-
Notifications
You must be signed in to change notification settings - Fork 0
/
data_utils.py
157 lines (132 loc) · 5.45 KB
/
data_utils.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
import os
import json
import random
import numpy as np
from tqdm import tqdm
import torch
from torch.utils.data import Dataset, DataLoader
from torchvision import transforms
import sys
import itertools
class Transform(object):
def __init__(self):
pass
def __call__(self, array, max_sents):
if max_sents < array.shape[0]:
return torch.from_numpy(array[:max_sents,:])
else:
return torch.from_numpy(np.pad(array, [(0, max_sents - array.shape[0]), (0,0)], mode = 'constant', constant_values = 0.0))
class jsonEncoder(object):
def __init__(self, json_obj=None, mode = None):
self.json_obj = json_obj
self.mode = mode
@classmethod
def from_json(cls, path, review_filename, mode):
try:
return cls(open(os.path.join(path, 'Embeddings', review_filename)), mode=mode)
except FileNotFoundError:
return cls(None)
def __call__(self):
if not self.json_obj == None:
encoded = json.load(self.json_obj)
paper = np.asarray(encoded['paper'])
reviews = []
sentiment = []
if self.mode == 'RECOMMENDATION':
rec_score = []
sentiment = []
for i, review in enumerate(encoded['reviews']):
reviews.append(np.asarray(review.get('review_text')))
rec_score.append(review.get('RECOMMENDATION'))
sentiment.append(review.get('SENTIMENT'))
return paper, reviews, rec_score, sentiment
elif self.mode == 'DECISION':
sentiment = []
for i, review in enumerate(encoded['reviews']):
reviews.append(np.asarray(review.get('review_text')))
sentiment.append(review.get('SENTIMENT'))
return paper, reviews, encoded['DECISION'], sentiment
else:
sys.exit("Provide the mode from ['RECOMMENDATION, DECISION")
else:
return None
class Data(Dataset):
def __init__(self, data, mode = 'RECOMMENDATION', transform = None, max_paper_sentences=-1, max_review_sentences=-1):
self.data = data
self.mode = mode
self.max_paper_sentences, self.max_review_sentences = max_paper_sentences, max_review_sentences
self.transform = transform
@classmethod
def readData(cls, path, transform = None, jsonEncoder = jsonEncoder(), mode='RECOMMENDATION', slice_=100):
reviews_dir = os.listdir(os.path.join(path, 'reviews'))[:slice_]
papers, reviews, rec_scores, reviews_all, decision, sentiment = [], [], [], [], [], []
max_paper_sents = 0
max_review_sents = 0
pbar = tqdm(reviews_dir)
for review_dir in pbar:
pbar.set_description("Reading Embeddings...")
ret = jsonEncoder.from_json(path, review_dir, mode=mode)()
if ret == None:
continue
if ret[0].shape[0] > max_paper_sents:
max_paper_sents = ret[0].shape[0]
if mode == 'RECOMMENDATION':
for i, rev in enumerate(ret[1],0):
papers.append(ret[0])
reviews.append(rev)
sentiment.append(ret[3][i])
rec_scores.append(int(ret[2][i]))
if rev.shape[0] > max_review_sents:
max_review_sents = rev.shape[0]
elif mode == 'DECISION':
papers.append(ret[0])
decision.append(ret[2])
sentiment_merged = list(itertools.chain.from_iterable(ret[3]))
rev_merged = list(itertools.chain.from_iterable(ret[1]))
reviews_all.append(rev_merged)
sentiment.append(sentiment_merged)
if len(rev_merged) > max_review_sents:
max_review_sents = len(rev_merged)
else:
sys.exit("Provide a valid mode from [RECOMMENDATION, DECISION]")
if mode == 'RECOMMENDATION':
return cls((papers, reviews, sentiment, rec_scores), transform=transform, max_paper_sentences=max_paper_sents, max_review_sentences=max_review_sents, mode=mode)
if mode == 'DECISION':
return cls((papers, reviews_all, sentiment, decision), transform=transform, max_paper_sentences=max_paper_sents, max_review_sentences=max_review_sents, mode=mode)
def __getitem__(self, index):
if self.mode == 'RECOMMENDATION':
if self.transform:
return self.transform(np.asarray(self.data[0][index]), self.max_paper_sentences), \
self.transform(np.asarray(self.data[1][index]), self.max_review_sentences), \
self.transform(np.asarray(self.data[2][index]), self.max_review_sentences),\
self.data[3][index]
else:
sys.exit("Use the transform!")
elif self.mode == 'DECISION':
if self.transform:
return self.transform(np.asarray(self.data[0][index]), self.max_paper_sentences), \
self.transform(np.asarray(self.data[1][index]), self.max_review_sentences), \
self.transform(np.asarray(self.data[2][index]), self.max_review_sentences),\
self.data[3][index]
else:
pass
def __len__(self):
return len(self.data)
def getLoaders(data_path = './2018/', mode='RECOMMENDATION', batch_size=8, slice=[-1, -1, -1], valid_path='./2018/'):
print('Reading the training Dataset...')
train_dataset = Data.readData(data_path, mode=mode, slice_=slice[0], transform=Transform())
print('Reading the validation Dataset...')
valid_dataset = Data.readData(valid_path, mode=mode, slice_=slice[1], transform=Transform())
trainloader = DataLoader(train_dataset, batch_size = batch_size, shuffle = True, num_workers=4)
validloader = DataLoader(valid_dataset, batch_size = batch_size, shuffle=True, num_workers=4)
return trainloader, validloader, train_dataset.max_review_sentences, train_dataset.max_paper_sentences
if __name__ == "__main__":
trainloader, validloader, _, _ = getLoaders(batch_size=2, mode='DECISION', slice=[5,5,5])
print(len(trainloader), len(validloader))
for i, d in enumerate(zip(*(trainloader, validloader)), 0):
train, valid = d
print(train[0].shape)
print(train[1].shape)
print(train[2].shape)
print(train[3])
break