File size: 5,559 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 | import numpy as np
np.random.seed(42)
import random
random.seed(42)
from dataclasses import dataclass
from typing import Optional
import torch
import torch.nn.functional as F
from transformers.tokenization_utils_base import BatchEncoding, PreTrainedTokenizerBase
from pdb import set_trace
# collator for pair-wise cross-entropy fine-tuning
@dataclass
class DataCollatorContrastiveClassification:
tokenizer: PreTrainedTokenizerBase
max_length: Optional[int] = 128
pad_to_multiple_of: Optional[int] = None
return_tensors: str = "pt"
def __call__(self, input):
features_left = [x['features_left'] for x in input]
features_right = [x['features_right'] for x in input]
labels = [x['labels'] for x in input]
batch_left = self.tokenizer(features_left, padding=True, truncation=True, max_length=self.max_length, return_tensors=self.return_tensors)
batch_right = self.tokenizer(features_right, padding=True, truncation=True, max_length=self.max_length, return_tensors=self.return_tensors)
batch = batch_left
if 'token_type_ids' in batch.keys():
del batch['token_type_ids']
batch['input_ids_right'] = batch_right['input_ids']
batch['attention_mask_right'] = batch_right['attention_mask']
batch['labels'] = torch.LongTensor(labels)
return batch
# collator for self-supervised contrastive pre-training
@dataclass
class DataCollatorContrastivePretrainSelfSupervised:
tokenizer: PreTrainedTokenizerBase
max_length: Optional[int] = 128
pad_to_multiple_of: Optional[int] = None
return_tensors: str = "pt"
def __call__(self, input):
features_left = [x[0]['features'] for x in input]
labels = [x[0]['labels'] for x in input]
batch = self.tokenizer(features_left, padding=True, truncation=True, max_length=self.max_length, return_tensors=self.return_tensors)
if 'token_type_ids' in batch.keys():
del batch['token_type_ids']
batch['labels'] = torch.LongTensor(labels)
return batch
# collator for supervised contrastive pre-training for WDC Computers
@dataclass
class DataCollatorContrastivePretrain:
tokenizer: PreTrainedTokenizerBase
max_length: Optional[int] = 128
pad_to_multiple_of: Optional[int] = None
return_tensors: str = "pt"
def __call__(self, input):
features_left = [x[0]['features'] for x in input]
features_right = [x[1]['features'] for x in input]
labels = [x[0]['labels'] for x in input]
batch_left = self.tokenizer(features_left, padding=True, truncation=True, max_length=self.max_length, return_tensors=self.return_tensors)
batch_right = self.tokenizer(features_right, padding=True, truncation=True, max_length=self.max_length, return_tensors=self.return_tensors)
batch = batch_left
if 'token_type_ids' in batch.keys():
del batch['token_type_ids']
batch['input_ids_right'] = batch_right['input_ids']
batch['attention_mask_right'] = batch_right['attention_mask']
batch['labels'] = torch.LongTensor(labels)
return batch
# collator for supervised contrastive pre-training for Abt-Buy and Amazon-Google
# randomly chooses the sampling dataset when using source-aware sampling
@dataclass
class DataCollatorContrastivePretrainDeepmatcher:
tokenizer: PreTrainedTokenizerBase
max_length: Optional[int] = 128
pad_to_multiple_of: Optional[int] = None
return_tensors: str = "pt"
def __call__(self, input_both):
rnd = random.choice([0,1])
input = [x[rnd] for x in input_both]
features_left = [x[0]['features'] for x in input]
features_right = [x[1]['features'] for x in input]
labels = [x[0]['labels'] for x in input]
batch_left = self.tokenizer(features_left, padding=True, truncation=True, max_length=self.max_length, return_tensors=self.return_tensors)
batch_right = self.tokenizer(features_right, padding=True, truncation=True, max_length=self.max_length, return_tensors=self.return_tensors)
batch = batch_left
if 'token_type_ids' in batch.keys():
del batch['token_type_ids']
batch['input_ids_right'] = batch_right['input_ids']
batch['attention_mask_right'] = batch_right['attention_mask']
batch['labels'] = torch.LongTensor(labels)
return batch
# collator for pair-wise cross-entropy fine-tuning
@dataclass
class DataCollatorContrastiveClassification:
tokenizer: PreTrainedTokenizerBase
max_length: Optional[int] = 128
pad_to_multiple_of: Optional[int] = None
return_tensors: str = "pt"
def __call__(self, input):
features_left = [x['features_left'] for x in input]
features_right = [x['features_right'] for x in input]
labels = [x['labels'] for x in input]
batch_left = self.tokenizer(features_left, padding=True, truncation=True, max_length=self.max_length, return_tensors=self.return_tensors)
batch_right = self.tokenizer(features_right, padding=True, truncation=True, max_length=self.max_length, return_tensors=self.return_tensors)
batch = batch_left
if 'token_type_ids' in batch.keys():
del batch['token_type_ids']
batch['input_ids_right'] = batch_right['input_ids']
batch['attention_mask_right'] = batch_right['attention_mask']
batch['labels'] = torch.LongTensor(labels)
return batch |