2"]}
verify = {"id": d['id'], "prompt": Verify_Explore_split_input, "answer": "",
"tag": d["task_l2"]}
Solution_Explore_save.append(solution)
Verify_Explore_save.append(verify)
split_dic_save.append(split_dic)
else:
# print(len(Solution_Explore), 0)
pass
else:
# print(0)
pass
elif self.part == "solution_verification_explore":
split_dic = self.solution_explore_split(split_result)
pre_key1, pre_key = None, None
for k1, v1 in split_dic.items():
for k, v in v1.items():
# dict_keys(['Question_Repeat', 'Problem_Understand', 'Solution_Explore', 'Verify', 'Conclusion'])
v_list = v.split(".")
if ((v_list[0] in result_list) or ("\n" + v_list[0] in result_list) or (
"\n\n" + v_list[0] in result_list)) and pre_key is not None:
if (v_list[0] in result_list):
start = result_list.index(v_list[0])
elif "\n" + v_list[0] in result_list:
start = result_list.index("\n" + v_list[0])
else:
start = result_list.index("\n\n" + v_list[0])
if start > 0:
v_real = split_dic[pre_key1][pre_key] + ".".join(result_list[: start]) + "."
split_dic[pre_key1][pre_key] = v_real
result_list = result_list[start:]
else:
a1 = self.find_str_in_list(v_list[0], result_list)
a2 = self.find_str_in_list("\n\n" + v_list[0], result_list)
if a2 >= 0 and pre_key is not None:
if a2 > 0 or (a2 == 0 and not result_list[0].startswith("\n\n" + v_list[0])):
v_real = split_dic[pre_key1][pre_key] + ".".join(result_list[: a2]) + result_list[a2].replace("\n\n" + v_list[0], "")
split_dic[pre_key1][pre_key] = v_real
result_list[a2] = "\n\n" + v_list[0]
result_list = result_list[a2:]
elif a1 >= 0 and pre_key is not None:
if a1 > 0 or (a1 == 0 and not result_list[0].startswith(v_list[0])):
v_real = split_dic[pre_key1][pre_key] + ".".join(result_list[: a1]) + result_list[
a1].replace(v_list[0], "")
split_dic[pre_key1][pre_key] = v_real
result_list[a1] = v_list[0]
result_list = result_list[a1:]
if v_list[-1] == "":
v_list = v_list[:-1]
for i in range(len(v_list)):
flag = -i - 1
if v_list[flag] in result_list:
end = result_list.index(v_list[flag])
v_real = ".".join(result_list[: end + 1]) + "."
split_dic[k1][k] = v_real
result_list = result_list[end + 1:]
break
elif "\n\n" + v_list[flag] in result_list:
end = result_list.index("\n\n" + v_list[flag])
v_real = ".".join(result_list[: end + 1]) + "."
split_dic[k1][k] = v_real
result_list = result_list[end + 1:]
break
elif " " + v_list[flag] in result_list:
end = result_list.index(" " + v_list[flag])
v_real = ".".join(result_list[: end + 1]) + "."
split_dic[k1][k] = v_real
result_list = result_list[end + 1:]
break
else:
# print(flag)
pass
# print("after refine: ")
# print(split_dic)
pre_key1 = k1
pre_key = k
if result_list:
if split_dic[pre_key1][pre_key] in result:
v_real = split_dic[pre_key1][pre_key] + ".".join(result_list)
split_dic[pre_key1][pre_key] = v_real
else:
split_dic[pre_key1][pre_key] = ".".join(result_list)
Solution_Explore_save.append({"id": d['id'], "split_dic": split_dic, "tag": d["task_l2"]})
elif self.part == "verify":
split_dic = self.part1_split(split_result)
pre_key = None
for k, v in split_dic.items():
# dict_keys(['Question_Repeat', 'Problem_Understand', 'Solution_Explore', 'Verify', 'Conclusion'])
v_list = v.split(".")
if ((v_list[0] in result_list) or ("\n" + v_list[0] in result_list) or (
"\n\n" + v_list[0] in result_list)) and pre_key is not None:
if (v_list[0] in result_list):
start = result_list.index(v_list[0])
elif "\n" + v_list[0] in result_list:
start = result_list.index("\n" + v_list[0])
else:
start = result_list.index("\n\n" + v_list[0])
if start > 0:
v_real = split_dic[pre_key] + ".".join(result_list[: start]) + "."
split_dic[pre_key] = v_real
result_list = result_list[start:]
else:
a1 = self.find_str_in_list(v_list[0], result_list)
a2 = self.find_str_in_list("\n\n" + v_list[0], result_list)
if a2 >= 0 and pre_key is not None:
if a2 > 0 or (a2 == 0 and not result_list[0].startswith("\n\n" + v_list[0])):
v_real = split_dic[pre_key] + ".".join(result_list[: a2]) + result_list[
a2].replace("\n\n" + v_list[0], "")
split_dic[pre_key] = v_real
result_list[a2] = "\n\n" + v_list[0]
result_list = result_list[a2:]
elif a1 >= 0 and pre_key is not None:
if a1 > 0 or (a1 == 0 and not result_list[0].startswith(v_list[0])):
v_real = split_dic[pre_key] + ".".join(result_list[: a1]) + result_list[
a1].replace(v_list[0], "")
split_dic[pre_key] = v_real
result_list[a1] = v_list[0]
result_list = result_list[a1:]
if v_list[-1] == "":
v_list = v_list[:-1]
for i in range(len(v_list)):
flag = -i - 1
if v_list[flag] in result_list:
end = result_list.index(v_list[flag])
v_real = ".".join(result_list[: end + 1]) + "."
split_dic[k] = v_real
result_list = result_list[end + 1:]
break
elif "\n\n" + v_list[flag] in result_list:
end = result_list.index("\n\n" + v_list[flag])
v_real = ".".join(result_list[: end + 1]) + "."
split_dic[k] = v_real
result_list = result_list[end + 1:]
break
elif " " + v_list[flag] in result_list:
end = result_list.index(" " + v_list[flag])
v_real = ".".join(result_list[: end + 1]) + "."
split_dic[k] = v_real
result_list = result_list[end + 1:]
break
else:
print(flag)
# print("after refine: ")
# print(split_dic)
pre_key = k
if result_list:
if split_dic[pre_key] in result:
v_real = split_dic[pre_key] + ".".join(result_list)
split_dic[pre_key] = v_real
else:
split_dic[pre_key] = ".".join(result_list)
Verify_Explore_save.append({"id": d['id'], "split_dic": split_dic, "tag": d["task_l2"]})
# print(split_dic)
# start, end = 0, 0
except:
print(f"error occured in processing {d['id']}")
if self.part == "basic_structure":
return Solution_Explore_save, Verify_Explore_save, split_dic_save
elif self.part == "solution_verification_explore":
output = {}
for d in Solution_Explore_save:
output[d["id"]] = d["split_dic"]
return output
elif self.part == "verify":
output = {}
for d in Verify_Explore_save:
output[d["id"]] = d["split_dic"]
return output
def run(self):
if self.part == "basic_structure":
data = []
with open("data/20250209_deepseek_r1_math_26K.jsonl") as f:
for line in f.readlines():
data.append(json.loads(line))
Solution_Explore_save, Verify_Explore_save, split_dic_save = self.split_refine(data)
fw_Solution_Explore = open("R1_solution_explore.jsonl", "w", encoding="utf-8")
fw_Verify_Explore = open("R1_verify_explore.jsonl", "w", encoding="utf-8")
fw_split_part1 = open("R1_split_part1.jsonl", "w", encoding="utf-8")
j, cnt = 0, 0
for d in tqdm(data):
if Solution_Explore_save[j]["id"] == d["id"]:
change = "".join(split_dic_save[j].values())
origin = d["result"]
diff = difflib.SequenceMatcher(None, change, origin).quick_ratio()
jaccard_diff = self.jaccard_similarity(change, origin)
# print("diff: ", diff, "jaccard_diff: ", jaccard_diff)
if diff >= 0.90 or jaccard_diff >= 0.95:
cnt += 1
d["split_result_part1_formate"] = split_dic_save[j]
d["flag"] = True
fw_Solution_Explore.write(json.dumps(Solution_Explore_save[j], ensure_ascii=False) + "\n")
fw_Verify_Explore.write(json.dumps(Verify_Explore_save[j], ensure_ascii=False) + "\n")
fw_split_part1.write(json.dumps(d, ensure_ascii=False) + "\n")
else:
d["flag"] = False
j += 1
else:
d["flag"] = False
print(f"{cnt} data samples are kept and saved from {len(data)} data samples in total")
fw_Solution_Explore.close()
fw_Verify_Explore.close()
fw_split_part1.close()
elif self.part == "solution_verification_explore":
orgin_data = {}
with open("R1_split_part1.jsonl", "r") as f:
for d in f.readlines():
d = json.loads(d)
orgin_data[d["id"]] = d
data = []
with open("R1_solution_explore.jsonl", "r") as f:
for d in f.readlines():
js = json.loads(d)
js['task_l1'] = "math"
js['task_2'] = js['tag']
js['split_result'] = js['result']
js['result'] = js['prompt'].apply(
lambda x: str(x).split("# Mathematical Solution:\n")[1]
.split("\n\nRemember double check original mathematical solution hasn't been rewritten.")[0]
if "# Mathematical Solution:\n" in str(x) else ""
)
data.append(js)
# data = o.execute_select(sql)
Solution_Explore_split = self.split_refine(data)
keys = Solution_Explore_split.keys()
print(f"{len(keys)} data samples passed Solution_Explore_split")
cnt = 0
for k in tqdm(orgin_data.keys()):
d = orgin_data[k]
d["split_result_part2_formate"] = d["split_result_part1_formate"].copy()
if str(k) in keys:
# print("id: ", d["id"],"Reformate match: ", "".join(["".join(Solution_Explore_split[str(k)][kk].values()) for kk in Solution_Explore_split[str(k)].keys()]) == d["split_result_part2_formate"]['Solution_Explore'])
origin = d["split_result_part2_formate"]['Solution_Explore']
change = "".join(["".join(Solution_Explore_split[str(k)][kk].values()) for kk in Solution_Explore_split[str(k)].keys()])
diff = difflib.SequenceMatcher(None, change, origin).quick_ratio()
jaccard_diff = self.jaccard_similarity(change, origin)
# print("diff: ", diff, "jaccard_diff: ", jaccard_diff)
if diff >= 0.90 or jaccard_diff >=0.95:
cnt += 1
orgin_data[k]["flag"] = True
orgin_data[k]["split_result_part2_formate"]["Solution_Explore"] = Solution_Explore_split[str(k)]
else:
orgin_data[k]["flag"] = False
else:
orgin_data[k]["flag"] = False
print(f"{cnt} records remain after validation against the original solution")
data = []
with open("R1_verify_explore.jsonl", "r") as f:
for d in f.readlines():
js = json.loads(d)
js['task_l1'] = "math"
js['task_2'] = js['tag']
js['split_result'] = js['result']
js['result'] = js['prompt'].apply(
lambda x: str(x).split("### 2. Please split the following mathematical reflection text:\n")[1]
.split("\n\n##Output: ")[0]
if "### 2. Please split the following mathematical reflection text:\n" in str(x)
else ""
)
data.append(js)
self.part = "verify"
Verification_split = self.split_refine(data)
keys = Verification_split.keys()
print(f"{len(keys)} data samples passed Verification_split")
cnt = 0
for k in tqdm(orgin_data.keys()):
d = orgin_data[k]
d["split_result_part3_formate"] = d["split_result_part2_formate"].copy()
if str(k) in keys:
# print("id: ", d["id"],"Reformate match: ", "".join(Verification_split[str(k)].values()) == d["split_result_part3_formate"]['Verify'])
origin = d["split_result_part1_formate"]['Verify']
change = "".join(Verification_split[str(k)].values())
diff = difflib.SequenceMatcher(None, change, origin).quick_ratio()
jaccard_diff = self.jaccard_similarity(change, origin)
# print("diff: ", diff, "jaccard_diff: ", jaccard_diff)
d["split_result_part3_formate"]["Verify"] = Verification_split[str(k)]
if diff >= 0.90 or jaccard_diff >=0.95:
cnt += 1
orgin_data[k]["flag"] = True and orgin_data[k]["flag"]
# d["split_result_part3_formate"]["Verify"] = Verification_split[str(k)]
else:
orgin_data[k]["flag"] = False
else:
orgin_data[k]["flag"] = False
print(f"{cnt} records remain after validation against the original solution")
with open("R1_split_part2_filtered.jsonl", "w", encoding="utf-8") as fw_split_part2:
i = 0
for k, v in orgin_data.items():
if v["flag"]:
fw_split_part2.write(json.dumps(v, ensure_ascii=False) + "\n")
i += 1
print("final number of data samples: ", i)
import argparse
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument(
'--part',
type=str,
choices=['basic_structure', 'solution_verification_explore'],
required=True,
help="steps include 'basic_structure' and 'solution_verification_explore'"
)
args = parser.parse_args()
return args
if __name__ == "__main__":
args = parse_args()
partition = "20250209_deepseek_r1_math_26K"
if args.part == 'basic_structure':
# Split the result, reformat it, and upload the solution and verification sections for the content to be split.
part = "basic_structure"
stat = "r1"
op = AnswerSplitOperator(partition, part, stat)
op.run()
elif args.part == 'solution_verification_explore':
# further split the solution and verification part
part = "solution_verification_explore"
stat = "r1"
op = AnswerSplitOperator(partition, part, stat)
op.run()