-
Notifications
You must be signed in to change notification settings - Fork 10
/
reader.py
184 lines (147 loc) · 5.69 KB
/
reader.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
import os
import glob
import math
import random
import pickle
from collections import defaultdict
import hickle
from tqdm import tqdm, trange
import numpy as np
PAD_INDEX = 0
PAD_WORD = '<PAD>'
START_INDEX = 1
START_WORD = '<STR>'
END_INDEX = 2
END_WORD = '<END>'
UNK_INDEX = 3
UNK_WORD = '<UNK>'
def get_review_data(users, items, ratings, review_data):
new_users = []
new_items = []
new_ratings = []
new_photos = []
new_reviews = []
for user, item, rating in zip(users, items, ratings):
for photo_id, reviews in review_data[(user, item)]:
new_users.append(user)
new_items.append(item)
new_ratings.append(rating)
new_photos.append(photo_id)
new_reviews.append(reviews)
return new_users, new_items, new_ratings, new_photos, new_reviews
def batch_review_normalize(reviews, max_length=None):
batch_size = len(reviews)
if max_length:
reviews = [review[:max_length] for review in reviews]
else:
max_length = max([len(review) for review in reviews])
norm_reviews = np.zeros(shape=[batch_size, max_length], dtype=np.int32) # == PAD
for i, review in enumerate(reviews):
for j, word in enumerate(review):
norm_reviews[i, j] = word
return norm_reviews
class DataReader:
def __init__(self, data_dir, training_shuffle=True):
self.data_dir = data_dir
self.is_shuffle = training_shuffle
self.total_users = len(self._read_ids(os.path.join(data_dir, 'users.txt')))
self.total_items = len(self._read_ids(os.path.join(data_dir, 'items.txt')))
print('Total users: {}, total items: {}'.format(self.total_users, self.total_items))
train_data = self._read_data(os.path.join(data_dir, 'train.pkl'))
test_data = self._read_data(os.path.join(data_dir, 'test.pkl'))
self.train_rating, self.train_review = self._prepare_data(train_data, training=True)
self.test_rating, self.test_review = self._prepare_data(test_data)
self.global_rating = np.asarray(self.train_rating)[:, 2].mean()
print('Global rating: {:.2f}'.format(self.global_rating))
self.load_images()
def load_images(self):
self.train_id2idx = self._read_img_id2idx(os.path.join(self.data_dir, 'train.id_to_idx.pkl'))
self.train_img_features = self._read_img_feature(os.path.join(self.data_dir, 'img_feats/train'),
len(self.train_id2idx.keys()))
self.test_id2idx = self._read_img_id2idx(os.path.join(self.data_dir, 'test.id_to_idx.pkl'))
self.test_img_features = self._read_img_feature(os.path.join(self.data_dir, 'img_feats/test'),
len(self.test_id2idx.keys()))
def read_train_set(self, batch_size, rating_only=False):
if self.is_shuffle:
random.shuffle(self.train_rating)
if rating_only:
return self.batch_iterator(self.train_rating, batch_size, True, desc='Training')
return self.batch_iterator(self.train_review, batch_size, desc='Training')
def read_test_set(self, batch_size, rating_only=False):
if rating_only:
return self.batch_iterator(self.test_rating, batch_size, True, desc='Testing')
return self.batch_iterator(self.test_review, batch_size, desc='Testing')
def batch_iterator(self, data, batch_size, rating_only=False, desc=None):
num_batches = int(math.ceil(len(data) / batch_size))
self.iter = trange(num_batches, desc=desc)
for cur_batch in self.iter:
begin = batch_size * cur_batch
end = batch_size * cur_batch + batch_size
if end > len(data):
end = len(data)
batch_users = []
batch_items = []
batch_ratings = []
batch_photos = []
batch_reviews = []
for exp in data[begin:end]:
batch_users.append(exp[0])
batch_items.append(exp[1])
batch_ratings.append(exp[2])
if not rating_only:
batch_photos.append(exp[3])
batch_reviews.append(exp[4])
if rating_only:
yield batch_users, batch_items, batch_ratings
else:
yield batch_users, batch_items, batch_ratings, batch_photos, batch_reviews
@staticmethod
def _read_ids(file_path):
print('Reading data: %s' % file_path)
data = [0]
with open(file_path, 'r') as f:
for line in f:
data.append(int(line.split()[1]))
return set(data)
@staticmethod
def _read_img_feature(feat_dir, num_imgs):
print('Reading image features: %s' % feat_dir)
all_feats = np.ndarray([num_imgs, 196, 512], dtype=np.float32)
for file_path in tqdm(glob.glob('{}/*.hkl'.format(feat_dir))):
start = int(file_path.split('/')[-1].split('_')[0])
end = int(file_path.split('/')[-1].split('_')[1].split('.')[0])
all_feats[start:end, :] = hickle.load(file_path)
return all_feats
@staticmethod
def _read_img_id2idx(file_path):
print('Reading image id_to_idx: %s' % file_path)
with open(file_path, 'rb') as f:
return pickle.load(f)
@staticmethod
def _read_data(file_path):
print('Reading data: %s' % file_path)
data = []
with open(file_path, 'rb') as f:
try:
while True:
exp = pickle.load(f)
data.append(exp)
except EOFError:
pass
return data
@staticmethod
def _prepare_data(data, training=False):
rating_data = []
review_data = defaultdict(list)
for exp in data:
user = int(exp['User'])
item = int(exp['Item'])
rating = exp['Rating']
rating_data.append((user, item, rating))
for photo_id, photo_reviews in exp['Reviews'].items():
if training:
for photo_review in photo_reviews:
review_data[(user, item)].append((photo_id, photo_review))
else:
review_data[(user, item)].append((photo_id, photo_reviews))
return rating_data, review_data