-
Notifications
You must be signed in to change notification settings - Fork 122
/
fdedup_support.py
621 lines (562 loc) · 23 KB
/
fdedup_support.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
# (C) Copyright IBM Corp. 2024.
# Licensed under the Apache License, Version 2.0 (the “License”);
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
# http://www.apache.org/licenses/LICENSE-2.0
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an “AS IS” BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
################################################################################
import pickle
import time
from typing import Any, Iterator, Union
import numpy as np
import ray
from data_processing.data_access import SnapshotUtils
from data_processing.utils import GB, RANDOM_SEED, TransformUtils, get_logger
from data_processing_ray.runtime.ray import RayUtils
from ray.actor import ActorHandle
from ray.util import ActorPool
from scipy.integrate import quad as integrate
NO_SIMILARITY = -1
REQUEST_LEN = 4096
LONG_BUCKET = 5000
LONG_BUCKET_PRINT = 1000
def fuzzy_optimal_param(
threshold: float,
num_perm: int,
false_positive_weight: float,
false_negative_weight: float,
) -> tuple[int, int]:
"""
Computes parameters for fuzzy dedup
:param threshold: filtering threshold
:param num_perm: number of permutations
:param false_positive_weight: false positive weight
:param false_negative_weight: false negative weight
:return: number of buckets and bucket length
"""
def _false_positive_probability(ths: float, b: int, r: int) -> float:
"""
Compute false positive probability
:param ths: filtering threshold
:param b: permutation
:param r: rel permutation
:return: probability
"""
_probability = lambda s: 1 - (1 - s ** float(r)) ** float(b)
a, err = integrate(_probability, 0.0, ths)
return a
def _false_negative_probability(ths: float, b: int, r: int) -> float:
"""
Compute false negative probability
:param ths: filtering threshold
:param b: permutation
:param r: rel permutation
:return: probability
"""
_probability = lambda s: 1 - (1 - (1 - s ** float(r)) ** float(b))
a, err = integrate(_probability, ths, 1.0)
return a
min_error = float("inf")
opt = (0, 0)
for perm in range(1, num_perm + 1):
max_r = int(num_perm / perm)
for rel in range(1, max_r + 1):
fp = _false_positive_probability(threshold, perm, rel)
fn = _false_negative_probability(threshold, perm, rel)
error = fp * false_positive_weight + fn * false_negative_weight
if error < min_error:
min_error = error
opt = (perm, rel)
return opt
class MurmurMH:
def __init__(self, num_perm: int, seed: int = RANDOM_SEED):
self.seed = seed
self.num_perm = num_perm
self.permutations = self._init_permutations(seed, num_perm)
def minhash(self, shingle_count: int, shingles: Iterator[str]) -> np.array:
def generator():
for shingle in shingles:
yield TransformUtils.str_to_int(shingle)
hash_values = np.fromiter(generator(), dtype=np.uint64, count=shingle_count)
result = np.zeros(self.permutations.shape, dtype=np.uint32)
for i, perm in enumerate(self.permutations):
result[i] = np.right_shift((perm * hash_values).T, 32).astype(np.uint32).min(axis=0, keepdims=False)
return result
@staticmethod
def _init_permutations(seed: int, num_perm: int) -> np.array:
# see https://en.wikipedia.org/wiki/Universal_hashing#Avoiding_modular_arithmetic
max_int = np.uint64((1 << 64) - 1)
gen = np.random.RandomState(seed)
# get self.num_perm pseudo random numbers between 2 and max_int (excl)
permutations = np.array([gen.randint(0, max_int, dtype=np.uint64) for _ in range(num_perm)], dtype=np.uint64).T
# make all even pseudo random numbers odd by adding 1
permutations[permutations % 2 == 0] += 1
return permutations
@staticmethod
def jaccard(mh1: np.array, mh2: np.array) -> float:
return np.count_nonzero(mh1 == mh2)
@ray.remote(scheduling_strategy="SPREAD")
class DocCollector:
"""
An actor collecting de duped document IDs
"""
def __init__(self, params: dict[str, Any]):
"""
Initializer
"""
self.logger = get_logger(__name__)
self.actor_id = params.get("id")
self.removed = set()
data_access_factory = params.get("data_access")
self.data_access = data_access_factory.create_data_access()
snapshot = params.get("snapshot", None)
if snapshot is None:
self.ids = {}
else:
try:
bids, _ = self.data_access.get_file(snapshot)
self.ids = pickle.loads(bids)
except Exception as e:
self.logger.warning(f"Failed to load doc collector {self.actor_id} with exception {e}")
raise e
def add_documents(self, dr: tuple[list[tuple[int, int]], list[int]]) -> None:
"""
Add documents and removed document
:param dr: documents to keep and documents to remove
:return:
"""
docs = dr[0]
rm = dr[1]
# process documents to remove
for did in rm:
self.ids.pop(did, None)
self.removed.update(rm)
# process documents to keep
for key, val in docs:
if key in self.removed:
continue
if key in self.ids and val == NO_SIMILARITY:
# Do not update existing docs with NO_SIMILARITY
continue
else:
self.ids[key] = val
def filter(self, docs: list[int]) -> dict[int, int]:
"""
Filter documents
:param docs: documents to filter
:return: documents to keep
"""
result = {}
for doc_id in docs:
r = self.ids.get(doc_id, None)
if r is not None:
result[doc_id] = r
return result
def snapshot(self) -> None:
"""
Snapshotting itself
"""
try:
b_doc = pickle.dumps(self.ids)
self.data_access.save_file(
f"{SnapshotUtils.get_snapshot_folder(self.data_access)}docs/doc_collector_{self.actor_id}", b_doc
)
except Exception as e:
self.logger.warning(f"Failed to snapshot doc collector {self.actor_id} with exception {e}")
raise e
def get_size(self) -> tuple[int, float, int, float]:
"""
get sizes
:return: number of ids, its memory utilization, number of removed, its memory utilization
"""
return (
len(self.ids),
TransformUtils.deep_get_size(self.ids) / GB,
len(self.removed),
TransformUtils.deep_get_size(self.removed) / GB,
)
@ray.remote(scheduling_strategy="SPREAD")
class DocsMinHash:
"""
An actor storing min hashes for a doc id
"""
def __init__(self, params: dict[str, Any]):
"""
Initialize
:param params: parameters
"""
self.logger = get_logger(__name__)
self.actor_id = params.get("id")
data_access_factory = params.get("data_access")
self.data_access = data_access_factory.create_data_access()
snapshot = params.get("snapshot", None)
if snapshot is None:
self.docs = {}
else:
try:
bdocs, _ = self.data_access.get_file(snapshot)
self.docs = pickle.loads(bdocs)
except Exception as e:
self.logger.warning(f"Failed to load minhash collector {self.actor_id} with exception {e}")
raise e
def add_minhashes(self, updates: list[tuple[int, int, np.array]]) -> None:
"""
Add minhashes
:param updates: minhash for doc_id a tuple of doc len and array of hashes
:return: None
"""
for doc_id, length, minhash in updates:
self.docs[doc_id] = np.concatenate(([length], minhash))
def get_minhashes(self, doc_ids: list[int]) -> list[tuple[int, int, np.array]]:
"""
Get minhashes for a list of documents
:param doc_ids: list of doc ids
:return: doc id, len, minhashes
"""
result = []
for doc_id in doc_ids:
info = self.docs.get(doc_id)
if info is not None:
result.append((doc_id, info[0], info[1:]))
return result
def snapshot(self) -> None:
"""
Snapshotting itself
"""
try:
b_doc = pickle.dumps(self.docs)
self.data_access.save_file(
f"{SnapshotUtils.get_snapshot_folder(self.data_access)}minhash/minhash_collector_{self.actor_id}",
b_doc,
)
except Exception as e:
self.logger.warning(f"Failed to snapshot minhash collector {self.actor_id} with exception {e}")
raise e
def get_size(self) -> tuple[int, float]:
"""
Get size of used min hashes
:return: number of docs, its memory utilization
"""
return len(self.docs), TransformUtils.deep_get_size(self.docs) / GB
@ray.remote(scheduling_strategy="SPREAD")
class BucketsHash:
"""
Actor storing buckets information
"""
def __init__(self, params: dict[str, Any]):
"""
Initialization
"""
from ray.util.metrics import Counter
self.submitter = None
self.n_buckets = 0
self.bucket_memory = 0
self.logger = get_logger(__name__)
self.actor_id = params.get("id")
data_access_factory = params.get("data_access")
self.data_access = data_access_factory.create_data_access()
snapshot = params.get("snapshot", None)
if snapshot is None:
self.buckets = {}
else:
try:
b_buckets, _ = self.data_access.get_file(snapshot)
self.buckets = pickle.loads(b_buckets)
except Exception as e:
self.logger.warning(f"Failed to load buckets collector {self.actor_id} with exception {e}")
raise e
self.bucket_created_counter = Counter("bucket_created", "Amount of buckets created")
self.long_bucket_submit_counter = Counter("long_bucket_submitted", "Amount of long buckets submitted")
self.short_bucket_submit_counter = Counter("short_bucket_submitted", "Amount of short buckets submitted")
def add_buckets(self, bck: list[tuple[int, list[int]]]) -> None:
"""
Add additional buckets to hash
:param bck: bucket information
:return: None
"""
for bucket in bck:
b_hash = bucket[0]
buckets_for_hash = self.buckets.get(b_hash)
if buckets_for_hash:
if type(buckets_for_hash) == int:
self.buckets[b_hash] = [buckets_for_hash] + bucket[1]
else:
buckets_for_hash.extend(bucket[1])
else:
if len(bucket[1]) == 1:
self.buckets[b_hash] = bucket[1][0]
else:
self.buckets[b_hash] = bucket[1]
self.bucket_created_counter.inc(1)
def add_processing_submitter(self, submitter: ActorHandle) -> None:
"""
Add process submitter
:param submitter: reference to submitter
:return:
"""
self.submitter = submitter
def process_buckets(self) -> None:
"""
Process buckets to generate documents
:return: None
"""
# Remember usage
self.n_buckets = len(self.buckets)
self.bucket_memory = TransformUtils.deep_get_size(self.buckets) / GB
# split buckets into short and long. Long buckets can take very long to process
long_buckets = []
short_buckets = []
while len(self.buckets) > 0:
doc_id, bucket = self.buckets.popitem()
if type(bucket) == list and len(bucket) > LONG_BUCKET:
# Its long
long_buckets.append(bucket)
else:
short_buckets.append(bucket)
self.logger.info(f"processing buckets {len(long_buckets)} long, {len(short_buckets)} short")
# process long buckets first - we are submitting them one at a time
for bucket in long_buckets:
if len(bucket) > 2 * LONG_BUCKET:
# For very long buckets, split them
self.logger.info(f"Splitting bucket of length len(bucket) into chunks")
smaller_bucket = [
bucket[i * LONG_BUCKET : (i + 1) * LONG_BUCKET]
for i in range((len(bucket) + LONG_BUCKET - 1) // LONG_BUCKET)
]
for b in smaller_bucket:
ray.get(self.submitter.submit_for_processing.remote([b]))
self.long_bucket_submit_counter.inc(1)
else:
ray.get(self.submitter.submit_for_processing.remote([bucket]))
self.long_bucket_submit_counter.inc(1)
self.logger.info("Done submitting long buckets")
# And now the rest of buckets
bucket_chunks = [short_buckets[i * 100 : (i + 1) * 100] for i in range((len(short_buckets) + 99) // 100)]
for b in bucket_chunks:
ray.get(self.submitter.submit_for_processing.remote(b))
self.short_bucket_submit_counter.inc(len(b))
def snapshot(self) -> None:
"""
Snapshotting itself
"""
try:
b_buckets = pickle.dumps(self.buckets)
self.data_access.save_file(
f"{SnapshotUtils.get_snapshot_folder(self.data_access)}buckets/buckets_collector_{self.actor_id}",
b_buckets,
)
except Exception as e:
self.logger.warning(f"Failed to snapshot buckets collector {self.actor_id} with exception {e}")
raise e
def get_size(self) -> tuple[int, float]:
"""
Get buckets resource utilization
:return: number of buckets and memory utilization
"""
return self.n_buckets, self.bucket_memory
@ray.remote(scheduling_strategy="SPREAD")
class BucketsHashProcessor:
"""
Actor for processing buckets
"""
def __init__(self, params: dict[str, Any]):
"""
Init method
:param params - dictionary of parameters containing the following keys
remote_docs - handles to the remote docs
remote_minhashes - handles to the remote minhashes
mn_min_hash - MurmurMH class
threshold - threshold
statistics - statistics actor
"""
from ray.util.metrics import Counter
self.threshold = params["threshold"]
self.mn_min_hash = params["mn_min_hash"]
self.remote_docs = params["remote_docs"]
self.remote_minhashes = params["remote_minhashes"]
self.stats = params["statistics"]
self.logger = get_logger(__name__)
self.bucket_processed_counter = Counter("bucket_processed", "Amount of buckets processed")
def _submit_generated_docs(self, docs: dict[int, int], removed: set[int]) -> None:
"""
Submit generated documents
:param docs: docs to submit
:param removed: removed documents
:return: None
"""
# Remove doc ids that are already removed
for did in removed:
docs.pop(did, None)
# Build remote requests
request = [([], []) for _ in range(len(self.remote_docs))]
for key, value in docs.items():
req_tuple = request[key % len(self.remote_docs)]
req_tuple[0].append((key, value))
for did in removed:
req_tuple = request[did % len(self.remote_docs)]
req_tuple[1].append(did)
# Submit requests and wait for replies
remote_replies = []
i = 0
for req in request:
if len(req[0]) > 0 or len(req[1]) > 0: # Only submit if the request has data
remote_replies.append(self.remote_docs[i].add_documents.remote(req))
i += 1
# Process replies
RayUtils.wait_for_execution_completion(logger=self.logger, replies=remote_replies)
# get minhashes and length for docs in the bucket
def _get_minhashes_docs(self, doc_ids: list[int]) -> dict[int, tuple[int, list[int]]]:
"""
Get minhashes for documents by submitting requests to an appropriate doc collectors
:param doc_ids: doc ids
:return: doc ids with hashes
"""
request = [[] for _ in range(len(self.remote_minhashes))]
for value in doc_ids:
request[value % len(self.remote_minhashes)].append(value)
remote_replies = []
i = 0
for req in request:
if len(req) > 0: # Only submit if the length is greater then 0
remote_replies.append(self.remote_minhashes[i].get_minhashes.remote(req))
i += 1
# Process replies
hashes = {}
while remote_replies:
# Wait for replies
ready, not_ready = ray.wait(remote_replies)
reply = ray.get(ready)[0]
for r in reply:
hashes[r[0]] = (r[1], r[2])
remote_replies = not_ready
return hashes
def process_buckets(self, buckets: list[Union[int, list[int]]]) -> None:
"""
process buckets to generate documents
:param buckets: buckets
:return: none
"""
t_start = time.time()
docs = {}
removed = set()
for bucket in buckets:
if type(bucket) == int:
# This hash has a single document
if bucket not in docs:
docs[bucket] = NO_SIMILARITY
self.bucket_processed_counter.inc(1)
continue
# multiple documents
start = time.time()
bucket_len = len(bucket)
very_long = bucket_len > LONG_BUCKET
hashes = self._get_minhashes_docs(bucket)
set_list = []
unvisited = set(bucket)
# combine similar documents
index = 0
while len(unvisited) > 0:
current_doc_id = unvisited.pop()
current_mh = hashes[current_doc_id][1]
current_set = set()
for other_doc_id in bucket:
if other_doc_id in unvisited:
other_mh = hashes[other_doc_id][1]
if self.mn_min_hash.jaccard(current_mh, other_mh) >= self.threshold:
current_set.add(current_doc_id)
current_set.add(other_doc_id)
unvisited.discard(other_doc_id)
if len(current_set) > 0:
set_list.append(current_set)
index += 1
if index % LONG_BUCKET_PRINT == 0:
self.logger.info(f"processing very long {bucket_len} bucket, {index} documents so far")
if index > LONG_BUCKET_PRINT:
self.logger.info(f"done processing very long {bucket_len}")
# process created sets
for current_set in set_list:
for d in current_set:
bucket.remove(d)
removed.update(current_set)
for i, doc_id in enumerate(current_set):
if i == 0:
cluster_id = doc_id
remaining = doc_id
min_len = hashes[doc_id][0]
max_len = min_len
continue
c_len = hashes[doc_id][0]
if c_len > max_len:
max_len = c_len
remaining = doc_id
continue
if c_len <= min_len:
min_len = c_len
cluster_id = doc_id
docs[remaining] = cluster_id
removed.discard(remaining)
# if we did not find docs in connections, submit them as NO_SIMILARITY
for d in bucket:
if d not in docs:
docs[d] = NO_SIMILARITY
if very_long:
self.logger.info(
f"Processed long ({bucket_len}) bucket in {round((time.time() - start) / 60.,3)} "
f"min; "
f"docs chains {len(set_list)}"
)
self.bucket_processed_counter.inc(1)
# Submit docs
self._submit_generated_docs(docs, removed)
# peg stats
self.stats.add_stats.remote({"generated doc_ids": len(docs), "bucket processing time": time.time() - t_start})
@ray.remote(scheduling_strategy="SPREAD")
class BucketsHashProcessorInvoker(object):
"""
Bucket hash processing coordinator (singleton)
"""
def __init__(self, bucket_processors: list[ActorHandle]) -> None:
self.n_processors = len(bucket_processors)
self.pool = ActorPool(bucket_processors)
self.submitted = 0
self.processed = 0
self.logger = get_logger(__name__)
self.start = time.time()
def submit_for_processing(self, buckets: list[Union[int, list[int]]]) -> None:
# Get completed results
if self.submitted < self.n_processors: # still have room
self.pool.submit(lambda a, v: a.process_buckets.remote(v), buckets)
self.logger.debug("Submitted bucket processing request")
self.submitted += 1
return
else:
while True:
# we can have several workers fail here
try:
self.pool.get_next_unordered()
break
except Exception as e:
self.logger.error(f"Failed to process request worker exception {e}")
self.processed += 1
self.processed += 1
if self.processed % 100 == 0:
self.logger.info(f"processed {self.processed} buckets in {(time.time() - self.start)/60} min")
self.logger.debug("Completed bucket processing request")
self.pool.submit(lambda a, v: a.process_buckets.remote(v), buckets)
self.submitted += 1
self.logger.debug("Submitted bucket processing request")
return
def wait_for_completion(self) -> None:
self.logger.info(f"Waiting bucket processing completion. Submitted requests {self.submitted}")
while self.pool.has_next():
try:
self.pool.get_next_unordered()
except Exception as e:
self.logger.error(f"Failed to process request worker exception {e}")
self.processed += 1
if self.processed % 100 == 0:
self.logger.info(f"processed {self.processed} buckets in {(time.time() - self.start)/60} min")