File size: 19,298 Bytes
0797029 | 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 | import numpy as np
np.random.seed(42)
import random
random.seed(42)
import pandas as pd
import json
from pathlib import Path
import glob
import gzip
import pickle
from copy import deepcopy
import torch
import torch.nn.functional as F
from transformers import AutoTokenizer, AutoConfig
import nlpaug.augmenter.word as naw
import nlpaug.augmenter.char as nac
from sklearn.preprocessing import LabelEncoder
from pdb import set_trace
def assign_clusterid(identifier, cluster_id_dict, cluster_id_amount):
try:
result = cluster_id_dict[identifier]
except KeyError:
result = cluster_id_amount
return result
# Methods for serializing examples by dataset
def serialize_sample_lspc(sample):
string = ''
string = f'{string}[COL] brand [VAL] {" ".join(sample[f"brand"].split(" ")[:5])}'.strip()
string = f'{string} [COL] title [VAL] {" ".join(sample[f"title"].split(" ")[:50])}'.strip()
string = f'{string} [COL] description [VAL] {" ".join(sample[f"description"].split(" ")[:100])}'.strip()
string = f'{string} [COL] specTableContent [VAL] {" ".join(sample[f"specTableContent"].split(" ")[:200])}'.strip()
return string
def serialize_sample_abtbuy(sample):
string = ''
string = f'{string}[COL] brand [VAL] {" ".join(sample[f"brand"].split())}'.strip()
string = f'{string} [COL] title [VAL] {" ".join(sample[f"name"].split())}'.strip()
string = f'{string} [COL] price [VAL] {" ".join(str(sample[f"price"]).split())}'.strip()
string = f'{string} [COL] description [VAL] {" ".join(sample[f"description"].split()[:100])}'.strip()
return string
def serialize_sample_amazongoogle(sample):
string = ''
string = f'{string}[COL] brand [VAL] {" ".join(sample[f"manufacturer"].split())}'.strip()
string = f'{string} [COL] title [VAL] {" ".join(sample[f"title"].split())}'.strip()
string = f'{string} [COL] price [VAL] {" ".join(str(sample[f"price"]).split())}'.strip()
string = f'{string} [COL] description [VAL] {" ".join(sample[f"description"].split()[:100])}'.strip()
return string
# Class for Data Augmentation
class Augmenter():
def __init__(self, aug):
stopwords = ['[COL]', '[VAL]', 'title', 'name', 'description', 'manufacturer', 'brand', 'specTableContent']
aug_typo = nac.KeyboardAug(stopwords=stopwords, aug_char_p=0.1, aug_word_p=0.1)
aug_swap = naw.RandomWordAug(action="swap", stopwords=stopwords, aug_p=0.1)
aug_del = naw.RandomWordAug(action="delete", stopwords=stopwords, aug_p=0.1)
aug_crop = naw.RandomWordAug(action="crop", stopwords=stopwords, aug_p=0.1)
aug_sub = naw.RandomWordAug(action="substitute", stopwords=stopwords, aug_p=0.1)
aug_split = naw.SplitAug(stopwords=stopwords, aug_p=0.1)
aug = aug.strip('-')
if aug == 'all':
self.augs = [aug_typo, aug_swap, aug_split, aug_sub, aug_del, aug_crop, None]
if aug == 'typo':
self.augs = [aug_typo, None]
if aug == 'swap':
self.augs = [aug_swap, None]
if aug == 'delete':
self.augs = [aug_del, None]
if aug == 'crop':
self.augs = [aug_crop, None]
if aug == 'substitute':
self.augs = [aug_sub, None]
if aug == 'split':
self.augs = [aug_split, None]
def apply_aug(self, string):
aug = random.choice(self.augs)
if aug is None:
return string
else:
return aug.augment(string)
# Dataset class for general Contrastive Pre-training for WDC computers
class ContrastivePretrainDataset(torch.utils.data.Dataset):
def __init__(self, path, deduction_set, tokenizer='huawei-noah/TinyBERT_General_4L_312D', max_length=128, intermediate_set=None, clean=False, dataset='lspc', only_interm=False, aug=False):
self.max_length = max_length
self.tokenizer = AutoTokenizer.from_pretrained(tokenizer, additional_special_tokens=('[COL]', '[VAL]'))
self.dataset = dataset
self.aug = aug
if self.aug:
self.augmenter = Augmenter(self.aug)
data = pd.read_pickle(path)
if dataset == 'abt-buy':
data['brand'] = ''
if dataset == 'amazon-google':
data['description'] = ''
if intermediate_set is not None:
interm_data = pd.read_pickle(intermediate_set)
if only_interm:
data = interm_data
else:
data = data.append(interm_data)
data = data.reset_index(drop=True)
data = data.fillna('')
data = self._prepare_data(data)
self.data = data
def __getitem__(self, idx):
# for every example in batch, sample one positive from the dataset
example = self.data.loc[idx].copy()
selection = self.data[self.data['labels'] == example['labels']]
# if len(selection) > 1:
# selection = selection.drop(idx)
pos = selection.sample(1).iloc[0].copy()
# apply augmentation if set
if self.aug:
example['features'] = self.augmenter.apply_aug(example['features'])
pos['features'] = self.augmenter.apply_aug(pos['features'])
return (example, pos)
def __len__(self):
return len(self.data)
def _prepare_data(self, data):
if self.dataset == 'lspc':
data['features'] = data.apply(serialize_sample_lspc, axis=1)
elif self.dataset == 'abt-buy':
data['features'] = data.apply(serialize_sample_abtbuy, axis=1)
elif self.dataset == 'amazon-google':
data['features'] = data.apply(serialize_sample_amazongoogle, axis=1)
label_enc = LabelEncoder()
data['labels'] = label_enc.fit_transform(data['cluster_id'])
self.label_encoder = label_enc
data = data[['features', 'labels']]
return data
# Dataset class for Contrastive Pre-training for Abt-Buy and Amazon-Google
# builds correspondence graph from train+val and builds source-aware sampling datasets
# if split=False, corresponds to not using source-aware sampling
class ContrastivePretrainDatasetDeepmatcher(torch.utils.data.Dataset):
def __init__(self, path, deduction_set, tokenizer='huawei-noah/TinyBERT_General_4L_312D', max_length=128, intermediate_set=None, clean=False, dataset='abt-buy', aug=False, split=True):
self.max_length = max_length
self.tokenizer = AutoTokenizer.from_pretrained(tokenizer, additional_special_tokens=('[COL]', '[VAL]'))
self.dataset = dataset
self.aug = aug
if self.aug:
self.augmenter = Augmenter(self.aug)
data = pd.read_pickle(path)
if dataset == 'abt-buy':
data['brand'] = ''
if dataset == 'amazon-google':
data['description'] = ''
if clean:
train_data = pd.read_json(deduction_set, lines=True)
if dataset == 'abt-buy':
val = pd.read_csv('../../data/interim/abt-buy/abt-buy-valid.csv')
elif dataset == 'amazon-google':
val = pd.read_csv('../../data/interim/amazon-google/amazon-google-valid.csv')
# use 80% of train and val set positives to build correspondence graph
val_set = train_data[train_data['pair_id'].isin(val['pair_id'])]
val_set_pos = val_set[val_set['label'] == 1]
val_set_pos = val_set_pos.sample(frac=0.80)
val_ids = set()
val_ids.update(val_set['pair_id'])
train_data = train_data[~train_data['pair_id'].isin(val_ids)]
train_data = train_data[train_data['label'] == 1]
train_data = train_data.sample(frac=0.80)
train_data = train_data.append(val_set_pos)
# build the connected components by applying binning
bucket_list = []
for i, row in train_data.iterrows():
left = f'{row["id_left"]}'
right = f'{row["id_right"]}'
found = False
for bucket in bucket_list:
if left in bucket and row['label'] == 1:
bucket.add(right)
found = True
break
elif right in bucket and row['label'] == 1:
bucket.add(left)
found = True
break
if not found:
bucket_list.append(set([left, right]))
cluster_id_amount = len(bucket_list)
#assign labels to connected components and single nodes (at this point single nodes have same label)
cluster_id_dict = {}
for i, id_set in enumerate(bucket_list):
for v in id_set:
cluster_id_dict[v] = i
data = data.set_index('id', drop=False)
data['cluster_id'] = data['id'].apply(assign_clusterid, args=(cluster_id_dict, cluster_id_amount))
#data = data[data['cluster_id'] != cluster_id_amount]
single_entities = data[data['cluster_id'] == cluster_id_amount].copy()
index = single_entities.index
if dataset == 'abt-buy':
left_index = [x for x in index if 'abt' in x]
right_index = [x for x in index if 'buy' in x]
elif dataset == 'amazon-google':
left_index = [x for x in index if 'amazon' in x]
right_index = [x for x in index if 'google' in x]
# assing increasing integer label to single nodes
single_entities = single_entities.reset_index(drop=True)
single_entities['cluster_id'] = single_entities['cluster_id'] + single_entities.index
single_entities = single_entities.set_index('id', drop=False)
single_entities_left = single_entities.loc[left_index]
single_entities_right = single_entities.loc[right_index]
# source aware sampling, build one sample per dataset
if split:
data1 = data.copy().drop(single_entities['id'])
data1 = data1.append(single_entities_left)
data2 = data.copy().drop(single_entities['id'])
data2 = data2.append(single_entities_right)
else:
data1 = data.copy().drop(single_entities['id'])
data1 = data1.append(single_entities_left)
data1 = data1.append(single_entities_right)
data2 = data.copy().drop(single_entities['id'])
data2 = data2.append(single_entities_left)
data2 = data2.append(single_entities_right)
if intermediate_set is not None:
interm_data = pd.read_pickle(intermediate_set)
if dataset != 'lspc':
cols = data.columns
if 'name' in cols:
interm_data = interm_data.rename(columns={'title':'name'})
if 'manufacturer' in cols:
interm_data = interm_data.rename(columns={'brand':'manufacturer'})
interm_data['cluster_id'] = interm_data['cluster_id']+10000
data1 = data1.append(interm_data)
data2 = data2.append(interm_data)
data1 = data1.reset_index(drop=True)
data2 = data2.reset_index(drop=True)
label_enc = LabelEncoder()
cluster_id_set = set()
cluster_id_set.update(data1['cluster_id'])
cluster_id_set.update(data2['cluster_id'])
label_enc.fit(list(cluster_id_set))
data1['labels'] = label_enc.transform(data1['cluster_id'])
data2['labels'] = label_enc.transform(data2['cluster_id'])
self.label_encoder = label_enc
data1 = data1.reset_index(drop=True)
data1 = data1.fillna('')
data1 = self._prepare_data(data1)
data2 = data2.reset_index(drop=True)
data2 = data2.fillna('')
data2 = self._prepare_data(data2)
diff = abs(len(data1)-len(data2))
if len(data1) > len(data2):
if len(data2) < diff:
sample = data2.sample(diff, replace=True)
else:
sample = data2.sample(diff)
data2 = data2.append(sample)
data2 = data2.reset_index(drop=True)
elif len(data2) > len(data1):
if len(data1) < diff:
sample = data1.sample(diff, replace=True)
else:
sample = data1.sample(diff)
data1 = data1.append(sample)
data1 = data1.reset_index(drop=True)
self.data1 = data1
self.data2 = data2
def __getitem__(self, idx):
# for every example, sample one positive from the respective sampling dataset
example1 = self.data1.loc[idx].copy()
selection1 = self.data1[self.data1['labels'] == example1['labels']]
# if len(selection1) > 1:
# selection1 = selection1.drop(idx)
pos1 = selection1.sample(1).iloc[0].copy()
example2 = self.data2.loc[idx].copy()
selection2 = self.data2[self.data2['labels'] == example2['labels']]
# if len(selection2) > 1:
# selection2 = selection2.drop(idx)
pos2 = selection2.sample(1).iloc[0].copy()
# apply augmentation if set
if self.aug:
example1['features'] = self.augmenter.apply_aug(example1['features'])
pos1['features'] = self.augmenter.apply_aug(pos1['features'])
example2['features'] = self.augmenter.apply_aug(example2['features'])
pos2['features'] = self.augmenter.apply_aug(pos2['features'])
return ((example1, pos1), (example2, pos2))
def __len__(self):
return len(self.data1)
def _prepare_data(self, data):
if self.dataset == 'lspc':
data['features'] = data.apply(serialize_sample_lspc, axis=1)
elif self.dataset == 'abt-buy':
data['features'] = data.apply(serialize_sample_abtbuy, axis=1)
elif self.dataset == 'amazon-google':
data['features'] = data.apply(serialize_sample_amazongoogle, axis=1)
data = data[['features', 'labels']]
return data
# Dataset class for pair-wise cross-entropy fine-tuning
class ContrastiveClassificationDataset(torch.utils.data.Dataset):
def __init__(self, path, dataset_type, size=None, tokenizer='huawei-noah/TinyBERT_General_4L_312D', max_length=128, dataset='lspc', aug=False):
self.max_length = max_length
self.tokenizer = AutoTokenizer.from_pretrained(tokenizer, additional_special_tokens=('[COL]', '[VAL]'))
self.dataset_type = dataset_type
self.dataset = dataset
self.aug = aug
if self.aug:
self.augmenter = Augmenter(self.aug)
if dataset == 'serialized':
data = json.loads(path)
self.data = data
return
if dataset == 'lspc':
data = pd.read_pickle(path)
else:
data = pd.read_json(path, lines=True)
if dataset == 'abt-buy':
data['brand_left'] = ''
data['brand_right'] = ''
if dataset == 'amazon-google':
data['description_left'] = ''
data['description_right'] = ''
data = data.fillna('')
if self.dataset_type != 'test':
if dataset == 'lspc':
validation_ids = pd.read_csv(f'../../data/raw/wdc-lspc/validation-sets/computers_valid_{size}.csv')
elif dataset == 'abt-buy':
validation_ids = pd.read_csv(f'../../data/interim/abt-buy/abt-buy-valid.csv')
elif dataset == 'amazon-google':
validation_ids = pd.read_csv(f'../../data/interim/amazon-google/amazon-google-valid.csv')
if self.dataset_type == 'train':
data = data[~data['pair_id'].isin(validation_ids['pair_id'])]
else:
data = data[data['pair_id'].isin(validation_ids['pair_id'])]
data = data.reset_index(drop=True)
data = self._prepare_data(data)
self.data = data
def __getitem__(self, idx):
example = self.data.loc[idx].copy()
if self.aug:
example['features_left'] = self.augmenter.apply_aug(example['features_left'])
example['features_right'] = self.augmenter.apply_aug(example['features_right'])
return example
def __len__(self):
return len(self.data)
def _prepare_data(self, data):
if self.dataset == 'lspc':
data['features_left'] = data.apply(self.serialize_sample_lspc, args=('left',), axis=1)
data['features_right'] = data.apply(self.serialize_sample_lspc, args=('right',), axis=1)
elif self.dataset == 'abt-buy':
data['features_left'] = data.apply(self.serialize_sample_abtbuy, args=('left',), axis=1)
data['features_right'] = data.apply(self.serialize_sample_abtbuy, args=('right',), axis=1)
elif self.dataset == 'amazon-google':
data['features_left'] = data.apply(self.serialize_sample_amazongoogle, args=('left',), axis=1)
data['features_right'] = data.apply(self.serialize_sample_amazongoogle, args=('right',), axis=1)
data = data[['features_left', 'features_right', 'label']]
data = data.rename(columns={'label': 'labels'})
return data
def serialize_sample_lspc(self, sample, side):
string = ''
string = f'{string}[COL] brand [VAL] {" ".join(sample[f"brand_{side}"].split(" ")[:5])}'.strip()
string = f'{string} [COL] title [VAL] {" ".join(sample[f"title_{side}"].split(" ")[:50])}'.strip()
string = f'{string} [COL] description [VAL] {" ".join(sample[f"description_{side}"].split(" ")[:100])}'.strip()
string = f'{string} [COL] specTableContent [VAL] {" ".join(sample[f"specTableContent_{side}"].split(" ")[:200])}'.strip()
return string
def serialize_sample_abtbuy(self, sample, side):
string = ''
string = f'{string}[COL] brand [VAL] {" ".join(sample[f"brand_{side}"].split())}'.strip()
string = f'{string} [COL] title [VAL] {" ".join(sample[f"name_{side}"].split())}'.strip()
string = f'{string} [COL] price [VAL] {" ".join(str(sample[f"price_{side}"]).split())}'.strip()
string = f'{string} [COL] description [VAL] {" ".join(sample[f"description_{side}"].split()[:100])}'.strip()
return string
def serialize_sample_amazongoogle(self, sample, side):
string = ''
string = f'{string}[COL] brand [VAL] {" ".join(sample[f"manufacturer_{side}"].split())}'.strip()
string = f'{string} [COL] title [VAL] {" ".join(sample[f"title_{side}"].split())}'.strip()
string = f'{string} [COL] price [VAL] {" ".join(str(sample[f"price_{side}"]).split())}'.strip()
string = f'{string} [COL] description [VAL] {" ".join(sample[f"description_{side}"].split()[:100])}'.strip()
return string |