-
Notifications
You must be signed in to change notification settings - Fork 1
/
dataset_random_group.py
199 lines (157 loc) · 7.42 KB
/
dataset_random_group.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
import torch
from torch.nn import functional as F
from nltk.corpus import wordnet as wn
import os
import sys
import time
import math
import copy
import random
import numpy as np
from collections import defaultdict
from torch.utils.data import Dataset
from wsd_models.util import *
def preprocess_context(tokenizer, text_data, max_len=-1):
data = defaultdict(dict)
ordered_ids = []
print('preprocessing data...')
num_line = 1
with open(text_data, 'r', encoding='utf8') as f:
line = f.readline()
while True:
if num_line % 1000 == 0:
print(num_line)
num_line += 1
line = f.readline()
if not line:
break
line = line.strip().split('\t')
sentence, target_index_start, target_index_end, target_id, lemma, target_pos, sense_key = line
c_ids = [torch.tensor([[x]]) for x in tokenizer.encode(tokenizer.cls_token)]
o_masks = [-1] * len(c_ids)
out_of_bound = False
for ind, word in enumerate(sentence.strip().split(' ')):
word_ids = [torch.tensor([[x]]) for x in tokenizer.encode(word.lower())]
c_ids.extend(word_ids)
if ind >= int(target_index_start) and ind < int(target_index_end):
if max_len > 0 and (len(o_masks) > max_len - 3):
out_of_bound = True
o_masks.extend([int(target_index_start)]*len(word_ids))
else:
o_masks.extend([-1]*len(word_ids))
if out_of_bound:
continue
target_lemma = generate_key(lemma, target_pos)
if target_lemma not in data.keys():
data[target_lemma]['sense'] = []
data[target_lemma]['sentence'] = []
c_ids.extend([torch.tensor([[x]]) for x in tokenizer.encode(tokenizer.sep_token)])
c_attn_mask = [1]*len(c_ids)
o_masks.extend([-1]*3)
c_ids, c_attn_masks, o_masks = normalize_length(c_ids, c_attn_mask, o_masks, max_len, pad_id=tokenizer.encode(tokenizer.pad_token)[0])
data[target_lemma]['sense'].append(sense_key)
data[target_lemma]['sentence'].append([torch.cat(c_ids,dim=-1), torch.tensor(c_attn_masks).unsqueeze(dim=0), torch.tensor(o_masks).unsqueeze(dim=0), \
target_lemma, target_id, sense_key])
ordered_ids.append(target_id)
# batch data here
keywords = list(data.keys())
return data, keywords, ordered_ids
class SemDataset(Dataset):
def __init__(self, data, batch_size):
self.data = data
self.batch_size = batch_size
self.keywords = list(data.keys())
self.keycounts = []
for key in self.keywords:
self.keycounts.append(max(1.0, len(data[key]['sense'])/(batch_size+0.0)))
self.key_cum_sum = np.cumsum(self.keycounts)
self.key_sum = np.sum(self.keycounts)
#generate even sampling mapping
self.seq_inds = {}
random.shuffle(self.keywords)
ind = 0
for key in self.keywords:
new_order = np.arange(len(self.data[key]['sense']))
np.random.shuffle(new_order)
for i in range(0,len(self.data[key]['sense']),batch_size):
self.seq_inds[ind] = []
for j in range(i, min(i+batch_size, len(self.data[key]['sense']))):
self.seq_inds[ind].append([key, new_order[j]])
ind += 1
self.length = ind
def __len__(self):
return self.length
def __getitem__(self, idx):
bz_data = [self.data[key]['sentence'][j] for key, j in self.seq_inds[idx]]
context_ids = torch.cat([x for x, _, _, _, _, _ in bz_data], dim=0)
context_attn_mask = torch.cat([x for _, x, _, _, _, _ in bz_data], dim=0)
context_output_mask = torch.cat([x for _, _, x, _, _, _ in bz_data], dim=0)
example_keys = []
instances = []
labels = []
for _, _, _, key, instance, label in bz_data:
example_keys.append(key)
instances.append(instance)
labels.append(label)
return context_ids, context_attn_mask, context_output_mask, example_keys, instances, labels
class EvalDataset(Dataset):
def __init__(self, data, ordered_ids, batch_size=1):
self.data = {}
len_sen = 0
for key in data.keys():
for sen in data[key]['sentence']:
context_ids, context_attn_mask, context_output_mask, example_keys, instances, labels = sen
self.data[instances] = sen
len_sen += 1
self.len_sen = len_sen
self.ordered_ids = ordered_ids
assert len(self.ordered_ids) == self.len_sen, 'length of instances does not match length of test sentence'
def __len__(self):
return self.len_sen
def __getitem__(self, idx):
return self.data[self.ordered_ids[idx]]
def tokenize_glosses(gloss_arr, tokenizer, max_len):
glosses = []
masks = []
for gloss_text in gloss_arr:
g_ids = [torch.tensor([[x]]) for x in tokenizer.encode(tokenizer.cls_token)+tokenizer.encode(gloss_text)+tokenizer.encode(tokenizer.sep_token)]
g_attn_mask = [1] * len(g_ids)
g_fake_mask = [-1] * len(g_ids)
g_ids, g_attn_mask, _ = normalize_length(g_ids, g_attn_mask, g_fake_mask, max_len, pad_id=tokenizer.encode(tokenizer.pad_token)[0])
g_ids = torch.cat(g_ids, dim=-1)
g_attn_mask = torch.tensor(g_attn_mask)
glosses.append(g_ids)
masks.append(g_attn_mask)
return glosses, masks
def load_and_preprocess_glosses(lemma_words, tokenizer, wn_senses, max_len=-1):
sense_glosses = {}
for target_key in lemma_words:
if target_key not in sense_glosses:
sensekey_arr = wn_senses[target_key]
gloss_arr = [wn.lemma_from_key(s).synset().definition() for s in sensekey_arr]
gloss_ids, gloss_masks = tokenize_glosses(gloss_arr, tokenizer, max_len)
gloss_ids = torch.cat(gloss_ids, dim=0)
gloss_masks = torch.stack(gloss_masks, dim=0)
sense_glosses[target_key] = (gloss_ids, gloss_masks, sensekey_arr)
return sense_glosses
def load_and_preprocess_glosses_train(lemma_words, tokenizer, wn_senses, max_len=-1):
sense_glosses = {}
sense_gloss_numlists = {}
for target_key in lemma_words:
if target_key not in sense_glosses:
sensekey_arr = wn_senses[target_key]
gloss_arr = [wn.lemma_from_key(s).synset().definition() for s in sensekey_arr]
sense_arr_numlist = [1.] * len(gloss_arr)
gloss_ids, gloss_masks = tokenize_glosses(gloss_arr, tokenizer, max_len)
gloss_ids = torch.cat(gloss_ids, dim=0)
gloss_masks = torch.stack(gloss_masks, dim=0)
sense_glosses[target_key] = (gloss_ids, gloss_masks, sensekey_arr)
sense_gloss_numlists[target_key] = sense_arr_numlist
return sense_glosses, sense_gloss_numlists
if __name__ == "__main__":
text_data = './preprocess/semcor.csv'
tokenizer = load_tokenizer('bert-base')
batched_data, keywords, ordered_ids = preprocess_context(tokenizer, text_data, max_len=128)
wn_senses = load_wn_senses('./data/WSD_Evaluation_Framework/Data_Validation/candidatesWN30.txt')
load_and_preprocess_glosses(keywords, tokenizer, wn_senses, max_len=32)
SemDataset(batched_data, 4)