-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathrun_bm25.py
More file actions
492 lines (431 loc) · 15 KB
/
run_bm25.py
File metadata and controls
492 lines (431 loc) · 15 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
import json
from typing import List
import tiktoken
from rank_bm25 import BM25Okapi
from raptor import RetrievalAugmentationConfig, RetrievalAugmentation, BaseSummarizationModel, BaseEmbeddingModel, BaseQAModel
import os
from structured_rag.utils import get_nondummy_ancestors
from structured_rag import SHTGenerator, SHTGeneratorConfig, get_context_len
from run_raptor import generate_context as raptor_generate_context
import logging
import logging_config
logging.disable(logging.INFO)
from datetime import datetime
def bm25_indexer(chunks: List[str], query: str) -> List[int]:
'''
Return a list of the indexes sorted by the decreasing order of bm25 score. The indexes are 0-based. The 0th chunk has index 0. The tokenizer is "cl100k_base".
Args:
- chunks (List[str])
- query (str)
Return:
- sorted list of the indexes
'''
tokenizer = tiktoken.get_encoding("cl100k_base")
tokenized_chunks = [tokenizer.encode(c) for c in chunks]
tokenized_query = tokenizer.encode(query)
bm25 = BM25Okapi(tokenized_chunks)
scores = bm25.get_scores(tokenized_query)
assert len(scores) == len(chunks)
sorted_indices = sorted(range(len(scores)), key=lambda index: (-scores[index], index))
return sorted_indices
def get_tree(
dataset,
name,
chunk_size,
summary_len,
summarization_model,
is_intrinsic,
is_baseline,
):
node_embedding_model = "sbert"
root_dir = os.path.join(os.path.dirname(os.path.abspath(__file__)), "data", dataset)
if is_intrinsic:
root_dir = os.path.join(root_dir, "intrinsic")
if is_baseline:
root_dir = os.path.join(root_dir, "baselines")
if not is_baseline:
tree_path = os.path.join(root_dir, f"{node_embedding_model}.{summarization_model}.c{chunk_size}.s{summary_len}", "sht", name+".json")
with open(tree_path, 'r') as file:
tree = json.load(file)
return tree
else:
tree_path = os.path.join(root_dir, "raptor_tree", name+".pkl")
return tree_path
def get_index_path(
dataset,
chunk_size,
summary_len,
summarization_model,
distance_metric,
embed_hierarchy,
is_intrinsic,
is_baseline,
is_raptor,
):
node_embedding_model = "bm25"
query_embedding_model = "bm25"
root_dir = os.path.join(os.path.dirname(os.path.abspath(__file__)), "data", dataset)
if is_intrinsic:
root_dir = os.path.join(root_dir, "intrinsic")
if is_baseline:
root_dir = os.path.join(root_dir, "baselines")
if not is_baseline:
index_path = os.path.join(root_dir, f"{node_embedding_model}.{summarization_model}.c{chunk_size}.s{summary_len}", f"{query_embedding_model}.{distance_metric}.h{int(embed_hierarchy)}", "index.jsonl")
else:
index_path = os.path.join(root_dir, f"{node_embedding_model}.{summarization_model}.c{chunk_size}.s{summary_len}", f"{query_embedding_model}.{distance_metric}.raptor{int(is_raptor)}", "index.jsonl")
os.makedirs(os.path.dirname(index_path), exist_ok=True)
return index_path
def get_context_path(
dataset,
chunk_size,
summary_len,
summarization_model,
distance_metric,
embed_hierarchy,
context_hierarchy,
context_raw,
context_len,
is_intrinsic,
is_baseline,
is_raptor,
is_ordered,
):
index_path = get_index_path(
dataset,
chunk_size,
summary_len,
summarization_model,
distance_metric,
embed_hierarchy,
is_intrinsic,
is_baseline,
is_raptor,
)
if not is_baseline:
context_path = os.path.join(os.path.dirname(index_path), f"l{int(context_raw)}.h{int(context_hierarchy)}", f"context{context_len}", "context.jsonl")
else:
context_path = os.path.join(os.path.dirname(index_path), f"o{int(is_ordered)}", f"context{context_len}", "context.jsonl")
os.makedirs(os.path.dirname(context_path), exist_ok=True)
return context_path
class CustomQAModel(BaseQAModel):
def __init__(self):
pass
def answer_question(self, context, question):
raise ValueError("should not achieve this part")
class CustomSummarizationModel(BaseSummarizationModel):
def __init__(self):
pass
def summarize(self, context, max_tokens=100, stop_sequence=None):
raise ValueError("should not achieve this part")
class CustomEmbeddingModel(BaseEmbeddingModel):
def __init__(self):
pass
def create_embedding(self, text):
raise ValueError("should not achieve this part")
def index_sht(
query,
query_id,
tree,
embed_hierarchy,
):
chunks = []
m_id_to_pos = dict()
for node in tree["nodes"]:
if node["is_dummy"]:
continue
node_id = node["id"]
ancestors = get_nondummy_ancestors(tree["nodes"], node_id)
if len(ancestors) > 0:
assert sorted(ancestors) == ancestors
assert all([(aid >= 0 and aid < len(tree["nodes"]) for aid in ancestors)])
assert all([not tree["nodes"][aid]["is_dummy"] for aid in ancestors])
assert all([tree["nodes"][aid]["type"] != "text" for aid in ancestors])
assert all([tree["nodes"][aid]["heading"] != "" for aid in ancestors])
assert sorted(ancestors) == ancestors
ancestor_string = "\n\n".join([tree["nodes"][aid]["heading"] for aid in ancestors])
if ancestor_string != "":
ancestor_string += "\n\n"
heading_string = node["heading"]
if heading_string != "":
heading_string += "\n\n"
if node["type"] == "text":
assert set(node["embeddings"].keys()) == set(["texts", "hybrid"])
assert node["heading"] == "" and heading_string == ""
else:
assert node["type"] in ["head", "list"]
assert set(node["embeddings"].keys()) == set(["texts", "hybrid", "heading"])
assert node["heading"] != ""
for chunk_id, text in enumerate(node["texts"]):
id = len(chunks)
m_id_to_pos[id] = {
"node_id": node["id"],
"chunk_id": chunk_id,
}
if embed_hierarchy:
chunks.append(ancestor_string + heading_string + text)
else:
chunks.append(heading_string + text)
sorted_indexes = bm25_indexer(chunks, query)
index_info = {
"id": query_id,
"indexes": [m_id_to_pos[i] for i in sorted_indexes]
}
return index_info
def index_raptor(
query,
query_id,
tree,
is_raptor
):
custom_summarizer = CustomSummarizationModel()
custom_embedder = CustomEmbeddingModel()
custom_qa = CustomQAModel()
custom_RAConfig = RetrievalAugmentationConfig(
summarization_model=custom_summarizer,
qa_model=custom_qa,
embedding_model=custom_embedder,
)
RA = RetrievalAugmentation(config=custom_RAConfig, tree=tree)
if not is_raptor:
nodes = RA.tree.leaf_nodes.values()
else:
nodes = RA.tree.all_nodes.values()
sorted_nodes = sorted(nodes, key=lambda v: v.index)
assert list(range(len(sorted_nodes))) == [v.index for v in sorted_nodes]
chunks = [v.text for v in sorted_nodes]
indexes = bm25_indexer(chunks, query)
assert len(indexes) == len(sorted_nodes)
assert set(indexes) == set(range(len(sorted_nodes)))
index_info = {
"id": query_id,
"indexes": indexes,
}
return index_info
def index(
dataset,
chunk_size,
summary_len,
summarization_model,
embed_hierarchy,
distance_metric,
context_hierarchy,
context_raw,
context_len,
is_intrinsic,
is_baseline,
is_raptor,
is_ordered,
):
print("Indexing")
index_path = get_index_path(
dataset,
chunk_size,
summary_len,
summarization_model,
distance_metric,
embed_hierarchy,
is_intrinsic,
is_baseline,
is_raptor,
)
os.makedirs(os.path.dirname(index_path), exist_ok=True)
queries_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "data", dataset, "queries.json")
with open(queries_path, 'r') as file:
queries = json.load(file)
existing_query_ids = list()
if os.path.exists(index_path):
with open(index_path, 'r') as file:
for l in file:
index_id = (json.loads(l))["id"]
assert index_id not in existing_query_ids, f"Duplicate query id: {index_id}"
existing_query_ids.append(index_id)
for query_info in queries:
query = query_info["query"]
query_id = query_info["id"]
name = query_info["file_name"]
if query_id in existing_query_ids:
cur_time = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
print(f"\t[{cur_time}]⚠️ Skipping {query_info['id']}")
continue
cur_time = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
print(f"[{cur_time}] Processing {query_info['id']}...")
try:
tree = get_tree(
dataset,
name,
chunk_size,
summary_len,
summarization_model,
is_intrinsic,
is_baseline,
)
if not is_baseline:
index_info = index_sht(query, query_id, tree, embed_hierarchy)
else:
index_info = index_raptor(query, query_id, tree, is_raptor)
with open(index_path, 'a') as file:
file.write(json.dumps(index_info) + "\n")
existing_query_ids.append(query_id)
cur_time = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
print(f"\t[{cur_time}]✅ Finished processing {query_info['id']} with is_baseline={is_baseline} is_raptor={is_raptor}")
except Exception as e:
cur_time = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
print(f"\t[{cur_time}]✅ Finished processing {query_info['id']} with is_baseline={is_baseline} is_raptor={is_raptor}")
continue
def generate_context(
dataset,
chunk_size,
summary_len,
summarization_model,
embed_hierarchy,
distance_metric,
context_hierarchy,
context_raw,
context_len,
is_intrinsic,
is_baseline,
is_raptor,
is_ordered,
):
context_path = get_context_path(
dataset,
chunk_size,
summary_len,
summarization_model,
distance_metric,
embed_hierarchy,
context_hierarchy,
context_raw,
context_len,
is_intrinsic,
is_baseline,
is_raptor,
is_ordered,
)
# if os.path.exists(context_path):
# logging.debug(f"Context file already existed: {context_path}!")
# return
existing_query_ids = list()
if os.path.exists(context_path):
with open(context_path, 'r') as file:
for l in file:
context_info = json.loads(l)
query_id = context_info["id"]
assert query_id not in existing_query_ids
existing_query_ids.append(context_info["id"])
if not is_baseline:
queries_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "data", dataset, "queries.json")
with open(queries_path, 'r') as file:
queries = json.load(file)
index_path = get_index_path(
dataset,
chunk_size,
summary_len,
summarization_model,
distance_metric,
embed_hierarchy,
is_intrinsic,
is_baseline,
is_raptor,
)
indexes = []
with open(index_path, 'r') as file:
for l in file:
indexes.append(json.loads(l))
assert len(indexes) == len(queries)
for index_info, query_info in zip(indexes, queries):
assert index_info["id"] == query_info["id"]
if query_info["id"] in existing_query_ids:
print(f"\tquery_id {query_id} already has context, skip!")
continue
query = query_info["query"]
query_id = query_info["id"]
name = query_info["file_name"]
true_context_len = get_context_len(
context_ratio=context_len,
dataset=dataset,
sht_json_filename=name,
min_context_len=round(max(chunk_size, summary_len) * 1.5)
)
generator_config = SHTGeneratorConfig(
use_hierarchy=context_hierarchy,
use_raw_chunks=context_raw,
context_len=true_context_len
)
generator = SHTGenerator(config=generator_config)
sht = get_tree(
dataset,
name,
chunk_size,
summary_len,
summarization_model,
is_intrinsic,
is_baseline,
)
context = generator.generate(
candid_indexes=index_info["indexes"],
nodes=sht["nodes"]
)
context_info = {
"id": query_id,
"context": context
}
assert query_id not in existing_query_ids
existing_query_ids.append(query_id)
with open(context_path, 'a') as file:
file.write(json.dumps(context_info) + "\n")
else:
raptor_generate_context(
dataset=dataset,
query_embedding_model="bm25",
is_ordered=is_ordered,
is_raptor=is_raptor,
context_len=context_len,
)
if __name__ == "__main__":
chunk_size = 100
summary_len = 100
summarization_model = "gpt-4o-mini"
embed_hierarchy = True
distance_metric = "cosine"
context_hierarchy = True
context_raw = True
context_len = 1000
is_intrinsic = False
for is_baseline in [True, False]:
for is_raptor in [True, False]:
print(is_baseline, is_raptor)
if (is_baseline == False and is_raptor == False):
continue
dataset = "finance"
index(
dataset,
chunk_size,
summary_len,
summarization_model,
embed_hierarchy,
distance_metric,
context_hierarchy,
context_raw,
context_len,
is_intrinsic,
is_baseline,
is_raptor,
is_ordered=None,
)
# for is_ordered in [False, True]:
# generate_context(
# dataset,
# chunk_size,
# summary_len,
# summarization_model,
# embed_hierarchy,
# distance_metric,
# context_hierarchy,
# context_raw,
# context_len,
# is_intrinsic,
# is_baseline,
# is_raptor,
# is_ordered,
# )