AnAutomaticPencil commited on
Commit
a6326c7
1 Parent(s): 9373819

data preprocessing update

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. README.md +11 -345
  2. src/ABSA.py +41 -0
  3. src/CC.py +42 -0
  4. src/CI.py +43 -0
  5. src/DCRG.py +144 -0
  6. src/DS.py +39 -0
  7. src/DST.py +43 -0
  8. src/DT.py +41 -0
  9. src/ER.py +38 -0
  10. src/ID.py +32 -0
  11. src/MCQA.py +41 -0
  12. src/MRC.py +35 -0
  13. src/NLI.py +33 -0
  14. src/QCR.py +40 -0
  15. src/README.md +13 -0
  16. src/RRR.py +36 -0
  17. src/SF.py +36 -0
  18. src/SP.py +38 -0
  19. src/T2S.py +54 -0
  20. src/modules/preprocess/__pycache__/config.cpython-312.pyc +0 -0
  21. src/modules/preprocess/__pycache__/config.cpython-38.pyc +0 -0
  22. src/modules/preprocess/__pycache__/const.cpython-312.pyc +0 -0
  23. src/modules/preprocess/__pycache__/const.cpython-38.pyc +0 -0
  24. src/modules/preprocess/__pycache__/logger.cpython-312.pyc +0 -0
  25. src/modules/preprocess/__pycache__/logger.cpython-38.pyc +0 -0
  26. src/modules/preprocess/config.py +68 -0
  27. src/modules/preprocess/const.py +73 -0
  28. src/modules/preprocess/logger.py +30 -0
  29. src/modules/preprocess/preprocess.py +343 -0
  30. src/modules/preprocess/preprocessor/SerialPreprocessor.py +428 -0
  31. src/modules/preprocess/preprocessor/__pycache__/SerialPreprocessor.cpython-312.pyc +0 -0
  32. src/modules/preprocess/preprocessor/__pycache__/SerialPreprocessor.cpython-38.pyc +0 -0
  33. src/modules/preprocess/preprocessor/__pycache__/base.cpython-312.pyc +0 -0
  34. src/modules/preprocess/preprocessor/__pycache__/base.cpython-38.pyc +0 -0
  35. src/modules/preprocess/preprocessor/__pycache__/knowledge_funcs.cpython-312.pyc +0 -0
  36. src/modules/preprocess/preprocessor/__pycache__/knowledge_funcs.cpython-38.pyc +0 -0
  37. src/modules/preprocess/preprocessor/__pycache__/label_funs.cpython-312.pyc +0 -0
  38. src/modules/preprocess/preprocessor/__pycache__/label_funs.cpython-38.pyc +0 -0
  39. src/modules/preprocess/preprocessor/__pycache__/process_turn_funcs.cpython-38.pyc +0 -0
  40. src/modules/preprocess/preprocessor/__pycache__/prompt_funcs.cpython-312.pyc +0 -0
  41. src/modules/preprocess/preprocessor/__pycache__/prompt_funcs.cpython-38.pyc +0 -0
  42. src/modules/preprocess/preprocessor/base.py +150 -0
  43. src/modules/preprocess/preprocessor/knowledge_funcs.py +505 -0
  44. src/modules/preprocess/preprocessor/label_funs.py +324 -0
  45. src/modules/preprocess/preprocessor/process_turn_funcs.py +22 -0
  46. src/modules/preprocess/preprocessor/prompt_funcs.py +5 -0
  47. src/preprocess.sh +11 -0
  48. src/preprocess/ASTE.py +97 -0
  49. src/preprocess/AlphaNLI.py +84 -0
  50. src/preprocess/Banking77.py +32 -0
README.md CHANGED
@@ -1,347 +1,13 @@
1
- # DialogZoo
2
-
3
- To replicate data construction, three steps are required:
4
- * Download data: ```bash scripts/download.sh```
5
- * Convert origin data into our unified format: ```bash scripts/convert_to_unified.sh```
6
  ```
7
- {
8
- # Optional values: `single` or `multi`. Indicates whether it is a single-turn or multi-turn dialogue.
9
- "turn": str,
10
-
11
- # The domains involved in the dialogue (a list because some dialogues involve multiple domains).
12
- "domain": [],
13
-
14
- # The language of the dialogue, based on the original dataset annotations (e.g., en, fr, etc.).
15
- "locale": str,
16
-
17
- # The dialogue, represented as a list where each element is a dictionary for a single turn.
18
- "dialog": [
19
- {
20
- # The roles involved in each turn. Some datasets may have multiple roles per turn, so it's a list.
21
- # For datasets without role annotations:
22
- # * Use `ROLE` for single-turn data.
23
- # * Use `ROLE1`, `ROLE2`, etc., for multi-turn data.
24
- "roles": [str, ...],
25
-
26
- # The text of the current turn.
27
- "utterance": str,
28
-
29
- # Used for the "answer" in QA tasks.
30
- "start": int,
31
- "end": int,
32
- "dialog_turn": int
33
-
34
- # Rewritten text corresponding to the current turn.
35
- "rewritten": str,
36
-
37
- # Dialogue state, represented as a list where each element includes:
38
- # Domain: Some datasets constrain slot-value pairs within specific domains.
39
- # Intent: Some datasets constrain slot-value pairs within specific intents.
40
- # Slot-value pairs: A list where each element includes a slot and its corresponding values.
41
- # Slot name: A string.
42
- # Values: A list where a slot may have multiple values.
43
- # Each value includes four parts: the value itself, the normalized value,
44
- # the character index in the current turn's text, and more.
45
- # Relation: Some slots are equal to a value, while others are greater than a value.
46
- # Defaults to "equal" if not specified.
47
- # Requested slots: A list of slots that need to be queried but are not filled in the current state.
48
- "belief_state": [
49
- {
50
- # Intent
51
- "intent": str,
52
- # Slot-value pairs
53
- "informed_slot_value_table": [
54
- {
55
- # Slot name
56
- "slot": str,
57
- # Values
58
- "values": [{
59
- # Actual value
60
- "value": str,
61
- # Normalized value
62
- "cononical_value": str
63
- }, ...],
64
- # Slot-value relation
65
- "relation": str,
66
- },
67
- ...
68
- ],
69
- # Requested slots
70
- "requested_slots": [],
71
- # Domain
72
- "domain": str,
73
- }, ...
74
- ],
75
-
76
- # Dialogue actions, represented as a list where each element includes:
77
- # Domain: Some datasets constrain slot-value pairs within specific domains.
78
- # Action: The actions involved in the current turn.
79
- # Slot-value pairs: Same as in dialogue state.
80
- "dialog_acts": [
81
- {
82
- # Action
83
- "act": str,
84
- # Slot-value pairs
85
- "slot_value_table": [
86
- {
87
- # Slot name
88
- "slot": str,
89
- # Slot-value relation
90
- "relation": str,
91
- # Values
92
- "values": [
93
- {
94
- # Actual value
95
- "value": str,
96
- # Normalized value
97
- "cononical_value": str,
98
- # Start position
99
- "start": int,
100
- # End position
101
- "end": int,
102
- },...
103
- ]
104
- },
105
- ...
106
- ],
107
- # Domain
108
- "domain": str,
109
- },
110
- ...
111
- ],
112
-
113
- # Slot filling
114
- "slots_to_fill": {
115
- "intent": str,
116
- "slot_value_table": [
117
- {
118
- "slot": str,
119
- "values": [
120
- {
121
- "value": str,
122
- "start": int,
123
- "end": int
124
- }
125
- ],
126
- "relation": str, # '=', '<=', and so on
127
- }
128
- ]
129
- },
130
-
131
- # Named entity recognition
132
- "named_entity_recognition": [
133
- {
134
- "type": str,
135
- "values": [
136
- {
137
- "value": str,
138
- "start": int,
139
- "end": int
140
- }, ...
141
- ]
142
- }, ...
143
- ],
144
-
145
- "characters": [
146
- {
147
- "value": str,
148
- "start": int,
149
- "end": int
150
- }
151
- ]
152
-
153
- # Intent detection
154
- "active_intents": [str],
155
-
156
- # Query
157
- "query" {
158
- ...
159
- },
160
-
161
- # Query result
162
- "querying_result": {
163
- ...
164
- },
165
-
166
- # Recorded satisfied main items
167
- "main_items": [],
168
-
169
- # Aspect Sentiment Triplet Extraction task, represented as a list where each element includes three parts:
170
- # Target entity.
171
- # Related sentiment.
172
- # Words reflecting the sentiment.
173
- "aspects": [
174
- {
175
- # Target entity
176
- "target": {
177
- # Entity value
178
- "value": str,
179
- # Start position in the current turn's text
180
- "start": int,
181
- # End position in the current turn's text
182
- "end": int
183
- },
184
-
185
- # Category of the target entity
186
- "category": str,
187
-
188
- # Words reflecting the sentiment
189
- "opinion": {
190
- # Sentiment word
191
- "value": str,
192
- # Start position in the current turn's text
193
- "start": int,
194
- # End position in the current turn's text
195
- "end": int
196
- },
197
- # Related sentiment
198
- "sentiment": str
199
- }
200
- ],
201
-
202
- "emotions": [
203
- {
204
- "emotion": str,
205
- "sentiment": "positive", "negative", or "ambiguous",
206
- "evidences": [
207
- {
208
- "turn": int,
209
- "span": str,
210
- "start": int,
211
- "end": int
212
- }
213
- ],
214
- "evidence_types": [str]
215
- }
216
- ],
217
-
218
- "kg_label": str,
219
-
220
- # Knowledge that may be required for each turn, used to select knowledge.
221
- "knowledge_to_select": str,
222
-
223
- # SQL
224
- "sql": str,
225
-
226
- # Rewritten text
227
- "rewritten": str,
228
-
229
- "roles_to_select": [str],
230
- },
231
-
232
- ],
233
-
234
- # Summary derived from the entire dialogue.
235
- "summary": str,
236
-
237
- # Entity relations determined from the entire dialogue.
238
- "instance_relations": [
239
- {
240
- "instance1": str,
241
- "instance2": str,
242
- "relations": [
243
- {
244
- "relation": str,
245
- "trigger": str
246
- }, ...
247
- ]
248
- }, ...
249
- ]
250
-
251
- # Role relations determined from the entire dialogue.
252
- "role_relations": [
253
- {
254
- "turn": int,
255
- "relation": str
256
- }
257
- ],
258
-
259
- # Used in FriendsPersona to determine a character's persona based on the entire dialogue.
260
- "role_personas": [
261
- {
262
- "name": str,
263
- "personas": [
264
- {
265
- "persona": str,
266
- "sentiment": int
267
- }, ...
268
- ]
269
- }
270
- ],
271
-
272
- # External knowledge required for the dialogue.
273
- "knowledge": {
274
- # `text`, `persona`, `kg`, or `schema`.
275
- "type": str,
276
-
277
- # For `text`.
278
- "value": str,
279
-
280
- # For `persona`, persona of all roles, used for personachat.
281
- "value": [
282
- {
283
- # Role name, matching the dialogue turn.
284
- "role": str,
285
-
286
- # Persona description, which may include several sentences.
287
- "description": []
288
- },
289
- ...
290
- ]
291
-
292
- # For `kg`.
293
- "value": {
294
- # `directed` or `undirected`.
295
- "direction": str,
296
-
297
- # Graph.
298
- "graph": [
299
- {
300
- # Source node.
301
- "source": str,
302
-
303
- # Target node.
304
- "target": str,
305
-
306
- # Relation.
307
- "relation": str
308
- },
309
- ...
310
- ]
311
- }
312
-
313
- # For `schema`.
314
- "value": {
315
- ...
316
- }
317
-
318
- # For `dialogue`.
319
- "value": {
320
- "dialog": [],
321
- "relations": []
322
- }
323
-
324
- # For `wiki`.
325
- "value": {
326
- ...
327
- }
328
-
329
- # For `sql`.
330
- "value": [
331
- {
332
- "turn": int,
333
- "sql": str,
334
- "result": ...
335
- }, ...
336
- ],
337
-
338
- # For dialogues based on specific article excerpts, this field indicates the article and section titles.
339
- "value": {
340
- "article title": str,
341
- "section title": str
342
- },
343
- }
344
- }
345
-
346
  ```
347
- * Linearize: ```bash scripts/convert_to_seq.sh```
 
1
+ # Code directory
 
 
 
 
2
  ```
3
+ .
4
+ |-- pretrain: pre-training package
5
+ |-- utils
6
+ | |-- data
7
+ | |-- logger
8
+ | |-- model
9
+ | |-- tokenizer
10
+ | `-- trainer
11
+ |-- 😀[TODO: some other tasks directories]😀
12
+ `-- README.md
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
13
  ```
 
src/ABSA.py ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys
2
+ sys.path.append("modules/preprocess")
3
+
4
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
5
+ from const import (
6
+ ABSA_TERM_CATEGORY_SENTIMENT, ABSA_TERM_OPINION_SENTIMENT, ABSA_CATEGORY_SENTIMENT, ABSA_TERM_SENTIMENT
7
+ )
8
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
9
+ from preprocessor.knowledge_funcs import (
10
+ None_knowledge,
11
+
12
+ )
13
+ from preprocessor.label_funs import (
14
+ extract_aspects_wrapper,
15
+ )
16
+ import sys
17
+
18
+ if __name__ == "__main__":
19
+ TASK = ABSA_TERM_SENTIMENT
20
+ input_data_path = sys.argv[1]
21
+ output_data_path = sys.argv[2]
22
+
23
+ serial_proc = SerialPreprocessor(
24
+ SerialConfig(
25
+ input_data_path,
26
+ output_data_path,
27
+ TASK,
28
+ logger_name=TASK,
29
+ task_bos_token=f"[{TASK}]",
30
+ prompt_func=const_prompt_func_wrapper(
31
+ "Extract all the aspects."
32
+ ),
33
+ # knowledge_func=concat_list_knowledge_wrapper("person 2 persona: ", " | "),
34
+ knowledge_func=None_knowledge,
35
+ label_func=extract_aspects_wrapper(" | ", ", "),
36
+ )
37
+ )
38
+
39
+ serial_proc.launch()
40
+
41
+
src/CC.py ADDED
@@ -0,0 +1,42 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys
2
+ sys.path.append("modules/preprocess")
3
+
4
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
5
+ from const import (
6
+ CHIT_CHAT
7
+ )
8
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
9
+ from preprocessor.knowledge_funcs import (
10
+ concat_list_knowledge_wrapper,
11
+
12
+ )
13
+ from preprocessor.label_funs import (
14
+ extract_turn_utterance,
15
+ )
16
+ import sys
17
+
18
+ if __name__ == "__main__":
19
+ TASK = CHIT_CHAT
20
+ input_data_path = sys.argv[1]
21
+ output_data_path = sys.argv[2]
22
+
23
+ serial_proc = SerialPreprocessor(
24
+ SerialConfig(
25
+ input_data_path,
26
+ output_data_path,
27
+ TASK,
28
+ logger_name=TASK,
29
+ task_bos_token=f"[{TASK}]",
30
+ prompt_func=const_prompt_func_wrapper(
31
+ "Response based on the dialogue context and given self persona"
32
+ ),
33
+ knowledge_func=concat_list_knowledge_wrapper("person 2 persona: ", " | "),
34
+ # knowledge_func=None_knowledge,
35
+ label_func=extract_turn_utterance,
36
+ roles_to_build_example=[["person 2"]],
37
+ )
38
+ )
39
+
40
+ serial_proc.launch()
41
+
42
+
src/CI.py ADDED
@@ -0,0 +1,43 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys
2
+ sys.path.append("modules/preprocess")
3
+
4
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
5
+ from const import (
6
+ CHARACTER_IDENTIFICATION
7
+ )
8
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
9
+ from preprocessor.knowledge_funcs import (
10
+ None_knowledge,
11
+
12
+ )
13
+ from preprocessor.label_funs import (
14
+ extract_characters,
15
+ )
16
+ from preprocessor.process_turn_funcs import introduce_mention_to_utterance_wrapper
17
+ import sys
18
+
19
+ if __name__ == "__main__":
20
+ TASK = CHARACTER_IDENTIFICATION
21
+ input_data_path = sys.argv[1]
22
+ output_data_path = sys.argv[2]
23
+
24
+ serial_proc = SerialPreprocessor(
25
+ SerialConfig(
26
+ input_data_path,
27
+ output_data_path,
28
+ TASK,
29
+ logger_name=TASK,
30
+ task_bos_token=f"[{TASK}]",
31
+ prompt_func=const_prompt_func_wrapper(
32
+ "Predict all characters mentioned in the given dialogue marked by [ mention ] tag based on the context."
33
+ ),
34
+ knowledge_func=None_knowledge,
35
+ # knowledge_func=None_knowledge,
36
+ label_func=extract_characters,
37
+ all_turns_process_func=introduce_mention_to_utterance_wrapper(" [ ", " ] "),
38
+ )
39
+ )
40
+
41
+ serial_proc.launch()
42
+
43
+
src/DCRG.py ADDED
@@ -0,0 +1,144 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys
2
+
3
+ sys.path.append("modules/preprocess")
4
+
5
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
6
+ from const import (
7
+ DIALOGUE_CONTEXT_TO_RESPONSE_GENERATION,
8
+ DOCUMENT_GROUNDED_CONVERSATION,
9
+ MULTI_REF_SEP,
10
+ )
11
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
12
+ from preprocessor.knowledge_funcs import (
13
+ extract_dialogue_knowledge_wrapper,
14
+ origin_knowledge,
15
+ None_knowledge,
16
+ extract_kg_knowledge_wrapper,
17
+ extract_turn_knowledge_wrapper,
18
+ )
19
+ from preprocessor.label_funs import (
20
+ extract_turn_utterance,
21
+ )
22
+ import sys
23
+
24
+ if __name__ == "__main__":
25
+ input_data_path = sys.argv[1]
26
+ output_data_path = sys.argv[2]
27
+ TASK = DOCUMENT_GROUNDED_CONVERSATION
28
+
29
+ if len(sys.argv) <= 3:
30
+ based_on = "dialogue"
31
+ else:
32
+ based_on = sys.argv[3]
33
+
34
+ if len(sys.argv) < 5:
35
+ if based_on == "turn-document":
36
+ serial_proc = SerialPreprocessor(
37
+ SerialConfig(
38
+ input_data_path,
39
+ output_data_path,
40
+ TASK,
41
+ logger_name=TASK,
42
+ task_bos_token=f"[{TASK}]",
43
+ prompt_func=const_prompt_func_wrapper(
44
+ "Response based on the dialogue context and given knowledge"
45
+ ),
46
+ # knowledge_func=extract_kg_knowledge_wrapper(": ", " | ", "; ", " "),
47
+ # knowledge_func=extract_dialogue_knowledge_wrapper(": ", " | ", ", "),
48
+ # knowledge_func=None_knowledge,
49
+ knowledge_func=origin_knowledge,
50
+ turn_knowledge_func=extract_turn_knowledge_wrapper(
51
+ ": ", " | ", ", "
52
+ ),
53
+ label_func=extract_turn_utterance,
54
+ roles_to_build_example=[["user1"], ["user2"]],
55
+ # dev_and_test_roles_to_build_example=[["user2"]],
56
+ roles_in_history=None,
57
+ multi_ref_sep=None,
58
+ )
59
+ )
60
+ elif based_on == "document":
61
+ serial_proc = SerialPreprocessor(
62
+ SerialConfig(
63
+ input_data_path,
64
+ output_data_path,
65
+ TASK,
66
+ logger_name=TASK,
67
+ task_bos_token=f"[{TASK}]",
68
+ prompt_func=const_prompt_func_wrapper(
69
+ "Response based on the dialogue context and given knowledge"
70
+ ),
71
+ # knowledge_func=extract_kg_knowledge_wrapper(": ", " | ", "; ", " "),
72
+ knowledge_func=extract_dialogue_knowledge_wrapper(
73
+ ": ", " | ", ", "
74
+ ),
75
+ # knowledge_func=None_knowledge,
76
+ # knowledge_func=origin_knowledge,
77
+ label_func=extract_turn_utterance,
78
+ roles_to_build_example=[
79
+ ["third-person"],
80
+ ["Listener"],
81
+ ["Speaker"],
82
+ ],
83
+ dev_and_test_roles_to_build_example=[
84
+ ["third-person"],
85
+ ["Listener"],
86
+ ],
87
+ )
88
+ )
89
+ elif based_on == "None":
90
+ serial_proc = SerialPreprocessor(
91
+ SerialConfig(
92
+ input_data_path,
93
+ output_data_path,
94
+ TASK,
95
+ logger_name=TASK,
96
+ task_bos_token=f"[{TASK}]",
97
+ prompt_func=const_prompt_func_wrapper(
98
+ "Response based on the dialogue context and given knowledge"
99
+ ),
100
+ knowledge_func=None_knowledge,
101
+ label_func=extract_turn_utterance,
102
+ roles_to_build_example=[["SYSTEM"]],
103
+ )
104
+ )
105
+ else:
106
+ serial_proc = SerialPreprocessor(
107
+ SerialConfig(
108
+ input_data_path,
109
+ output_data_path,
110
+ TASK,
111
+ logger_name=TASK,
112
+ task_bos_token=f"[{TASK}]",
113
+ prompt_func=const_prompt_func_wrapper(
114
+ "Response based on the dialogue context and given knowledge"
115
+ ),
116
+ knowledge_func=extract_kg_knowledge_wrapper(": ", " | ", "; ", " "),
117
+ # knowledge_func=extract_dialogue_knowledge_wrapper(": ", " | ", ", "),
118
+ # knowledge_func=None_knowledge,
119
+ label_func=extract_turn_utterance,
120
+ roles_to_build_example=[["SYSTEM"], ["USER"]],
121
+ dev_and_test_roles_to_build_example=[["SYSTEM"]],
122
+ )
123
+ )
124
+ else:
125
+ serial_proc = SerialPreprocessor(
126
+ SerialConfig(
127
+ input_data_path,
128
+ output_data_path,
129
+ TASK,
130
+ logger_name=TASK,
131
+ task_bos_token=f"[{TASK}]",
132
+ prompt_func=const_prompt_func_wrapper(
133
+ "Response based on the dialogue context and given knowledge"
134
+ ),
135
+ # knowledge_func=extract_kg_knowledge_wrapper(": ", " | ", "; ", " "),
136
+ knowledge_func=extract_dialogue_knowledge_wrapper(": ", " | ", ", "),
137
+ label_func=extract_turn_utterance,
138
+ roles_to_build_example=[["SYSTEM"]],
139
+ roles_in_history=[["USER"]],
140
+ multi_ref_sep=MULTI_REF_SEP,
141
+ )
142
+ )
143
+
144
+ serial_proc.launch()
src/DS.py ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys
2
+ sys.path.append("modules/preprocess")
3
+
4
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
5
+ from const import (
6
+ DIALOGUE_SUMMARY,
7
+ )
8
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
9
+ from preprocessor.knowledge_funcs import (
10
+ None_knowledge,
11
+ )
12
+ from preprocessor.label_funs import (
13
+ extract_summary,
14
+ )
15
+ import sys
16
+
17
+ if __name__ == "__main__":
18
+ TASK = DIALOGUE_SUMMARY
19
+ input_data_path = sys.argv[1]
20
+ output_data_path = sys.argv[2]
21
+
22
+ serial_proc = SerialPreprocessor(
23
+ SerialConfig(
24
+ input_data_path,
25
+ output_data_path,
26
+ TASK,
27
+ logger_name=TASK,
28
+ task_bos_token=f"[{TASK}]",
29
+ prompt_func=const_prompt_func_wrapper(
30
+ "Summarize the dialogue."
31
+ ),
32
+ knowledge_func=None_knowledge,
33
+ label_func=extract_summary,
34
+ )
35
+ )
36
+
37
+ serial_proc.launch()
38
+
39
+
src/DST.py ADDED
@@ -0,0 +1,43 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys
2
+
3
+ sys.path.append("modules/preprocess")
4
+
5
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
6
+ from const import (
7
+ DIALOGUE_STATE_TRACKING,
8
+ )
9
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
10
+ from preprocessor.knowledge_funcs import None_knowledge, extract_turn_domains_wrapper
11
+ from preprocessor.label_funs import (
12
+ extract_belief_state_wrapper,
13
+ )
14
+ import os, shutil
15
+
16
+ if __name__ == "__main__":
17
+ TASK = DIALOGUE_STATE_TRACKING
18
+ input_data_path = sys.argv[1]
19
+ output_data_path = sys.argv[2]
20
+ serial_proc = SerialPreprocessor(
21
+ SerialConfig(
22
+ input_data_path,
23
+ output_data_path,
24
+ TASK,
25
+ logger_name=TASK,
26
+ task_bos_token=f"[{TASK}]",
27
+ prompt_func=const_prompt_func_wrapper(
28
+ "Generate the dialogue state based on the given dialogue context."
29
+ ),
30
+ knowledge_func=None_knowledge,
31
+ label_func=extract_belief_state_wrapper(", ", " | ", "; ", ": "),
32
+ roles_to_build_example=[["USER"]],
33
+ )
34
+ )
35
+
36
+ serial_proc.launch()
37
+
38
+ for split in ["train", "dev", "test"]:
39
+ if os.path.isfile(os.path.join(input_data_path, f"{split}_ontology.json")):
40
+ shutil.copyfile(
41
+ os.path.join(input_data_path, f"{split}_ontology.json"),
42
+ os.path.join(output_data_path, f"{split}_ontology.json"),
43
+ )
src/DT.py ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys
2
+
3
+ sys.path.append("modules/preprocess")
4
+
5
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
6
+ from const import (
7
+ DATA_TO_TEXT, MULTI_REF_SEP
8
+ )
9
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
10
+ from preprocessor.knowledge_funcs import (
11
+ extract_dict_knowledge_wrapper,
12
+ )
13
+ from preprocessor.label_funs import (
14
+ extract_turn_utterance,
15
+ )
16
+ import sys
17
+
18
+ if __name__ == "__main__":
19
+ # 2. Emotion Recognition
20
+ TASK = DATA_TO_TEXT
21
+ input_data_path = sys.argv[1]
22
+ output_data_path = sys.argv[2]
23
+
24
+ serial_proc = SerialPreprocessor(
25
+ SerialConfig(
26
+ input_data_path,
27
+ output_data_path,
28
+ TASK,
29
+ logger_name=TASK,
30
+ task_bos_token=f"[{TASK}]",
31
+ prompt_func=const_prompt_func_wrapper(
32
+ "Generate the corresponding text based on the given knowledge."
33
+ ),
34
+ knowledge_func=extract_dict_knowledge_wrapper(": ", " | "),
35
+ label_func=extract_turn_utterance,
36
+ roles_in_history=[],
37
+ multi_ref_sep=MULTI_REF_SEP
38
+ )
39
+ )
40
+
41
+ serial_proc.launch()
src/ER.py ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys
2
+ sys.path.append("modules/preprocess")
3
+
4
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
5
+ from const import (
6
+ EMOTION_RECOGNITION,
7
+ )
8
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
9
+ from preprocessor.knowledge_funcs import (
10
+ None_knowledge,
11
+ )
12
+ from preprocessor.label_funs import (
13
+ extract_turn_emotion_wrapper,
14
+ )
15
+ import sys
16
+
17
+ if __name__ == "__main__":
18
+ # 2. Emotion Recognition
19
+ TASK = EMOTION_RECOGNITION
20
+ input_data_path = sys.argv[1]
21
+ output_data_path = sys.argv[2]
22
+
23
+ serial_proc = SerialPreprocessor(
24
+ SerialConfig(
25
+ input_data_path,
26
+ output_data_path,
27
+ TASK,
28
+ logger_name=TASK,
29
+ task_bos_token=f"[{TASK}]",
30
+ prompt_func=const_prompt_func_wrapper(
31
+ "Recognize correct emotions based on the dialogue context."
32
+ ),
33
+ knowledge_func=None_knowledge,
34
+ label_func=extract_turn_emotion_wrapper(", "),
35
+ )
36
+ )
37
+
38
+ serial_proc.launch()
src/ID.py ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys
2
+
3
+ sys.path.append("modules/preprocess")
4
+
5
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
6
+ from const import INTENT_DETECTION, MULTI_REF_SEP
7
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
8
+ from preprocessor.knowledge_funcs import None_knowledge
9
+ from preprocessor.label_funs import (
10
+ extract_intents_wrapper,
11
+ )
12
+
13
+ if __name__ == "__main__":
14
+ TASK = INTENT_DETECTION
15
+ input_data_path = sys.argv[1]
16
+ output_data_path = sys.argv[2]
17
+ serial_proc = SerialPreprocessor(
18
+ SerialConfig(
19
+ input_data_path,
20
+ output_data_path,
21
+ TASK,
22
+ logger_name=TASK,
23
+ task_bos_token=f"[{TASK}]",
24
+ prompt_func=const_prompt_func_wrapper(
25
+ "Detect the intent based on the given dialogue context."
26
+ ),
27
+ knowledge_func=None_knowledge,
28
+ label_func=extract_intents_wrapper(" | "),
29
+ )
30
+ )
31
+
32
+ serial_proc.launch()
src/MCQA.py ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys
2
+
3
+ sys.path.append("modules/preprocess")
4
+
5
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
6
+ from const import (
7
+ MULTIPLE_CHOICE_QUESTION_ANSWERING,
8
+ )
9
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
10
+ from preprocessor.knowledge_funcs import (
11
+ # extract_dict_knowledge_wrapper
12
+ extract_dialogue_knowledge_wrapper,
13
+ # None_knowledge,
14
+ )
15
+ from preprocessor.label_funs import (
16
+ extract_options,
17
+ )
18
+
19
+ if __name__ == "__main__":
20
+ TASK = MULTIPLE_CHOICE_QUESTION_ANSWERING
21
+ input_data_path = sys.argv[1]
22
+ output_data_path = sys.argv[2]
23
+ sort_or_not = len(sys.argv) > 3
24
+
25
+ prompt = "Generate the best choice."
26
+
27
+ serial_proc = SerialPreprocessor(
28
+ SerialConfig(
29
+ input_data_path,
30
+ output_data_path,
31
+ TASK,
32
+ logger_name=TASK,
33
+ task_bos_token=f"[{TASK}]",
34
+ prompt_func=const_prompt_func_wrapper(prompt),
35
+ knowledge_func=extract_dialogue_knowledge_wrapper(": ", " | ", ", "),
36
+ # knowledge_func=extract_dict_knowledge_wrapper(": ", " | "),
37
+ label_func=extract_options,
38
+ )
39
+ )
40
+
41
+ serial_proc.launch()
src/MRC.py ADDED
@@ -0,0 +1,35 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys
2
+
3
+ sys.path.append("modules/preprocess")
4
+
5
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
6
+ from const import MACHINE_READING_COMPREHENSION, MULTI_REF_SEP
7
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
8
+ from preprocessor.knowledge_funcs import extract_dialogue_knowledge_wrapper
9
+ from preprocessor.label_funs import (
10
+ extract_turn_utterance,
11
+ )
12
+
13
+ if __name__ == "__main__":
14
+ TASK = MACHINE_READING_COMPREHENSION
15
+ input_data_path = sys.argv[1]
16
+ output_data_path = sys.argv[2]
17
+ serial_proc = SerialPreprocessor(
18
+ SerialConfig(
19
+ input_data_path,
20
+ output_data_path,
21
+ TASK,
22
+ logger_name=TASK,
23
+ task_bos_token=f"[{TASK}]",
24
+ prompt_func=const_prompt_func_wrapper(
25
+ "Answer the question based on the given document and dialogue context."
26
+ ),
27
+ knowledge_func=extract_dialogue_knowledge_wrapper(": ", " | ", ", "),
28
+ label_func=extract_turn_utterance,
29
+ roles_in_history=[["USER"]],
30
+ roles_to_build_example=[["SYSTEM"]],
31
+ multi_ref_sep=MULTI_REF_SEP,
32
+ )
33
+ )
34
+
35
+ serial_proc.launch()
src/NLI.py ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys
2
+
3
+ sys.path.append("modules/preprocess")
4
+
5
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
6
+ from const import (
7
+ NATURAL_LANGUAGE_INFERENCE,
8
+ )
9
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
10
+ from preprocessor.knowledge_funcs import extract_dict_knowledge_wrapper
11
+ from preprocessor.label_funs import (
12
+ extract_options,
13
+ )
14
+
15
+ if __name__ == "__main__":
16
+ TASK = NATURAL_LANGUAGE_INFERENCE
17
+ input_data_path = sys.argv[1]
18
+ output_data_path = sys.argv[2]
19
+ serial_proc = SerialPreprocessor(
20
+ SerialConfig(
21
+ input_data_path,
22
+ output_data_path,
23
+ TASK,
24
+ logger_name=TASK,
25
+ task_bos_token=f"[{TASK}]",
26
+ prompt_func=const_prompt_func_wrapper("Generate the better hypothesis."),
27
+ knowledge_func=extract_dict_knowledge_wrapper(": ", " | "),
28
+ # knowledge_func=None_knowledge,
29
+ label_func=extract_options,
30
+ )
31
+ )
32
+
33
+ serial_proc.launch()
src/QCR.py ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys
2
+
3
+ sys.path.append("modules/preprocess")
4
+
5
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
6
+ from const import (
7
+ QUESTION_IN_CONTEXT_REWRITING,
8
+ )
9
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
10
+ from preprocessor.knowledge_funcs import extract_dict_knowledge_wrapper, None_knowledge
11
+ from preprocessor.label_funs import (
12
+ extrac_rewritten,
13
+ )
14
+
15
+ if __name__ == "__main__":
16
+ TASK = QUESTION_IN_CONTEXT_REWRITING
17
+ input_data_path = sys.argv[1]
18
+ output_data_path = sys.argv[2]
19
+ has_knowledge = len(sys.argv) > 3
20
+
21
+ kf = (
22
+ extract_dict_knowledge_wrapper(": ", " | ") if has_knowledge else None_knowledge
23
+ )
24
+ serial_proc = SerialPreprocessor(
25
+ SerialConfig(
26
+ input_data_path,
27
+ output_data_path,
28
+ TASK,
29
+ logger_name=TASK,
30
+ task_bos_token=f"[{TASK}]",
31
+ prompt_func=const_prompt_func_wrapper(
32
+ "Rewrite the user utterance of current turn based on the given dialogue context."
33
+ ),
34
+ knowledge_func=kf,
35
+ # knowledge_func=None_knowledge,
36
+ label_func=extrac_rewritten,
37
+ )
38
+ )
39
+
40
+ serial_proc.launch()
src/README.md ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Code directory
2
+ ```
3
+ .
4
+ |-- pretrain: pre-training package
5
+ |-- utils
6
+ | |-- data
7
+ | |-- logger
8
+ | |-- model
9
+ | |-- tokenizer
10
+ | `-- trainer
11
+ |-- 😀[TODO: some other tasks directories]😀
12
+ `-- README.md
13
+ ```
src/RRR.py ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys
2
+
3
+ sys.path.append("modules/preprocess")
4
+
5
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
6
+ from const import (
7
+ ROLE_RELATION_RECOGNITION,
8
+ )
9
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
10
+ from preprocessor.knowledge_funcs import (
11
+ None_knowledge,
12
+ )
13
+ from preprocessor.label_funs import (
14
+ extract_role_relation_without_turn_wrapper,
15
+ )
16
+
17
+ if __name__ == "__main__":
18
+ TASK = ROLE_RELATION_RECOGNITION
19
+ input_data_path = sys.argv[1]
20
+ output_data_path = sys.argv[2]
21
+ serial_proc = SerialPreprocessor(
22
+ SerialConfig(
23
+ input_data_path,
24
+ output_data_path,
25
+ TASK,
26
+ logger_name=TASK,
27
+ task_bos_token=f"[{TASK}]",
28
+ prompt_func=const_prompt_func_wrapper(
29
+ "Judge the relation of two roles in the given dialogue."
30
+ ),
31
+ knowledge_func=None_knowledge,
32
+ label_func=extract_role_relation_without_turn_wrapper("; "),
33
+ )
34
+ )
35
+
36
+ serial_proc.launch()
src/SF.py ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys
2
+
3
+ sys.path.append("modules/preprocess")
4
+
5
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
6
+ from const import (
7
+ SLOT_FILLING,
8
+ )
9
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
10
+ from preprocessor.knowledge_funcs import (
11
+ None_knowledge,
12
+ )
13
+ from preprocessor.label_funs import (
14
+ extract_slots_without_intents_wrapper,
15
+ )
16
+
17
+ if __name__ == "__main__":
18
+ TASK = SLOT_FILLING
19
+ input_data_path = sys.argv[1]
20
+ output_data_path = sys.argv[2]
21
+ serial_proc = SerialPreprocessor(
22
+ SerialConfig(
23
+ input_data_path,
24
+ output_data_path,
25
+ TASK,
26
+ logger_name=TASK,
27
+ task_bos_token=f"[{TASK}]",
28
+ prompt_func=const_prompt_func_wrapper(
29
+ "Fill all the slots based on the given utterance."
30
+ ),
31
+ knowledge_func=None_knowledge,
32
+ label_func=extract_slots_without_intents_wrapper(", ", "; "),
33
+ )
34
+ )
35
+
36
+ serial_proc.launch()
src/SP.py ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys
2
+
3
+ sys.path.append("modules/preprocess")
4
+
5
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
6
+ from const import (
7
+ DIALOGUE_SUMMARY,
8
+ )
9
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
10
+ from preprocessor.knowledge_funcs import (
11
+ None_knowledge,
12
+ )
13
+ from preprocessor.label_funs import (
14
+ extract_sql,
15
+ )
16
+ import sys
17
+
18
+ if __name__ == "__main__":
19
+ TASK = "Semantic Parsing"
20
+ input_data_path = sys.argv[1]
21
+ output_data_path = sys.argv[2]
22
+
23
+ serial_proc = SerialPreprocessor(
24
+ SerialConfig(
25
+ input_data_path,
26
+ output_data_path,
27
+ TASK,
28
+ logger_name=TASK,
29
+ task_bos_token=f"[{TASK}]",
30
+ prompt_func=const_prompt_func_wrapper(
31
+ "Parse the sentence into intents and slots."
32
+ ),
33
+ knowledge_func=None_knowledge,
34
+ label_func=extract_sql,
35
+ )
36
+ )
37
+
38
+ serial_proc.launch()
src/T2S.py ADDED
@@ -0,0 +1,54 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import sys
2
+ import os
3
+
4
+ sys.path.append("modules/preprocess")
5
+
6
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
7
+ from const import (
8
+ TEXT2SQL,
9
+ )
10
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
11
+ from preprocessor.knowledge_funcs import (
12
+ origin_knowledge,
13
+ extract_schema_knowledge_wrapper,
14
+ )
15
+ from preprocessor.label_funs import (
16
+ extract_sql,
17
+ )
18
+
19
+ import shutil
20
+
21
+ if __name__ == "__main__":
22
+ # 8. Text2SQL
23
+ TASK = TEXT2SQL
24
+ input_data_path = sys.argv[1]
25
+ output_data_path = sys.argv[2]
26
+
27
+ serial_proc = SerialPreprocessor(
28
+ SerialConfig(
29
+ input_data_path,
30
+ output_data_path,
31
+ TASK,
32
+ logger_name=TASK,
33
+ task_bos_token=f"[{TASK}]",
34
+ prompt_func=const_prompt_func_wrapper(
35
+ "Parse the SQL based on the given dialogue context and schema."
36
+ ),
37
+ knowledge_func=origin_knowledge,
38
+ turn_knowledge_func=extract_schema_knowledge_wrapper(),
39
+ label_func=extract_sql,
40
+ )
41
+ )
42
+
43
+ serial_proc.launch()
44
+
45
+ shutil.copyfile(
46
+ os.path.join(input_data_path, "tables.json"),
47
+ os.path.join(output_data_path, "tables.json"),
48
+ )
49
+
50
+ if not os.path.exists(os.path.join(output_data_path, "database")):
51
+ shutil.copytree(
52
+ os.path.join(input_data_path, "database"),
53
+ os.path.join(output_data_path, "database"),
54
+ )
src/modules/preprocess/__pycache__/config.cpython-312.pyc ADDED
Binary file (3.54 kB). View file
 
src/modules/preprocess/__pycache__/config.cpython-38.pyc ADDED
Binary file (2.6 kB). View file
 
src/modules/preprocess/__pycache__/const.cpython-312.pyc ADDED
Binary file (2.59 kB). View file
 
src/modules/preprocess/__pycache__/const.cpython-38.pyc ADDED
Binary file (2.37 kB). View file
 
src/modules/preprocess/__pycache__/logger.cpython-312.pyc ADDED
Binary file (1.21 kB). View file
 
src/modules/preprocess/__pycache__/logger.cpython-38.pyc ADDED
Binary file (764 Bytes). View file
 
src/modules/preprocess/config.py ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Base config supporting save and load. Refer to https://github.com/huggingface/transformers/blob/main/src/transformers/configuration_utils.py.
3
+ Author: md
4
+ """
5
+
6
+ from typing import Dict, Any
7
+ import copy
8
+ import json
9
+
10
+
11
+ class BaseConfig(object):
12
+ def __init__(
13
+ self,
14
+ logger_name: str = None,
15
+ log_file: str = None,
16
+ log_mode: str = "a",
17
+ formatter: str = "%(asctime)s | %(levelname)s | %(message)s",
18
+ ) -> None:
19
+ """
20
+ params:
21
+ ------
22
+
23
+ logger_name: logger name
24
+ log_file: the file to ouput log. If `None`, output to stdout
25
+ log_mode: mode to write to the log file, `a` is appending.
26
+ formatter: logging formatter.
27
+ """
28
+ self.logger_name = logger_name
29
+ self.log_file = log_file
30
+ self.log_mode = log_mode
31
+ self.formatter = formatter
32
+
33
+ def to_json_string(self) -> Dict[str, Any]:
34
+ output = self.to_dict()
35
+ return json.dumps(output)
36
+
37
+ def to_dict(self) -> Dict[str, Any]:
38
+ """
39
+ Serializes this instance to a Python dictionary.
40
+
41
+ Returns:
42
+ `Dict[str, Any]`: Dictionary of all the attributes that make up this configuration instance.
43
+ """
44
+ output = copy.deepcopy(self.__dict__)
45
+ # if "_auto_class" in output:
46
+ # del output["_auto_class"]
47
+
48
+ return output
49
+
50
+ def from_dict(self, config_dict: Dict[str, Any]) -> None:
51
+ self.__dict__.update(config_dict)
52
+
53
+ def save(self, save_path: str, indent: int = 4) -> None:
54
+ with open(save_path, "w") as writer:
55
+ json.dump(self.to_dict(), writer, indent=indent)
56
+
57
+ def load(self, load_path: str) -> None:
58
+ with open(load_path, "r") as reader:
59
+ self.__dict__.update(json.load(reader))
60
+
61
+
62
+ if __name__ == "__main__":
63
+ config = BaseConfig()
64
+
65
+ config.from_dict({"a": 1, "b": 2, "c": "test", "d": True, "e": 3.2})
66
+ config.save("../../test/test1.json")
67
+ config.load("../../test/test1.json")
68
+ config.save("../../test/test2.json")
src/modules/preprocess/const.py ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Some constant variables.
3
+ Author: md
4
+ """
5
+
6
+ # The split names
7
+ TRAIN_SPLIT = "train"
8
+ DEV_SPLIT = "dev"
9
+ TEST_SPLIT = "test"
10
+
11
+ # Universal dialogue format keywords
12
+ DIALOG = "dialog"
13
+ ROLES = "roles"
14
+ TARGET = "target"
15
+ SUMMARY = "summary"
16
+ KNOWLEDGE = "knowledge"
17
+ UTTERANCE = "utterance"
18
+ EMOTIONS = "emotions"
19
+ EMOTION = "emotion"
20
+ VALUE = "value"
21
+ ASPECTS = "aspects"
22
+ CATEGORY = "category"
23
+ OPINION = "opinion"
24
+ SENTIMENT = "sentiment"
25
+ CHARACTERS = "characters"
26
+ START = "start"
27
+ END = "end"
28
+ BELIEF_STATE = "belief_state"
29
+ DOMAIN = "domain"
30
+ INFORMED_SLOT_VALUE_TABLE = "informed_slot_value_table"
31
+ SLOT = "slot"
32
+ VALUES = "values"
33
+ RELATION = "relation"
34
+ KNOWLEDGE_TO_SELECT = "knowledge_to_select"
35
+ SQL = "sql"
36
+ SLOT_VALUE_TABLE = "slot_value_table"
37
+ SLOTS_TO_FILL = "slots_to_fill"
38
+ ROLE_RELATIONS = "role_relations"
39
+ REWRITTEN = "rewritten"
40
+ ROLES_TO_SELECT = "roles_to_select"
41
+ ACTIVE_INTENTS = "active_intents"
42
+
43
+
44
+ # TASK NAMES
45
+ DIALOGUE_SUMMARY = "Dialogue Summary"
46
+ EMOTION_RECOGNITION = "Emotion Recognition"
47
+ DIALOGUE_CONTEXT_TO_RESPONSE_GENERATION = "Dialogue Context-to-Response Generation"
48
+ ABSA = "ABSA"
49
+ ABSA_TERM_OPINION_SENTIMENT = "ABSA: term opinion sentiment"
50
+ ABSA_TERM_CATEGORY_SENTIMENT = "ABSA: term category sentiment"
51
+ ABSA_TERM_SENTIMENT = "ABSA: term sentiment"
52
+ ABSA_CATEGORY_SENTIMENT = "ABSA: category sentiment"
53
+ CHARACTER_IDENTIFICATION = "Character Identification"
54
+ DIALOGUE_STATE_TRACKING = "Dialogue State Tracking"
55
+ DOCUMENT_GROUNDED_CONVERSATION = "Document Grounded Conversation"
56
+ TEXT2SQL = "Text2SQL"
57
+ SLOT_FILLING = "Slot Filling"
58
+ ROLE_RELATION_RECOGNITION = "Role Relation Recognition"
59
+ QUESTION_IN_CONTEXT_REWRITING = "Question in Context Rewriting"
60
+ NATURAL_LANGUAGE_INFERENCE = "Natural Language Inference"
61
+ MACHINE_READING_COMPREHENSION = "Machine Reading Comprehension"
62
+ MULTIPLE_CHOICE_QUESTION_ANSWERING = "Multiple Choice Question Answering"
63
+ INTENT_DETECTION = "Intent Detection"
64
+ DATA_TO_TEXT = "Data-to-Text"
65
+ CHIT_CHAT = "Chit-Chat"
66
+
67
+ # Seq2Seq
68
+ MULTI_REF_SEP = "__multi_ref_sep__"
69
+ OPTION_LABEL = "option_label"
70
+ CANDIDATES = "candidates"
71
+
72
+ # MENTION
73
+ MENTION = "mention"
src/modules/preprocess/logger.py ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Author: md
3
+ """
4
+ import logging
5
+ import sys
6
+
7
+
8
+ def build_logger(
9
+ logger_name: str,
10
+ level: int,
11
+ log_file: str = None,
12
+ log_mode: str = "a",
13
+ formatter: str = "%(asctime)s | [%(name)s] | %(levelname)s | %(message)s",
14
+ ):
15
+ logger = logging.getLogger(logger_name)
16
+
17
+ logger.setLevel(level)
18
+
19
+ logger.handlers.clear()
20
+
21
+ formatter = logging.Formatter(formatter)
22
+ if log_file is not None:
23
+ handler = logging.FileHandler(log_file, log_mode)
24
+ else:
25
+ handler = logging.StreamHandler(sys.stdout)
26
+
27
+ handler.setFormatter(formatter)
28
+ logger.addHandler(handler)
29
+
30
+ return logger
src/modules/preprocess/preprocess.py ADDED
@@ -0,0 +1,343 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from preprocessor.SerialPreprocessor import SerialConfig, SerialPreprocessor
2
+ from const import (
3
+ DIALOGUE_SUMMARY,
4
+ EMOTION_RECOGNITION,
5
+ DIALOGUE_CONTEXT_TO_TEXT_GENERATION,
6
+ ABSA_TERM_OPINION_SENTIMENT,
7
+ ABSA_TERM_SENTIMENT,
8
+ ABSA_CATEGORY_SENTIMENT,
9
+ ABSA_TERM_CATEGORY_SENTIMENT,
10
+ CHARACTER_IDENTIFICATION,
11
+ DIALOGUE_STATE_TRACKING,
12
+ DOCUMENT_GROUNDED_CONVERSATION,
13
+ TEXT2SQL,
14
+ SLOT_FILLING,
15
+ )
16
+ from preprocessor.prompt_funcs import const_prompt_func_wrapper
17
+ from preprocessor.knowledge_funcs import (
18
+ None_knowledge,
19
+ concat_list_knowledge_wrapper,
20
+ extract_turn_knowledge_wrapper,
21
+ origin_knowledge,
22
+ extract_schema_knowledge_wrapper,
23
+ )
24
+ from preprocessor.label_funs import (
25
+ extract_summary,
26
+ extract_turn_emotion_wrapper,
27
+ extract_turn_utterance,
28
+ extract_aspects_wrapper,
29
+ rebuild_utterance_with_characters,
30
+ extract_belief_state_wrapper,
31
+ extract_sql,
32
+ extract_slots_without_intents_wrapper,
33
+ )
34
+ import os
35
+
36
+ if __name__ == "__main__":
37
+ # 1. Dialogue Summary
38
+ TASK = DIALOGUE_SUMMARY
39
+ input_path = r"E:\research\processed\DialogueSummary"
40
+ output_path = r"E:\research\seq\DialogueSummary"
41
+
42
+ for dataset in os.listdir(input_path):
43
+ input_data_path = os.path.join(input_path, dataset)
44
+ output_data_path = os.path.join(output_path, dataset)
45
+
46
+ serial_proc = SerialPreprocessor(
47
+ SerialConfig(
48
+ input_data_path,
49
+ output_data_path,
50
+ TASK,
51
+ logger_name=TASK,
52
+ task_bos_token=f"[{TASK}]",
53
+ prompt_func=const_prompt_func_wrapper(
54
+ "Give a summary of this dialogue."
55
+ ),
56
+ knowledge_func=None_knowledge,
57
+ label_func=extract_summary,
58
+ )
59
+ )
60
+
61
+ serial_proc.launch()
62
+
63
+ # 2. Emotion Recognition
64
+ TASK = EMOTION_RECOGNITION
65
+ input_path = r"E:\research\processed\EmotionRecognition"
66
+ output_path = r"E:\research\seq\EmotionRecognition"
67
+
68
+ for dataset in os.listdir(input_path):
69
+ input_data_path = os.path.join(input_path, dataset)
70
+ output_data_path = os.path.join(output_path, dataset)
71
+
72
+ serial_proc = SerialPreprocessor(
73
+ SerialConfig(
74
+ input_data_path,
75
+ output_data_path,
76
+ TASK,
77
+ logger_name=TASK,
78
+ task_bos_token=f"[{TASK}]",
79
+ prompt_func=const_prompt_func_wrapper(
80
+ "With given possible emotions, select the correct answer."
81
+ ),
82
+ knowledge_func=concat_list_knowledge_wrapper(
83
+ "possible choices: ", " | "
84
+ ),
85
+ label_func=extract_turn_emotion_wrapper(", "),
86
+ )
87
+ )
88
+
89
+ serial_proc.launch()
90
+
91
+ # 3. Dialogue Context-to-Text Generation
92
+ TASK = DIALOGUE_CONTEXT_TO_TEXT_GENERATION
93
+ input_path = r"E:\research\processed\Dialogue-Context-to-Text Generation"
94
+ output_path = r"E:\research\seq\Dialogue-Context-to-Text Generation"
95
+
96
+ for dataset in os.listdir(input_path):
97
+ input_data_path = os.path.join(input_path, dataset)
98
+ output_data_path = os.path.join(output_path, dataset)
99
+
100
+ serial_proc = SerialPreprocessor(
101
+ SerialConfig(
102
+ input_data_path,
103
+ output_data_path,
104
+ TASK,
105
+ logger_name=TASK,
106
+ task_bos_token=f"[{TASK}]",
107
+ prompt_func=const_prompt_func_wrapper(
108
+ "With given dialogue context, give the response."
109
+ ),
110
+ knowledge_func=None_knowledge,
111
+ label_func=extract_turn_utterance,
112
+ roles_to_build_example=[["Listener"], ["third-person"]],
113
+ )
114
+ )
115
+
116
+ serial_proc.launch()
117
+
118
+ # 4. Aspect Sentiment Analysis
119
+ # 4.1 ABSA: term opinion sentiment
120
+ TASK = ABSA_TERM_OPINION_SENTIMENT
121
+ input_path = r"E:\research\processed\ABSA-term opinion sentiment\ASTE"
122
+ output_path = r"E:\research\seq\Aspect-based Sentiment Analysis\ASTE"
123
+
124
+ for dataset in os.listdir(input_path):
125
+ input_data_path = os.path.join(input_path, dataset)
126
+ output_data_path = os.path.join(output_path, dataset)
127
+
128
+ serial_proc = SerialPreprocessor(
129
+ SerialConfig(
130
+ input_data_path,
131
+ output_data_path,
132
+ TASK,
133
+ logger_name=TASK,
134
+ task_bos_token=f"[{TASK}]",
135
+ prompt_func=const_prompt_func_wrapper("Give all the aspects."),
136
+ knowledge_func=None_knowledge,
137
+ label_func=extract_aspects_wrapper(" | ", ", "),
138
+ )
139
+ )
140
+
141
+ serial_proc.launch()
142
+
143
+ # 4.2 ABSA: term sentiment
144
+ TASK = ABSA_TERM_SENTIMENT
145
+ input_path = r"E:\research\processed\ABSA-term sentiment"
146
+ output_path = r"E:\research\seq\Aspect-based Sentiment Analysis"
147
+
148
+ for dataset in os.listdir(input_path):
149
+ input_data_path = os.path.join(input_path, dataset)
150
+ output_data_path = os.path.join(output_path, dataset)
151
+
152
+ serial_proc = SerialPreprocessor(
153
+ SerialConfig(
154
+ input_data_path,
155
+ output_data_path,
156
+ TASK,
157
+ logger_name=TASK,
158
+ task_bos_token=f"[{TASK}]",
159
+ prompt_func=const_prompt_func_wrapper("Give all the aspects."),
160
+ knowledge_func=None_knowledge,
161
+ label_func=extract_aspects_wrapper(" | ", ", "),
162
+ )
163
+ )
164
+
165
+ serial_proc.launch()
166
+
167
+ # 4.3 ABSA: category sentiment
168
+ TASK = ABSA_CATEGORY_SENTIMENT
169
+ input_path = r"E:\research\processed\ABSA-category sentiment"
170
+ output_path = r"E:\research\seq\Aspect-based Sentiment Analysis"
171
+
172
+ for dataset in os.listdir(input_path):
173
+ input_data_path = os.path.join(input_path, dataset)
174
+ output_data_path = os.path.join(output_path, dataset)
175
+
176
+ serial_proc = SerialPreprocessor(
177
+ SerialConfig(
178
+ input_data_path,
179
+ output_data_path,
180
+ TASK,
181
+ logger_name=TASK,
182
+ task_bos_token=f"[{TASK}]",
183
+ prompt_func=const_prompt_func_wrapper("Give all the aspects."),
184
+ knowledge_func=None_knowledge,
185
+ label_func=extract_aspects_wrapper(" | ", ", "),
186
+ )
187
+ )
188
+
189
+ serial_proc.launch()
190
+
191
+ # 4.4 ABSA: term category sentiment
192
+ TASK = ABSA_TERM_CATEGORY_SENTIMENT
193
+ input_path = r"E:\research\processed\ABSA-term category sentiment"
194
+ output_path = r"E:\research\seq\Aspect-based Sentiment Analysis"
195
+
196
+ for dataset in os.listdir(input_path):
197
+ input_data_path = os.path.join(input_path, dataset)
198
+ output_data_path = os.path.join(output_path, dataset)
199
+
200
+ serial_proc = SerialPreprocessor(
201
+ SerialConfig(
202
+ input_data_path,
203
+ output_data_path,
204
+ TASK,
205
+ logger_name=TASK,
206
+ task_bos_token=f"[{TASK}]",
207
+ prompt_func=const_prompt_func_wrapper("Give all the aspects."),
208
+ knowledge_func=None_knowledge,
209
+ label_func=extract_aspects_wrapper(" | ", ", "),
210
+ )
211
+ )
212
+
213
+ serial_proc.launch()
214
+
215
+ # 5. Character Identification
216
+ TASK = CHARACTER_IDENTIFICATION
217
+ input_path = r"E:\research\processed\CharacterIdentification"
218
+ output_path = r"E:\research\seq\CharacterIdentification"
219
+
220
+ for dataset in os.listdir(input_path):
221
+ input_data_path = os.path.join(input_path, dataset)
222
+ output_data_path = os.path.join(output_path, dataset)
223
+
224
+ serial_proc = SerialPreprocessor(
225
+ SerialConfig(
226
+ input_data_path,
227
+ output_data_path,
228
+ TASK,
229
+ logger_name=TASK,
230
+ task_bos_token=f"[{TASK}]",
231
+ prompt_func=const_prompt_func_wrapper("Generate with all characters."),
232
+ knowledge_func=concat_list_knowledge_wrapper("all speakers: ", " | "),
233
+ label_func=rebuild_utterance_with_characters,
234
+ )
235
+ )
236
+
237
+ serial_proc.launch()
238
+
239
+
240
+ # 6. Dialogue State Tracking
241
+ TASK = DIALOGUE_STATE_TRACKING
242
+ input_path = r"E:\research\processed\DialogueStateTracking"
243
+ output_path = r"E:\research\seq\DialogueStateTracking"
244
+
245
+ for dataset in os.listdir(input_path):
246
+ input_data_path = os.path.join(input_path, dataset)
247
+ output_data_path = os.path.join(output_path, dataset)
248
+
249
+ serial_proc = SerialPreprocessor(
250
+ SerialConfig(
251
+ input_data_path,
252
+ output_data_path,
253
+ TASK,
254
+ logger_name=TASK,
255
+ task_bos_token=f"[{TASK}]",
256
+ prompt_func=const_prompt_func_wrapper(
257
+ "With given dialogue context, give the dialogue state."
258
+ ),
259
+ knowledge_func=None_knowledge,
260
+ label_func=extract_belief_state_wrapper(", ", " | ", "; ", ": "),
261
+ roles_to_build_example=[["USER"]],
262
+ )
263
+ )
264
+
265
+ serial_proc.launch()
266
+
267
+ # 7. Document Grounded Conversation
268
+ TASK = DOCUMENT_GROUNDED_CONVERSATION
269
+ input_path = r"E:\research\processed\DocumentGroundedConversations"
270
+ output_path = r"E:\research\seq\DocumentGroundedConversation"
271
+
272
+ for dataset in os.listdir(input_path):
273
+ input_data_path = os.path.join(input_path, dataset)
274
+ output_data_path = os.path.join(output_path, dataset)
275
+
276
+ serial_proc = SerialPreprocessor(
277
+ SerialConfig(
278
+ input_data_path,
279
+ output_data_path,
280
+ TASK,
281
+ logger_name=TASK,
282
+ task_bos_token=f"[{TASK}]",
283
+ prompt_func=const_prompt_func_wrapper(
284
+ "With given dialogue context, give the response."
285
+ ),
286
+ knowledge_func=origin_knowledge,
287
+ turn_knowledge_func=extract_turn_knowledge_wrapper(": ", " | ", "; "),
288
+ label_func=extract_turn_utterance,
289
+ )
290
+ )
291
+
292
+ serial_proc.launch()
293
+ # 8. Text2SQL
294
+ TASK = TEXT2SQL
295
+ input_path = r"E:\research\processed\Text2SQL"
296
+ output_path = r"E:\research\seq\Text2SQL"
297
+
298
+ for dataset in os.listdir(input_path):
299
+ input_data_path = os.path.join(input_path, dataset)
300
+ output_data_path = os.path.join(output_path, dataset)
301
+
302
+ serial_proc = SerialPreprocessor(
303
+ SerialConfig(
304
+ input_data_path,
305
+ output_data_path,
306
+ TASK,
307
+ logger_name=TASK,
308
+ task_bos_token=f"[{TASK}]",
309
+ prompt_func=const_prompt_func_wrapper(
310
+ "With given dialogue context, give the sql."
311
+ ),
312
+ knowledge_func=origin_knowledge,
313
+ turn_knowledge_func=extract_schema_knowledge_wrapper(),
314
+ label_func=extract_sql,
315
+ )
316
+ )
317
+
318
+ serial_proc.launch()
319
+
320
+ TASK = SLOT_FILLING
321
+ input_path = r"E:\research\processed\SlotFilling\MultiDoGo"
322
+ output_path = r"E:\research\seq\SlotFilling\MultiDoGo"
323
+
324
+ for dataset in os.listdir(input_path):
325
+ input_data_path = os.path.join(input_path, dataset)
326
+ output_data_path = os.path.join(output_path, dataset)
327
+
328
+ serial_proc = SerialPreprocessor(
329
+ SerialConfig(
330
+ input_data_path,
331
+ output_data_path,
332
+ TASK,
333
+ logger_name=TASK,
334
+ task_bos_token=f"[{TASK}]",
335
+ prompt_func=const_prompt_func_wrapper(
336
+ "With given utterance, fill the slots."
337
+ ),
338
+ knowledge_func=None_knowledge,
339
+ label_func=extract_slots_without_intents_wrapper(", ", " | "),
340
+ )
341
+ )
342
+
343
+ serial_proc.launch()
src/modules/preprocess/preprocessor/SerialPreprocessor.py ADDED
@@ -0,0 +1,428 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Several preprocessor classes.
3
+ Author: md
4
+ """
5
+
6
+ from preprocessor.base import BasePreprocessorConfig, BasePreprocessor
7
+ from const import (
8
+ DIALOGUE_SUMMARY,
9
+ DIALOGUE_CONTEXT_TO_RESPONSE_GENERATION,
10
+ DIALOG,
11
+ KNOWLEDGE,
12
+ UTTERANCE,
13
+ ROLES,
14
+ EMOTION_RECOGNITION,
15
+ VALUE,
16
+ ABSA,
17
+ CHARACTER_IDENTIFICATION,
18
+ DIALOGUE_STATE_TRACKING,
19
+ DOCUMENT_GROUNDED_CONVERSATION,
20
+ TEXT2SQL,
21
+ SLOT_FILLING,
22
+ ROLE_RELATION_RECOGNITION,
23
+ QUESTION_IN_CONTEXT_REWRITING,
24
+ NATURAL_LANGUAGE_INFERENCE,
25
+ MACHINE_READING_COMPREHENSION,
26
+ MULTIPLE_CHOICE_QUESTION_ANSWERING,
27
+ INTENT_DETECTION,
28
+ DATA_TO_TEXT,
29
+ CHIT_CHAT,
30
+ TRAIN_SPLIT,
31
+ )
32
+ from typing import Dict, List, Callable
33
+ from copy import deepcopy
34
+
35
+
36
+ class SerialConfig(BasePreprocessorConfig):
37
+ def __init__(
38
+ self,
39
+ input_dir: str,
40
+ output_dir: str,
41
+ task: str,
42
+ task_bos_token: str = "<s>",
43
+ knowledge_bos_token: str = "[EK]",
44
+ prompt_bos_token: str = "[C]",
45
+ use_role: bool = True,
46
+ turn_sep: str = None,
47
+ roles_to_build_example: List = None,
48
+ dev_and_test_roles_to_build_example: List = None,
49
+ prompt_func: Callable = None,
50
+ knowledge_func: Callable = None,
51
+ label_func: Callable = None,
52
+ turn_knowledge_func: Callable = None,
53
+ roles_in_history: List[List] = None,
54
+ cur_turn_process_func: Callable = None,
55
+ all_turns_process_func: Callable = None,
56
+ multi_ref_sep: str = None,
57
+ *args,
58
+ **kwargs,
59
+ ) -> None:
60
+ super().__init__(input_dir, output_dir, task, *args, **kwargs)
61
+
62
+ self.use_role = use_role
63
+ self.turn_sep = turn_sep
64
+ self.roles_to_build_example = roles_to_build_example
65
+ self.prompt_func = prompt_func
66
+ self.task_bos_token = task_bos_token
67
+ self.knowledge_bos_token = knowledge_bos_token
68
+ self.prompt_bos_token = prompt_bos_token
69
+ self.knowledge_func = knowledge_func
70
+ self.label_func = label_func
71
+ self.turn_knowledge_func = turn_knowledge_func
72
+ self.roles_in_history = roles_in_history
73
+ self.multi_ref_sep = multi_ref_sep
74
+ self.dev_and_test_roles_to_build_example = dev_and_test_roles_to_build_example
75
+ self.cur_turn_process_func = cur_turn_process_func
76
+ self.all_turns_process_func = all_turns_process_func
77
+
78
+
79
+ def concat_roles(roles):
80
+ return ", ".join(roles)
81
+
82
+
83
+ def concat_dial_history(config: SerialConfig, history: List[Dict]):
84
+ # utterance_list = [
85
+ # f"{concat_roles(turn[ROLES])}: {turn[UTTERANCE].strip()}"
86
+ # if config.use_role
87
+ # else turn[UTTERANCE].strip()
88
+ # for turn in history
89
+ # ]
90
+
91
+ utterance_list = []
92
+ for turn in history:
93
+ if (
94
+ config.roles_in_history is not None
95
+ and turn[ROLES] not in config.roles_in_history
96
+ ):
97
+ continue
98
+
99
+ if config.use_role:
100
+ utterance_list.append(
101
+ f"{concat_roles(turn[ROLES])}: {turn[UTTERANCE].strip()}"
102
+ )
103
+ else:
104
+ utterance_list.append(turn[UTTERANCE].strip())
105
+
106
+ if not utterance_list:
107
+ return "None"
108
+
109
+ turn_sep = " "
110
+ if config.turn_sep is not None:
111
+ turn_sep = f" {config.turn_sep} "
112
+
113
+ return turn_sep.join(utterance_list)
114
+
115
+
116
+ def concat_history_knowledge_prompt(
117
+ config: SerialConfig, history: str, knowledge: str = "", prompt: str = ""
118
+ ):
119
+ """Concat `history`, `knowledge` and `prompt`.
120
+
121
+ NOTE: the order is fixed now.
122
+ """
123
+ text = ""
124
+
125
+ if config.task_bos_token is not None:
126
+ text = f"{config.task_bos_token} "
127
+
128
+ text += history
129
+
130
+ if knowledge is not None:
131
+ text += f" {config.knowledge_bos_token} {knowledge}"
132
+
133
+ if prompt is not None:
134
+ text += f" {config.prompt_bos_token} {prompt}"
135
+
136
+ return text
137
+
138
+
139
+ def clean(text):
140
+ return text.replace("\r\n", " ").replace("\n", " ").replace("\r", " ")
141
+
142
+
143
+ def add_prefix_to_label(prefix, split, label):
144
+ tgt = f"{prefix} {label}" if split == "train" else label
145
+ return tgt
146
+
147
+
148
+ class SerialPreprocessor(BasePreprocessor):
149
+ def __init__(self, config: SerialConfig) -> None:
150
+ super().__init__(config)
151
+
152
+ def extract_knowledge(self, example: Dict):
153
+ if self.config.knowledge_func is None:
154
+ knowledge = None
155
+
156
+ elif (
157
+ KNOWLEDGE not in example
158
+ or not self.config.knowledge_func.__code__.co_argcount
159
+ ):
160
+ knowledge = self.config.knowledge_func()
161
+ else:
162
+ knowledge = self.config.knowledge_func(example[KNOWLEDGE][VALUE])
163
+
164
+ return knowledge
165
+
166
+ def preprocess_for_dialogue_level(self, split: str, example: Dict, knowledge: str):
167
+ label = self.config.label_func(example)
168
+ tgt = add_prefix_to_label(self.config.task_bos_token, split, label)
169
+
170
+ history = concat_dial_history(self.config, example[DIALOG])
171
+
172
+ if self.config.prompt_func is None:
173
+ prompt = ""
174
+ elif not self.config.prompt_func.__code__.co_argcount:
175
+ prompt = self.config.prompt_func()
176
+
177
+ src = concat_history_knowledge_prompt(self.config, history, knowledge, prompt)
178
+
179
+ return [{"src": clean(src), "tgt": clean(tgt)}]
180
+
181
+ def preprocess_for_label_level(self, split: str, example: Dict, knowledge: str):
182
+ label_generator = self.config.label_func(example)
183
+
184
+ examples = []
185
+ for turn_id, label, extra_args in label_generator:
186
+ tgt = add_prefix_to_label(self.config.task_bos_token, split, label)
187
+
188
+ hist = deepcopy(example[DIALOG])
189
+ if self.config.all_turns_process_func is not None:
190
+ hist[turn_id] = self.config.all_turns_process_func(
191
+ hist[turn_id], *extra_args
192
+ )
193
+
194
+ history = concat_dial_history(self.config, hist)
195
+
196
+ if self.config.prompt_func is None:
197
+ prompt = ""
198
+ elif not self.config.prompt_func.__code__.co_argcount:
199
+ prompt = self.config.prompt_func()
200
+
201
+ src = concat_history_knowledge_prompt(
202
+ self.config, history, knowledge, prompt
203
+ )
204
+
205
+ examples.append({"src": clean(src), "tgt": clean(tgt)})
206
+
207
+ return examples
208
+
209
+ def get_label(
210
+ self, turn, include_current_turn, turn_idx, split, origin_knowledge=None
211
+ ):
212
+ # skip the roles not requiring to build examples
213
+ if (
214
+ split != TRAIN_SPLIT
215
+ and self.config.dev_and_test_roles_to_build_example is not None
216
+ ):
217
+ roles_to_build_example = self.config.dev_and_test_roles_to_build_example
218
+ else:
219
+ roles_to_build_example = self.config.roles_to_build_example
220
+ if (
221
+ roles_to_build_example is not None
222
+ and turn[ROLES] not in roles_to_build_example
223
+ ):
224
+ return None
225
+
226
+ # skip the first turn if not including current turn
227
+ if not include_current_turn and turn_idx == 0:
228
+ return None
229
+
230
+ if self.config.task != DIALOGUE_STATE_TRACKING:
231
+ try:
232
+ label = self.config.label_func(turn, split=split)
233
+ except:
234
+ label = self.config.label_func(turn, origin_knowledge, split=split)
235
+ else:
236
+ label = self.config.label_func(
237
+ turn, self.ontologies[split], do_train=(split == TRAIN_SPLIT)
238
+ )
239
+
240
+ return label
241
+
242
+ def preprocess_for_turn_level(
243
+ self,
244
+ split: str,
245
+ example: Dict,
246
+ knowledge: str,
247
+ include_current_turn=False,
248
+ origin_knowledge=None,
249
+ ):
250
+ examples = []
251
+ multiref = []
252
+ for turn_idx, turn in enumerate(example[DIALOG]):
253
+ label = self.get_label(
254
+ turn, include_current_turn, turn_idx, split, origin_knowledge
255
+ )
256
+
257
+ if label is None:
258
+ continue
259
+
260
+ multiref.append(label)
261
+ # requre to merge and arrive at the final consecutive label
262
+ if (
263
+ self.config.multi_ref_sep is not None
264
+ and split != "train"
265
+ and turn_idx < len(example[DIALOG]) - 1
266
+ and self.get_label(
267
+ example[DIALOG][turn_idx + 1],
268
+ include_current_turn,
269
+ turn_idx + 1,
270
+ split,
271
+ )
272
+ is not None
273
+ ):
274
+ continue
275
+
276
+ if self.config.multi_ref_sep is not None and split != "train":
277
+ label = self.config.multi_ref_sep.join(multiref)
278
+
279
+ tgt = add_prefix_to_label(self.config.task_bos_token, split, label)
280
+
281
+ end = (turn_idx + 1) if include_current_turn else turn_idx
282
+
283
+ hist = deepcopy(example[DIALOG][:end])
284
+ if self.config.cur_turn_process_func is not None:
285
+ hist[-1] = self.config.cur_turn_process_func(hist[-1])
286
+
287
+ history = concat_dial_history(self.config, hist)
288
+
289
+ if self.config.prompt_func is None:
290
+ prompt = ""
291
+ elif not self.config.prompt_func.__code__.co_argcount:
292
+ prompt = self.config.prompt_func()
293
+
294
+ if self.config.turn_knowledge_func is not None:
295
+ knowledge_to_use = self.config.turn_knowledge_func(knowledge, turn)
296
+ else:
297
+ knowledge_to_use = knowledge
298
+
299
+ src = concat_history_knowledge_prompt(
300
+ self.config, history, knowledge_to_use, prompt
301
+ )
302
+
303
+ examples.append({"src": clean(src), "tgt": clean(tgt)})
304
+
305
+ multiref = []
306
+
307
+ return examples
308
+
309
+ def preprocess_line(self, split: str, example: Dict) -> List[Dict]:
310
+ knowledge = self.extract_knowledge(example)
311
+
312
+ # 1. Dialogue Summary
313
+ if self.config.task == DIALOGUE_SUMMARY:
314
+ return self.preprocess_for_dialogue_level(split, example, knowledge)
315
+
316
+ # 2. Emotion Recognition
317
+ if self.config.task == EMOTION_RECOGNITION:
318
+ return self.preprocess_for_turn_level(
319
+ split, example, knowledge, include_current_turn=True
320
+ )
321
+
322
+ # 3. Dialogue Context-to-Text Generation
323
+ if self.config.task == DIALOGUE_CONTEXT_TO_RESPONSE_GENERATION:
324
+ return self.preprocess_for_turn_level(
325
+ split, example, knowledge, include_current_turn=False
326
+ )
327
+
328
+ # 4. ABSA
329
+ if self.config.task.startswith(ABSA):
330
+ return self.preprocess_for_turn_level(
331
+ split, example, knowledge, include_current_turn=True
332
+ )
333
+
334
+ # 5. Character Identification
335
+ if self.config.task == CHARACTER_IDENTIFICATION:
336
+ # return self.preprocess_for_turn_level(
337
+ # split, example, knowledge, include_current_turn=True
338
+ # )
339
+ # return self.preprocess_for_dialogue_level(split, example, knowledge)
340
+ return self.preprocess_for_label_level(split, example, knowledge)
341
+
342
+ # 6. Dialogue State Tracking
343
+ if self.config.task == DIALOGUE_STATE_TRACKING:
344
+ return self.preprocess_for_turn_level(
345
+ split, example, knowledge, include_current_turn=True
346
+ )
347
+
348
+ # 7. Document Grounded Conversation
349
+ if self.config.task == DOCUMENT_GROUNDED_CONVERSATION:
350
+ return self.preprocess_for_turn_level(
351
+ split, example, knowledge, include_current_turn=False
352
+ )
353
+
354
+ # 8. Text2SQL
355
+ if self.config.task == TEXT2SQL:
356
+ seq_examples = self.preprocess_for_turn_level(
357
+ split, example, knowledge, include_current_turn=True
358
+ )
359
+
360
+ for idx in range(len(seq_examples)):
361
+ seq_examples[idx]["db_id"] = knowledge["db_id"]
362
+
363
+ return seq_examples
364
+
365
+ # 9. Slot Filling
366
+ if self.config.task == SLOT_FILLING:
367
+ return self.preprocess_for_turn_level(
368
+ split, example, knowledge, include_current_turn=True
369
+ )
370
+
371
+ # 10. Relation Recognition
372
+ if self.config.task == ROLE_RELATION_RECOGNITION:
373
+ return self.preprocess_for_dialogue_level(split, example, knowledge)
374
+
375
+ # 11. Question in Context Rewriting
376
+ if self.config.task == QUESTION_IN_CONTEXT_REWRITING:
377
+ return self.preprocess_for_turn_level(
378
+ split, example, knowledge, include_current_turn=True
379
+ )
380
+
381
+ # 12. Natural Language Inference
382
+ if self.config.task == NATURAL_LANGUAGE_INFERENCE:
383
+ return self.preprocess_for_turn_level(
384
+ split,
385
+ example,
386
+ knowledge,
387
+ include_current_turn=True,
388
+ origin_knowledge=example[KNOWLEDGE][VALUE],
389
+ )
390
+
391
+ # 13. Machine Reading Comprehension
392
+ if self.config.task == MACHINE_READING_COMPREHENSION:
393
+ return self.preprocess_for_turn_level(split, example, knowledge)
394
+
395
+ # 14. Multiple Choice Question Answering
396
+ if self.config.task == MULTIPLE_CHOICE_QUESTION_ANSWERING:
397
+ return self.preprocess_for_turn_level(
398
+ split,
399
+ example,
400
+ knowledge,
401
+ include_current_turn=True,
402
+ origin_knowledge=example[KNOWLEDGE][VALUE],
403
+ )
404
+
405
+ # 15. Intent Detection
406
+ if self.config.task == INTENT_DETECTION:
407
+ return self.preprocess_for_turn_level(
408
+ split, example, knowledge, include_current_turn=True
409
+ )
410
+
411
+ # 16. Data-to-Text
412
+ if self.config.task == DATA_TO_TEXT:
413
+ return self.preprocess_for_turn_level(
414
+ split, example, knowledge, include_current_turn=True
415
+ )
416
+
417
+ # 17. Chit-Chat
418
+ if self.config.task == CHIT_CHAT:
419
+ return self.preprocess_for_turn_level(
420
+ split, example, knowledge, include_current_turn=False
421
+ )
422
+
423
+ if self.config.task == "Semantic Parsing":
424
+ seq_examples = self.preprocess_for_turn_level(
425
+ split, example, knowledge, include_current_turn=True
426
+ )
427
+
428
+ return seq_examples
src/modules/preprocess/preprocessor/__pycache__/SerialPreprocessor.cpython-312.pyc ADDED
Binary file (15 kB). View file
 
src/modules/preprocess/preprocessor/__pycache__/SerialPreprocessor.cpython-38.pyc ADDED
Binary file (7.83 kB). View file
 
src/modules/preprocess/preprocessor/__pycache__/base.cpython-312.pyc ADDED
Binary file (8.5 kB). View file
 
src/modules/preprocess/preprocessor/__pycache__/base.cpython-38.pyc ADDED
Binary file (4.48 kB). View file
 
src/modules/preprocess/preprocessor/__pycache__/knowledge_funcs.cpython-312.pyc ADDED
Binary file (20.9 kB). View file
 
src/modules/preprocess/preprocessor/__pycache__/knowledge_funcs.cpython-38.pyc ADDED
Binary file (15.1 kB). View file
 
src/modules/preprocess/preprocessor/__pycache__/label_funs.cpython-312.pyc ADDED
Binary file (14 kB). View file
 
src/modules/preprocess/preprocessor/__pycache__/label_funs.cpython-38.pyc ADDED
Binary file (9.78 kB). View file
 
src/modules/preprocess/preprocessor/__pycache__/process_turn_funcs.cpython-38.pyc ADDED
Binary file (937 Bytes). View file
 
src/modules/preprocess/preprocessor/__pycache__/prompt_funcs.cpython-312.pyc ADDED
Binary file (497 Bytes). View file
 
src/modules/preprocess/preprocessor/__pycache__/prompt_funcs.cpython-38.pyc ADDED
Binary file (468 Bytes). View file
 
src/modules/preprocess/preprocessor/base.py ADDED
@@ -0,0 +1,150 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Base preprocessor class.
3
+ Author: md
4
+ """
5
+
6
+ from config import BaseConfig
7
+ import os
8
+ from logger import build_logger
9
+ import logging
10
+ import json
11
+ from const import TRAIN_SPLIT, DEV_SPLIT, TEST_SPLIT, DIALOGUE_STATE_TRACKING
12
+ from typing import Dict
13
+ import shutil
14
+
15
+
16
+ class BasePreprocessorConfig(BaseConfig):
17
+ def __init__(
18
+ self,
19
+ input_dir: str,
20
+ output_dir: str,
21
+ task: str,
22
+ formatter="%(asctime)s | [%(name)s] | %(levelname)s | %(message)s",
23
+ *args,
24
+ **kwargs,
25
+ ) -> None:
26
+ super().__init__(*args, **kwargs)
27
+
28
+ self.input_dir = input_dir
29
+ self.output_dir = output_dir
30
+ self.task = task
31
+ self.formatter = formatter
32
+
33
+
34
+ class BasePreprocessor(object):
35
+ def __init__(self, config: BasePreprocessorConfig) -> None:
36
+ self.config = config
37
+ self.logger = build_logger(
38
+ config.logger_name,
39
+ logging.INFO,
40
+ config.log_file,
41
+ config.log_mode,
42
+ config.formatter,
43
+ )
44
+
45
+ if self.config.task == DIALOGUE_STATE_TRACKING:
46
+ self.ontologies = {
47
+ split: self.load_ontology(split)
48
+ for split in [TRAIN_SPLIT, DEV_SPLIT, TEST_SPLIT]
49
+ }
50
+
51
+ def load_ontology(self, split: str) -> Dict:
52
+ """
53
+ Load the ontology file.
54
+ """
55
+ ontology_file = os.path.join(self.config.input_dir, f"{split}_ontology.json")
56
+ if not os.path.exists(ontology_file):
57
+ return None
58
+ return json.load(open(ontology_file, "r", encoding="utf8"))
59
+
60
+ def preprocess_line(self, split: str, example: Dict) -> Dict:
61
+ """
62
+ Every preprocessor should customize this function for all `train`, `dev` and `test` split.
63
+ """
64
+ raise NotImplementedError("The preprocess line procedure is required!")
65
+
66
+ def _preprocess_file(
67
+ self, start, infile, src_writer, tgt_writer, split, encoding="UTF-8"
68
+ ):
69
+ with open(infile, "r", encoding=encoding) as reader:
70
+ for line in reader:
71
+ if line.strip():
72
+ example = json.loads(line)
73
+ if start:
74
+ start = False
75
+ elif split != "train":
76
+ tgt_writer.write("\n")
77
+ for processed_example in self.preprocess_line(split, example):
78
+ src_writer.write(f"{processed_example['src']}\n")
79
+ tgt_writer.write(f"{processed_example['tgt']}")
80
+
81
+ if "db_id" in processed_example and split != "train":
82
+ tgt_writer.write(f"\t{processed_example['db_id']}")
83
+ tgt_writer.write("\n")
84
+ return start
85
+
86
+ def preprocess(self, split: str) -> bool:
87
+ if not os.path.exists(self.config.output_dir):
88
+ os.makedirs(self.config.output_dir)
89
+
90
+ src_file = os.path.join(self.config.output_dir, f"{split}.src")
91
+ tgt_file = os.path.join(
92
+ self.config.output_dir,
93
+ f"{split}.tgt" if split == "train" else f"{split}.gold",
94
+ )
95
+ exist = False
96
+
97
+ with open(src_file, "w") as src_writer, open(tgt_file, "w") as tgt_writer:
98
+ start = True
99
+ for filename in os.listdir(self.config.input_dir):
100
+ if split not in filename or not filename.endswith(".jsonl"):
101
+ continue
102
+
103
+ exist = True
104
+ infile = os.path.join(self.config.input_dir, filename)
105
+
106
+ self.logger.info(f"preprocessing {infile}")
107
+ try:
108
+ start = self._preprocess_file(
109
+ start, infile, src_writer, tgt_writer, split
110
+ )
111
+ except UnicodeDecodeError:
112
+ start = self._preprocess_file(
113
+ start, infile, src_writer, tgt_writer, split, "ISO-8859-1"
114
+ )
115
+
116
+ return exist
117
+
118
+ def launch(self) -> None:
119
+ self.logger.info(f"Start to preprocess: {TRAIN_SPLIT}")
120
+ train = self.preprocess(TRAIN_SPLIT)
121
+ assert train
122
+
123
+ self.logger.info(f"Start to preprocess: {DEV_SPLIT}")
124
+ dev = self.preprocess(DEV_SPLIT)
125
+ self.logger.info(f"Start to preprocess: {TEST_SPLIT}")
126
+ test = self.preprocess(TEST_SPLIT)
127
+
128
+ if dev and not test:
129
+ self.logger.info("Copy dev to test")
130
+ shutil.copyfile(
131
+ os.path.join(self.config.output_dir, "dev.src"),
132
+ os.path.join(self.config.output_dir, "test.src"),
133
+ )
134
+ shutil.copyfile(
135
+ os.path.join(self.config.output_dir, "dev.gold"),
136
+ os.path.join(self.config.output_dir, "test.gold"),
137
+ )
138
+
139
+ if test and not dev:
140
+ self.logger.info("Copy test to dev")
141
+ shutil.copyfile(
142
+ os.path.join(self.config.output_dir, "test.src"),
143
+ os.path.join(self.config.output_dir, "dev.src"),
144
+ )
145
+ shutil.copyfile(
146
+ os.path.join(self.config.output_dir, "test.gold"),
147
+ os.path.join(self.config.output_dir, "dev.gold"),
148
+ )
149
+
150
+ self.logger.info("Preprocess successfully!")
src/modules/preprocess/preprocessor/knowledge_funcs.py ADDED
@@ -0,0 +1,505 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from typing import List, Dict, Optional, Tuple
2
+ import random
3
+ import difflib
4
+ from rapidfuzz import fuzz
5
+ import sqlite3
6
+ import functools
7
+ from const import KNOWLEDGE_TO_SELECT, UTTERANCE, ROLES, BELIEF_STATE, DOMAIN
8
+
9
+
10
+ def None_knowledge():
11
+ return "None"
12
+
13
+
14
+ def concat_list_knowledge_wrapper(prompt: str = "", sep: str = " | "):
15
+ def get_list_knowledge(str_list: List[str]):
16
+ return prompt + sep.join(str_list)
17
+
18
+ return get_list_knowledge
19
+
20
+
21
+ def origin_knowledge(knowledge):
22
+ return knowledge
23
+
24
+
25
+ def extract_turn_knowledge(
26
+ knowledge, section_prompt_op, section_sep, section_value_sep
27
+ ):
28
+ if isinstance(knowledge, dict):
29
+ sec_list = []
30
+ for section in sorted(knowledge.keys()):
31
+ sec_str = f"{section}{section_prompt_op}"
32
+ if isinstance(knowledge[section], str):
33
+ sec_str += knowledge[section]
34
+ elif isinstance(knowledge[section], list):
35
+ sec_str += section_value_sep.join(knowledge[section])
36
+ sec_list.append(sec_str)
37
+
38
+ return section_sep.join(sec_list)
39
+
40
+ elif isinstance(knowledge, str):
41
+ return knowledge
42
+
43
+ elif isinstance(knowledge, list):
44
+ return ";; ".join(
45
+ [
46
+ extract_turn_knowledge(
47
+ sec, section_prompt_op, section_sep, section_value_sep
48
+ )
49
+ for sec in knowledge
50
+ ]
51
+ )
52
+
53
+
54
+ def extract_turn_domains_wrapper(prompt: str = "", sep: str = ", "):
55
+ def extract_turn_domains(knowledge, turn):
56
+ bs = turn[BELIEF_STATE]
57
+ domains = []
58
+ for state in bs:
59
+ domain = state[DOMAIN]
60
+ if domain not in domains:
61
+ domains.append(domain)
62
+
63
+ return prompt + sep.join(domains)
64
+
65
+ return extract_turn_domains
66
+
67
+
68
+ def extract_turn_knowledge_wrapper(section_prompt_op, section_sep, section_value_sep):
69
+ def extract_turn_knowledge_func(knowledge, turn):
70
+ return extract_turn_knowledge(
71
+ [knowledge[sec] for sec in turn[KNOWLEDGE_TO_SELECT]],
72
+ section_prompt_op,
73
+ section_sep,
74
+ section_value_sep,
75
+ )
76
+
77
+ return extract_turn_knowledge_func
78
+
79
+
80
+ # Text2SQL
81
+ EXIST = {"atis", "geo", "advising", "yelp", "restaurants", "imdb", "academic"}
82
+
83
+ # fmt: off
84
+ _stopwords = {'who', 'ourselves', 'down', 'only', 'were', 'him', 'at', "weren't", 'has', 'few', "it's", 'm', 'again',
85
+ 'd', 'haven', 'been', 'other', 'we', 'an', 'own', 'doing', 'ma', 'hers', 'all', "haven't", 'in', 'but',
86
+ "shouldn't", 'does', 'out', 'aren', 'you', "you'd", 'himself', "isn't", 'most', 'y', 'below', 'is',
87
+ "wasn't", 'hasn', 'them', 'wouldn', 'against', 'this', 'about', 'there', 'don', "that'll", 'a', 'being',
88
+ 'with', 'your', 'theirs', 'its', 'any', 'why', 'now', 'during', 'weren', 'if', 'should', 'those', 'be',
89
+ 'they', 'o', 't', 'of', 'or', 'me', 'i', 'some', 'her', 'do', 'will', 'yours', 'for', 'mightn', 'nor',
90
+ 'needn', 'the', 'until', "couldn't", 'he', 'which', 'yourself', 'to', "needn't", "you're", 'because',
91
+ 'their', 'where', 'it', "didn't", 've', 'whom', "should've", 'can', "shan't", 'on', 'had', 'have',
92
+ 'myself', 'am', "don't", 'under', 'was', "won't", 'these', 'so', 'as', 'after', 'above', 'each', 'ours',
93
+ 'hadn', 'having', 'wasn', 's', 'doesn', "hadn't", 'than', 'by', 'that', 'both', 'herself', 'his',
94
+ "wouldn't", 'into', "doesn't", 'before', 'my', 'won', 'more', 'are', 'through', 'same', 'how', 'what',
95
+ 'over', 'll', 'yourselves', 'up', 'mustn', "mustn't", "she's", 're', 'such', 'didn', "you'll", 'shan',
96
+ 'when', "you've", 'themselves', "mightn't", 'she', 'from', 'isn', 'ain', 'between', 'once', 'here',
97
+ 'shouldn', 'our', 'and', 'not', 'too', 'very', 'further', 'while', 'off', 'couldn', "hasn't", 'itself',
98
+ 'then', 'did', 'just', "aren't"}
99
+ # fmt: on
100
+
101
+ _commonwords = {"no", "yes", "many"}
102
+
103
+
104
+ def is_number(s: str) -> bool:
105
+ try:
106
+ float(s.replace(",", ""))
107
+ return True
108
+ except:
109
+ return False
110
+
111
+
112
+ def is_stopword(s: str) -> bool:
113
+ return s.strip() in _stopwords
114
+
115
+
116
+ def is_commonword(s: str) -> bool:
117
+ return s.strip() in _commonwords
118
+
119
+
120
+ def is_common_db_term(s: str) -> bool:
121
+ return s.strip() in ["id"]
122
+
123
+
124
+ class Match(object):
125
+ def __init__(self, start: int, size: int) -> None:
126
+ self.start = start
127
+ self.size = size
128
+
129
+
130
+ def is_span_separator(c: str) -> bool:
131
+ return c in "'\"()`,.?! "
132
+
133
+
134
+ def split(s: str) -> List[str]:
135
+ return [c.lower() for c in s.strip()]
136
+
137
+
138
+ def prefix_match(s1: str, s2: str) -> bool:
139
+ i, j = 0, 0
140
+ for i in range(len(s1)):
141
+ if not is_span_separator(s1[i]):
142
+ break
143
+ for j in range(len(s2)):
144
+ if not is_span_separator(s2[j]):
145
+ break
146
+ if i < len(s1) and j < len(s2):
147
+ return s1[i] == s2[j]
148
+ elif i >= len(s1) and j >= len(s2):
149
+ return True
150
+ else:
151
+ return False
152
+
153
+
154
+ def get_effective_match_source(s: str, start: int, end: int) -> Match:
155
+ _start = -1
156
+
157
+ for i in range(start, start - 2, -1):
158
+ if i < 0:
159
+ _start = i + 1
160
+ break
161
+ if is_span_separator(s[i]):
162
+ _start = i
163
+ break
164
+
165
+ if _start < 0:
166
+ return None
167
+
168
+ _end = -1
169
+ for i in range(end - 1, end + 3):
170
+ if i >= len(s):
171
+ _end = i - 1
172
+ break
173
+ if is_span_separator(s[i]):
174
+ _end = i
175
+ break
176
+
177
+ if _end < 0:
178
+ return None
179
+
180
+ while _start < len(s) and is_span_separator(s[_start]):
181
+ _start += 1
182
+ while _end >= 0 and is_span_separator(s[_end]):
183
+ _end -= 1
184
+
185
+ return Match(_start, _end - _start + 1)
186
+
187
+
188
+ def get_matched_entries(
189
+ s: str, field_values: List[str], m_theta: float = 0.85, s_theta: float = 0.85
190
+ ) -> Optional[List[Tuple[str, Tuple[str, str, float, float, int]]]]:
191
+ if not field_values:
192
+ return None
193
+
194
+ if isinstance(s, str):
195
+ n_grams = split(s)
196
+ else:
197
+ n_grams = s
198
+
199
+ matched = dict()
200
+ for field_value in field_values:
201
+ if not isinstance(field_value, str):
202
+ continue
203
+ fv_tokens = split(field_value)
204
+ sm = difflib.SequenceMatcher(None, n_grams, fv_tokens)
205
+ match = sm.find_longest_match(0, len(n_grams), 0, len(fv_tokens))
206
+ if match.size > 0:
207
+ source_match = get_effective_match_source(
208
+ n_grams, match.a, match.a + match.size
209
+ )
210
+ if source_match and source_match.size > 1:
211
+ match_str = field_value[match.b : match.b + match.size]
212
+ source_match_str = s[
213
+ source_match.start : source_match.start + source_match.size
214
+ ]
215
+ c_match_str = match_str.lower().strip()
216
+ c_source_match_str = source_match_str.lower().strip()
217
+ c_field_value = field_value.lower().strip()
218
+ if (
219
+ c_match_str
220
+ and not is_number(c_match_str)
221
+ and not is_common_db_term(c_match_str)
222
+ ):
223
+ if (
224
+ is_stopword(c_match_str)
225
+ or is_stopword(c_source_match_str)
226
+ or is_stopword(c_field_value)
227
+ ):
228
+ continue
229
+ if c_source_match_str.endswith(c_match_str + "'s"):
230
+ match_score = 1.0
231
+ else:
232
+ if prefix_match(c_field_value, c_source_match_str):
233
+ match_score = (
234
+ fuzz.ratio(c_field_value, c_source_match_str) / 100
235
+ )
236
+ else:
237
+ match_score = 0
238
+ if (
239
+ is_commonword(c_match_str)
240
+ or is_commonword(c_source_match_str)
241
+ or is_commonword(c_field_value)
242
+ ) and match_score < 1:
243
+ continue
244
+ s_match_score = match_score
245
+ if match_score >= m_theta and s_match_score >= s_theta:
246
+ if field_value.isupper() and match_score * s_match_score < 1:
247
+ continue
248
+ matched[match_str] = (
249
+ field_value,
250
+ source_match_str,
251
+ match_score,
252
+ s_match_score,
253
+ match.size,
254
+ )
255
+
256
+ if not matched:
257
+ return None
258
+ else:
259
+ return sorted(
260
+ matched.items(),
261
+ key=lambda x: (1e16 * x[1][2] + 1e8 * x[1][3] + x[1][4]),
262
+ reverse=True,
263
+ )
264
+
265
+
266
+ @functools.lru_cache(maxsize=1000, typed=False)
267
+ def get_column_picklist(table_name: str, column_name: str, db_path: str) -> list:
268
+ fetch_sql = "SELECT DISTINCT `{}` FROM `{}`".format(column_name, table_name)
269
+ try:
270
+ conn = sqlite3.connect(db_path)
271
+ conn.text_factory = bytes
272
+ c = conn.cursor()
273
+ c.execute(fetch_sql)
274
+ picklist = set()
275
+ for x in c.fetchall():
276
+ if isinstance(x[0], str):
277
+ picklist.add(x[0].encode("utf-8"))
278
+ elif isinstance(x[0], bytes):
279
+ try:
280
+ picklist.add(x[0].decode("utf-8"))
281
+ except UnicodeDecodeError:
282
+ picklist.add(x[0].decode("latin-1"))
283
+ else:
284
+ picklist.add(x[0])
285
+ picklist = list(picklist)
286
+ finally:
287
+ conn.close()
288
+ return picklist
289
+
290
+
291
+ def get_database_matches(
292
+ question: str,
293
+ table_name: str,
294
+ column_name: str,
295
+ db_path: str,
296
+ top_k_matches: int = 2,
297
+ match_threshold: float = 0.85,
298
+ ) -> List[str]:
299
+ picklist = get_column_picklist(
300
+ table_name=table_name, column_name=column_name, db_path=db_path
301
+ )
302
+ matches = []
303
+ if picklist and isinstance(picklist[0], str):
304
+ matched_entries = get_matched_entries(
305
+ s=question,
306
+ field_values=picklist,
307
+ m_theta=match_threshold,
308
+ s_theta=match_threshold,
309
+ )
310
+ if matched_entries:
311
+ num_values_inserted = 0
312
+ for _match_str, (
313
+ field_value,
314
+ _s_match_str,
315
+ match_score,
316
+ s_match_score,
317
+ _match_size,
318
+ ) in matched_entries:
319
+ if "name" in column_name and match_score * s_match_score < 1:
320
+ continue
321
+ if table_name != "sqlite_sequence": # Spider database artifact
322
+ matches.append(field_value)
323
+ num_values_inserted += 1
324
+ if num_values_inserted >= top_k_matches:
325
+ break
326
+ return matches
327
+
328
+
329
+ def serialize_schema(
330
+ question: str,
331
+ db_path: str,
332
+ db_id: str,
333
+ db_column_names: Dict[str, str],
334
+ db_table_names: List[str],
335
+ schema_serialization_type: str = "peteshaw",
336
+ schema_serialization_randomized: bool = False,
337
+ schema_serialization_with_db_id: bool = True,
338
+ schema_serialization_with_db_content: bool = False,
339
+ normalize_query: bool = True,
340
+ ) -> str:
341
+ if schema_serialization_type == "verbose":
342
+ db_id_str = "Database: {db_id}. "
343
+ table_sep = ". "
344
+ table_str = "Table: {table}. Columns: {columns}"
345
+ column_sep = ", "
346
+ column_str_with_values = "{column} ({values})"
347
+ column_str_without_values = "{column}"
348
+ value_sep = ", "
349
+ elif schema_serialization_type == "peteshaw":
350
+ # see https://github.com/google-research/language/blob/master/language/nqg/tasks/spider/append_schema.py#L42
351
+ db_id_str = "{db_id}"
352
+ table_sep = ""
353
+ table_str = " | {table} : {columns}"
354
+ column_sep = " , "
355
+ column_str_with_values = "{column} ( {values} )"
356
+ column_str_without_values = "{column}"
357
+ value_sep = " , "
358
+ else:
359
+ raise NotImplementedError
360
+
361
+ def get_column_str(table_name: str, column_name: str) -> str:
362
+ column_name_str = column_name.lower() if normalize_query else column_name
363
+ if schema_serialization_with_db_content:
364
+ matches = get_database_matches(
365
+ question=question,
366
+ table_name=table_name,
367
+ column_name=column_name,
368
+ db_path=(db_path + "/" + db_id + "/" + db_id + ".sqlite"),
369
+ )
370
+ if matches:
371
+ return column_str_with_values.format(
372
+ column=column_name_str, values=value_sep.join(matches)
373
+ )
374
+ else:
375
+ return column_str_without_values.format(column=column_name_str)
376
+ else:
377
+ return column_str_without_values.format(column=column_name_str)
378
+
379
+ tables = [
380
+ table_str.format(
381
+ table=table_name.lower() if normalize_query else table_name,
382
+ columns=column_sep.join(
383
+ map(
384
+ lambda y: get_column_str(table_name=table_name, column_name=y[1]),
385
+ filter(
386
+ lambda y: y[0] == table_id,
387
+ zip(
388
+ db_column_names["table_id"],
389
+ db_column_names["column_name"],
390
+ ),
391
+ ),
392
+ )
393
+ ),
394
+ )
395
+ for table_id, table_name in enumerate(db_table_names)
396
+ ]
397
+ if schema_serialization_randomized:
398
+ random.shuffle(tables)
399
+ if schema_serialization_with_db_id:
400
+ serialized_schema = db_id_str.format(db_id=db_id) + table_sep.join(tables)
401
+ else:
402
+ serialized_schema = table_sep.join(tables)
403
+ return serialized_schema
404
+
405
+
406
+ def extract_schema_knowledge_wrapper(
407
+ schema_serialization_type: str = "peteshaw",
408
+ schema_serialization_randomized: bool = False,
409
+ schema_serialization_with_db_id: bool = True,
410
+ schema_serialization_with_db_content: bool = False,
411
+ normalize_query: bool = True,
412
+ ):
413
+ def extract_turn_schema_knowledge_func(knowledge, turn):
414
+ schema = knowledge["schema"]
415
+ db_column_names = {
416
+ "table_id": [table_id for table_id, _ in schema["column_names_original"]],
417
+ "column_name": [
418
+ column_name for _, column_name in schema["column_names_original"]
419
+ ],
420
+ }
421
+ return serialize_schema(
422
+ turn[UTTERANCE],
423
+ knowledge["db_path"],
424
+ knowledge["db_id"],
425
+ db_column_names,
426
+ schema["table_names_original"],
427
+ schema_serialization_type,
428
+ schema_serialization_randomized,
429
+ schema_serialization_with_db_id,
430
+ schema_serialization_with_db_content,
431
+ normalize_query,
432
+ )
433
+
434
+ return extract_turn_schema_knowledge_func
435
+
436
+
437
+ def extract_dict_knowledge(knowledge, key_prompt_op, pair_sep):
438
+ pair_list = []
439
+ for key in knowledge:
440
+ pair_list.append(f"{key}{key_prompt_op}{knowledge[key]}")
441
+
442
+ if not pair_list:
443
+ return "None"
444
+
445
+ return pair_sep.join(pair_list)
446
+
447
+
448
+ def extract_dict_knowledge_wrapper(key_prompt_op, pair_sep):
449
+ def extract_dict_knowledge_func(knowledge):
450
+ return extract_dict_knowledge(knowledge, key_prompt_op, pair_sep)
451
+
452
+ return extract_dict_knowledge_func
453
+
454
+
455
+ def extract_dialogue_knowledge(knowledge, key_prompt_op, pair_sep, role_sep):
456
+ pair_list = []
457
+ for key in knowledge:
458
+ if isinstance(knowledge[key], str):
459
+ pair_list.append(f"{key}{key_prompt_op}{knowledge[key]}")
460
+ elif isinstance(knowledge[key], list):
461
+ turns = []
462
+ for turn in knowledge[key]:
463
+ role_str = role_sep.join(turn[ROLES])
464
+ turns.append(f"{role_str}# {turn[UTTERANCE]}")
465
+ dial_str = " ".join(turns)
466
+ pair_list.append(f"{key}{key_prompt_op}{dial_str}")
467
+ if not pair_list:
468
+ return "None"
469
+ return pair_sep.join(pair_list)
470
+
471
+
472
+ def extract_dialogue_knowledge_wrapper(key_prompt_op, pair_sep, role_sep):
473
+ def extract_dialogue_knowledge_func(knowledge):
474
+ return extract_dialogue_knowledge(knowledge, key_prompt_op, pair_sep, role_sep)
475
+
476
+ return extract_dialogue_knowledge_func
477
+
478
+
479
+ def extract_kg_knowledge(
480
+ knowledge, key_prompt_op, pair_sep, intra_edge_sep, inner_edge_sep
481
+ ):
482
+ pair_list = []
483
+ for key in knowledge:
484
+ if isinstance(knowledge[key], str):
485
+ pair_list.append(f"{key}{key_prompt_op}{knowledge[key]}")
486
+ elif isinstance(knowledge[key], list):
487
+ edges = []
488
+ for edge in knowledge[key]:
489
+ edges.append(inner_edge_sep.join(edge))
490
+ kg_str = intra_edge_sep.join(edges)
491
+ pair_list.append(f"{key}{key_prompt_op}{kg_str}")
492
+ if not pair_list:
493
+ return "None"
494
+ return pair_sep.join(pair_list)
495
+
496
+
497
+ def extract_kg_knowledge_wrapper(
498
+ key_prompt_op, pair_sep, intra_edge_sep, inner_edge_sep
499
+ ):
500
+ def extract_kg_knowledge_func(knowledge):
501
+ return extract_kg_knowledge(
502
+ knowledge, key_prompt_op, pair_sep, intra_edge_sep, inner_edge_sep
503
+ )
504
+
505
+ return extract_kg_knowledge_func
src/modules/preprocess/preprocessor/label_funs.py ADDED
@@ -0,0 +1,324 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from const import (
2
+ SUMMARY,
3
+ EMOTIONS,
4
+ EMOTION,
5
+ UTTERANCE,
6
+ ASPECTS,
7
+ TARGET,
8
+ VALUE,
9
+ OPINION,
10
+ SENTIMENT,
11
+ CATEGORY,
12
+ CHARACTERS,
13
+ DIALOG,
14
+ START,
15
+ END,
16
+ BELIEF_STATE,
17
+ DOMAIN,
18
+ INFORMED_SLOT_VALUE_TABLE,
19
+ SLOT,
20
+ VALUES,
21
+ RELATION,
22
+ SQL,
23
+ SLOT_VALUE_TABLE,
24
+ SLOTS_TO_FILL,
25
+ ROLE_RELATIONS,
26
+ REWRITTEN,
27
+ ROLES_TO_SELECT,
28
+ ACTIVE_INTENTS,
29
+ TRAIN_SPLIT,
30
+ OPTION_LABEL,
31
+ CANDIDATES,
32
+ )
33
+ from typing import Dict
34
+ import re
35
+ import random
36
+ import copy
37
+ import json
38
+
39
+
40
+ def extract_summary(dial: Dict, **kwargs):
41
+ """
42
+ `dial` is the full dialog.
43
+ """
44
+ return dial[SUMMARY]
45
+
46
+
47
+ def extract_turn_emotion(turn: Dict, sep: str, **kwargs):
48
+ if EMOTIONS not in turn:
49
+ return None
50
+ return sep.join(map(lambda x: x[EMOTION], turn[EMOTIONS]))
51
+
52
+
53
+ def extract_turn_emotion_wrapper(sep: str):
54
+ def extract_turn_emotion_func(turn: Dict, **kwargs):
55
+ return extract_turn_emotion(turn, sep)
56
+
57
+ return extract_turn_emotion_func
58
+
59
+
60
+ def extract_turn_utterance(turn: Dict, **kwargs):
61
+ return turn[UTTERANCE]
62
+
63
+
64
+ def extract_aspects(turn: Dict, ext_aspect_sep: str, int_aspect_sep: str):
65
+ if not turn[ASPECTS]:
66
+ return "None"
67
+
68
+ aspects = turn[ASPECTS]
69
+
70
+ tgt_seq = []
71
+ for aspect in aspects:
72
+ aspect_seq = []
73
+ if TARGET in aspect:
74
+ aspect_seq.append(aspect[TARGET][VALUE])
75
+ if CATEGORY in aspect:
76
+ aspect_seq.append(aspect[CATEGORY])
77
+
78
+ if OPINION in aspect:
79
+ aspect_seq.append(aspect[OPINION][VALUE])
80
+
81
+ if SENTIMENT in aspect:
82
+ aspect_seq.append(aspect[SENTIMENT])
83
+
84
+ tgt_seq.append(int_aspect_sep.join(aspect_seq))
85
+
86
+ return ext_aspect_sep.join(tgt_seq)
87
+
88
+
89
+ def extract_aspects_wrapper(ext_aspect_sep: str, int_aspect_sep: str):
90
+ def extract_aspects_func(turn: Dict, **kwargs):
91
+ return extract_aspects(turn, ext_aspect_sep, int_aspect_sep)
92
+
93
+ return extract_aspects_func
94
+
95
+
96
+ def rebuild_utterance_with_characters(turn: Dict, split):
97
+ if split == "train":
98
+ utterance = turn[UTTERANCE]
99
+ parts = []
100
+ pre = 0
101
+
102
+ for character in turn[CHARACTERS]:
103
+ parts.append(utterance[pre : character[START]])
104
+ parts.append(
105
+ f"[{utterance[character[START]: character[END]]} | {character[VALUE]}]"
106
+ )
107
+ pre = character[END]
108
+
109
+ parts.append(utterance[pre:])
110
+ return "".join(parts)
111
+
112
+ else:
113
+ tuples = []
114
+ for character in turn[CHARACTERS]:
115
+ tuples.append(f"{character[VALUE]}, {character[START]}, {character[END]}")
116
+
117
+ if not tuples:
118
+ return "None"
119
+ return " | ".join(tuples)
120
+
121
+
122
+ def extract_characters(example):
123
+ for turn_id, turn in enumerate(example[DIALOG]):
124
+ if CHARACTERS not in turn:
125
+ continue
126
+
127
+ for character in turn[CHARACTERS]:
128
+ yield turn_id, character[VALUE], (character[END],)
129
+
130
+
131
+ def extract_belief_state(
132
+ turn,
133
+ value_sep,
134
+ domain_sep,
135
+ slot_sep,
136
+ domain_prompt_op,
137
+ ontology=None,
138
+ do_train=True,
139
+ ):
140
+ domain_bs = dict()
141
+ bs = turn[BELIEF_STATE]
142
+
143
+ # spare_bs = {domain: {slot for slot in ontology[domain]} for domain in ontology}
144
+
145
+ for state in bs:
146
+ domain = state[DOMAIN]
147
+ if domain not in domain_bs:
148
+ domain_bs[domain] = dict()
149
+
150
+ if INFORMED_SLOT_VALUE_TABLE not in state:
151
+ continue
152
+
153
+ for svp in state[INFORMED_SLOT_VALUE_TABLE]:
154
+ slot = svp[SLOT]
155
+ values = svp[VALUES]
156
+ relation = svp[RELATION]
157
+
158
+ if slot not in domain_bs[domain]:
159
+ domain_bs[domain][slot] = {"relation": relation, "values": []}
160
+ domain_bs[domain][slot]["values"] += list(map(lambda x: x[VALUE], values))
161
+
162
+ # spare_bs[domain].remove(slot)
163
+
164
+ domain_bs_list = []
165
+ for domain in domain_bs:
166
+ svp_list = []
167
+ for slot in domain_bs[domain]:
168
+ val_str = value_sep.join(domain_bs[domain][slot]["values"])
169
+ svp_list.append(f"{slot} {domain_bs[domain][slot]['relation']} {val_str}")
170
+
171
+ # control whether to add spare slots
172
+ # for slot in sorted(spare_bs[domain]):
173
+ # svp_list.append(f"{slot} = None")
174
+ if not svp_list:
175
+ continue
176
+ if do_train:
177
+ # shuffle for training
178
+ random.shuffle(svp_list)
179
+
180
+ # append a slot separator at the end to alleviate the problem of end point prediction of T5
181
+ svt_str = slot_sep.join(svp_list) + slot_sep
182
+
183
+ domain_bs_list.append(f"{domain}{domain_prompt_op}{svt_str.strip()}")
184
+
185
+ if not domain_bs_list:
186
+ return "None"
187
+
188
+ return domain_sep.join(domain_bs_list)
189
+
190
+
191
+ def extract_belief_state_wrapper(value_sep, domain_sep, slot_sep, domain_prompt_op):
192
+ def extract_belief_state_func(turn, ontology, do_train=True, **kwargs):
193
+ return extract_belief_state(
194
+ turn,
195
+ value_sep,
196
+ domain_sep,
197
+ slot_sep,
198
+ domain_prompt_op,
199
+ ontology,
200
+ do_train=do_train,
201
+ )
202
+
203
+ return extract_belief_state_func
204
+
205
+
206
+ def normalize(query: str) -> str:
207
+ def comma_fix(s):
208
+ # Remove spaces in front of commas
209
+ return s.replace(" , ", ", ")
210
+
211
+ def white_space_fix(s):
212
+ # Remove double and triple spaces
213
+ return " ".join(s.split())
214
+
215
+ def lower(s):
216
+ # Convert everything except text between (single or double) quotation marks to lower case
217
+ return re.sub(
218
+ r"\b(?<!['\"])(\w+)(?!['\"])\b", lambda match: match.group(1).lower(), s
219
+ )
220
+
221
+ def space_fix(sql: str):
222
+ def agg_fix(sql: str):
223
+ return re.sub(
224
+ r"(count|max|min|sum|avg)\s\(",
225
+ lambda match: match.group(0).replace(" ", ""),
226
+ sql,
227
+ )
228
+
229
+ def brackets_fix(sql: str):
230
+ sql = re.sub(r"\(\s", lambda match: match.group(0)[:-1], sql)
231
+ sql = re.sub(r"\s\)", lambda match: match.group(0)[1:], sql)
232
+
233
+ return sql
234
+
235
+ def double_chars_op_fix(sql: str):
236
+ return re.sub(
237
+ r"((>|<|!)\s=)",
238
+ lambda match: match.group(0).replace(" ", ""),
239
+ sql,
240
+ )
241
+
242
+ return double_chars_op_fix(brackets_fix(agg_fix(sql)))
243
+
244
+ return space_fix(comma_fix(white_space_fix(lower(query))))
245
+
246
+
247
+ def extract_sql(turn, split):
248
+ if SQL not in turn:
249
+ return None
250
+ _normalize = normalize if split == "train" else (lambda x: x)
251
+ return _normalize(turn[SQL])
252
+
253
+
254
+ def extract_slots_without_intents(turn, value_sep, slot_sep):
255
+ if SLOTS_TO_FILL not in turn or not turn[SLOTS_TO_FILL][SLOT_VALUE_TABLE]:
256
+ return "None"
257
+ slots = []
258
+ for svp in turn[SLOTS_TO_FILL][SLOT_VALUE_TABLE]:
259
+ slots.append(
260
+ svp[SLOT]
261
+ + " "
262
+ + svp[RELATION]
263
+ + " "
264
+ + value_sep.join(map(lambda x: x[VALUE], svp[VALUES]))
265
+ )
266
+
267
+ return (slot_sep.join(slots) + slot_sep).strip()
268
+
269
+
270
+ def extract_slots_without_intents_wrapper(value_sep, slot_sep):
271
+ def extract_slots_without_intents_func(turn, **kwargs):
272
+ return extract_slots_without_intents(turn, value_sep, slot_sep)
273
+
274
+ return extract_slots_without_intents_func
275
+
276
+
277
+ def extract_role_relation_without_turn(dialog, relation_sep):
278
+ return relation_sep.join(map(lambda x: x[RELATION], dialog[ROLE_RELATIONS]))
279
+
280
+
281
+ def extract_role_relation_without_turn_wrapper(relation_sep):
282
+ def extract_role_relation_without_turn_func(dialog, **kwargs):
283
+ return extract_role_relation_without_turn(dialog, relation_sep)
284
+
285
+ return extract_role_relation_without_turn_func
286
+
287
+
288
+ def extrac_rewritten(turn, **kwargs):
289
+ if REWRITTEN not in turn:
290
+ return None
291
+ return turn[REWRITTEN]
292
+
293
+
294
+ def extract_options(turn, knowledge, split=None):
295
+ if ROLES_TO_SELECT not in turn:
296
+ return None
297
+ if split == TRAIN_SPLIT:
298
+ return knowledge[turn[ROLES_TO_SELECT][0]]
299
+ else:
300
+ return json.dumps(
301
+ {OPTION_LABEL: turn[ROLES_TO_SELECT][0], CANDIDATES: knowledge}
302
+ )
303
+
304
+
305
+ # def extract_roles_wrapper(role_sep):
306
+ # def extract_roles_func(turn, knowledge, split=None):
307
+ # return extract_options(turn, know)
308
+
309
+ # return extract_roles_func
310
+
311
+
312
+ def extract_intents(turn, intent_sep):
313
+ if not turn[ACTIVE_INTENTS]:
314
+ return "None"
315
+ return intent_sep.join(
316
+ map(lambda intent: intent.replace("_", " "), turn[ACTIVE_INTENTS])
317
+ )
318
+
319
+
320
+ def extract_intents_wrapper(intent_sep):
321
+ def extract_intents_func(turn, **kwargs):
322
+ return extract_intents(turn, intent_sep)
323
+
324
+ return extract_intents_func
src/modules/preprocess/preprocessor/process_turn_funcs.py ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from const import UTTERANCE, CHARACTERS, START, END, MENTION
2
+
3
+
4
+ def introduce_mention_to_utterance(turn, insert_index, left_bracket, right_bracket):
5
+ turn[UTTERANCE] = (
6
+ turn[UTTERANCE][:insert_index]
7
+ + left_bracket
8
+ + MENTION
9
+ + right_bracket
10
+ + turn[UTTERANCE][insert_index:]
11
+ )
12
+
13
+ return turn
14
+
15
+
16
+ def introduce_mention_to_utterance_wrapper(left_bracket, right_bracket):
17
+ def introduce_mention_to_utterance_func(turn, insert_index, **kwargs):
18
+ return introduce_mention_to_utterance(
19
+ turn, insert_index, left_bracket, right_bracket
20
+ )
21
+
22
+ return introduce_mention_to_utterance_func
src/modules/preprocess/preprocessor/prompt_funcs.py ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ def const_prompt_func_wrapper(const_prompt):
2
+ def const_prompt_func():
3
+ return const_prompt
4
+
5
+ return const_prompt_func
src/preprocess.sh ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/bin/bash
2
+
3
+ cd preprocess
4
+
5
+ INPUT=$1
6
+ OUTPUT=$2
7
+
8
+ DATA="MAMS-ACSA"
9
+ echo "--> ${DATA}"
10
+ # python ${DATA}.py --input_dir "${INPUT}/TaskMaster/TM-1-2019" --output_dir "${OUTPUT}/${DATA}"
11
+ python ${DATA}.py --input_dir "${INPUT}/MAMS/data/${DATA}/raw" --output_dir "${OUTPUT}/MAMS/${DATA}"
src/preprocess/ASTE.py ADDED
@@ -0,0 +1,97 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from utils import write_jsonl_file, parse, read_line_labels
2
+ import os
3
+
4
+ sent_map = {
5
+ "POS": "positive",
6
+ "NEU": "neutral",
7
+ "NEG": "negative",
8
+ }
9
+
10
+
11
+ def get_char_index(lengths, tok_ids):
12
+ start = lengths[tok_ids[0]] + tok_ids[0]
13
+ end = (
14
+ lengths[tok_ids[-1] + 1]
15
+ - lengths[tok_ids[0]]
16
+ + start
17
+ + tok_ids[-1]
18
+ - tok_ids[0]
19
+ )
20
+ return start, end
21
+
22
+
23
+ def parse_aspects(utterance, aspects):
24
+ toks = utterance.split()
25
+ lengths = list(map(lambda x: len(x), toks))
26
+ lengths = [0] + lengths
27
+
28
+ for i in range(1, len(lengths)):
29
+ lengths[i] += lengths[i - 1]
30
+
31
+ parsed_aspects = []
32
+ for target, opinion, sentiment in aspects:
33
+ target_start, target_end = get_char_index(lengths, target)
34
+ opinion_start, opinion_end = get_char_index(lengths, opinion)
35
+ target_value = " ".join(toks[target[0] : target[-1] + 1])
36
+ opinion_value = " ".join(toks[opinion[0] : opinion[-1] + 1])
37
+
38
+ assert target_value == utterance[target_start:target_end]
39
+ assert opinion_value == utterance[opinion_start:opinion_end]
40
+
41
+ parsed_aspects.append(
42
+ {
43
+ "target": {
44
+ "value": target_value,
45
+ "start": target_start,
46
+ "end": target_end,
47
+ },
48
+ "opinion": {
49
+ "value": opinion_value,
50
+ "start": opinion_start,
51
+ "end": opinion_end,
52
+ },
53
+ "sentiment": sent_map[sentiment],
54
+ }
55
+ )
56
+
57
+ return parsed_aspects
58
+
59
+
60
+ def reformat(args, file):
61
+ for domain in os.listdir(args.input_dir):
62
+ path = os.path.join(os.path.join(args.input_dir, domain), f"{file}.txt")
63
+ data = read_line_labels(path)
64
+
65
+ dials = []
66
+ for line in data:
67
+ utterance, aspects = line.strip().split("####")
68
+ aspects = eval(aspects)
69
+
70
+ dial = {
71
+ "turn": "single",
72
+ "locale": "en",
73
+ "dialog": [
74
+ {
75
+ "roles": ["USER"],
76
+ "utterance": utterance,
77
+ "aspects": parse_aspects(utterance, aspects),
78
+ }
79
+ ],
80
+ }
81
+
82
+ dials.append(dial)
83
+
84
+ write_jsonl_file(
85
+ dials, os.path.join(os.path.join(args.output_dir, domain), f"{file}.jsonl")
86
+ )
87
+
88
+
89
+ def preprocess(args):
90
+ reformat(args, "train")
91
+ reformat(args, "dev")
92
+ reformat(args, "test")
93
+
94
+
95
+ if __name__ == "__main__":
96
+ args = parse()
97
+ preprocess(args)
src/preprocess/AlphaNLI.py ADDED
@@ -0,0 +1,84 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from utils import read_jsonl_file, write_jsonl_file, parse, read_line_labels
2
+ import os
3
+ import copy
4
+
5
+ label2nl = {"1": "First", "2": "Second"}
6
+
7
+
8
+ def preprocess_for_train_and_dev(args, file):
9
+ data_path = os.path.join(args.input_dir, f"{file}.jsonl")
10
+ data = read_jsonl_file(data_path)
11
+
12
+ label_path = os.path.join(args.input_dir, f"{file}-labels.lst")
13
+ labels = read_line_labels(label_path)
14
+
15
+ turns = []
16
+ for idx, example in enumerate(data):
17
+ turn = {
18
+ "turn": "multi",
19
+ "locale": "en",
20
+ "dialog": [
21
+ {"roles": ["First observation"], "utterance": example["obs1"]},
22
+ {
23
+ "roles": ["Second observation"],
24
+ "utterance": example["obs2"],
25
+ "roles_to_select": [f"hypothesis candidate {labels[idx]}"],
26
+ },
27
+ ],
28
+ }
29
+
30
+ # turn["dialog"].append(
31
+ # {
32
+ # "roles": ["First hypothesis"],
33
+ # "utterance": example["hyp1"],
34
+ # }
35
+ # )
36
+
37
+ # turn["dialog"].append(
38
+ # {
39
+ # "roles": ["Second hypothesis"],
40
+ # "utterance": example["hyp2"],
41
+ # "roles_to_select": [label2nl[labels[idx]] + " hypothesis"],
42
+ # }
43
+ # )
44
+
45
+ turn["knowledge"] = {
46
+ "type": "text",
47
+ "value": {
48
+ "hypothesis candidate 1": example["hyp1"],
49
+ "hypothesis candidate 2": example["hyp2"],
50
+ },
51
+ }
52
+
53
+ # turn["roles_to_select"] = ["HYPOTHESIS " + labels[idx]]
54
+ turns.append(turn)
55
+
56
+ # if labels[idx] == "1":
57
+ # pos_hyp = example["hyp1"]
58
+ # neg_hyp = example["hyp2"]
59
+ # else:
60
+ # pos_hyp = example["hyp2"]
61
+ # neg_hyp = example["hyp1"]
62
+
63
+ # # possitive hypothesis
64
+ # pos_turn = copy.deepcopy(turn)
65
+ # pos_turn["dialog"].append({"roles": ["HYPOTHESIS"], "utterance": pos_hyp, "class_label": True})
66
+
67
+ # # negative hypothesis
68
+ # neg_turn = copy.deepcopy(turn)
69
+ # neg_turn["dialog"].append({"roles": ["HYPOTHESIS"], "utterance": neg_hyp, "class_label": False})
70
+
71
+ # turns.append(pos_turn)
72
+ # turns.append(neg_turn)
73
+
74
+ write_jsonl_file(turns, os.path.join(args.output_dir, f"{file}.jsonl"))
75
+
76
+
77
+ def preprocess(args):
78
+ preprocess_for_train_and_dev(args, "train")
79
+ preprocess_for_train_and_dev(args, "dev")
80
+
81
+
82
+ if __name__ == "__main__":
83
+ args = parse()
84
+ preprocess(args)
src/preprocess/Banking77.py ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from utils import write_jsonl_file, read_csv_file, parse
2
+
3
+
4
+ def reformat(args, file):
5
+ path = args.input_dir + "/" + file + ".csv"
6
+ data = read_csv_file(path)
7
+ turns = []
8
+ for i in range(len(data)):
9
+ t = {
10
+ "turn": "single",
11
+ "locale": "en",
12
+ "dialog": [
13
+ {
14
+ "roles": ["USER"],
15
+ "utterance": data["text"][i],
16
+ "active_intents": [data["category"][i]],
17
+ }
18
+ ],
19
+ }
20
+ turns.append(t)
21
+
22
+ write_jsonl_file(turns, args.output_dir + "/" + file + ".jsonl")
23
+
24
+
25
+ def preprocess(args):
26
+ reformat(args, "train")
27
+ reformat(args, "test")
28
+
29
+
30
+ if __name__ == "__main__":
31
+ args = parse()
32
+ preprocess(args)