-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdataloader.py
301 lines (240 loc) · 12.6 KB
/
dataloader.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
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
"""
author: hzj
date: 2024-6-18
file info:
"""
import random
import torch.nn.functional as func
import numpy as np
import pandas as pd
import torch
from torch.utils.data import Dataset, DataLoader
import scipy.sparse as sp
from scipy.sparse.linalg import svds
import os
class GraphDataset(Dataset):
def __init__(self, flags_obj):
super(GraphDataset, self).__init__()
self.flags_obj = flags_obj
self.dataset_path = os.path.join(self.flags_obj.dataset, self.flags_obj.dataset_name)
self.num_users = 0
self.num_items = 0
# self.trainInteractionSize = 0
# self.testInteractionSize = 0
self.popularity = None
self.activity = None
self.lil_train_record = None
self.lil_test_record = None
self.symmetric_Graph = None # (n+m) x (n+m)，symmetric norm
self.origin_Graph = None # (n+m) x (n+m)，original
self.symmetric_sub_graph = None # n x m，symmetric norm
self.origin_sub_graph = None # n x m，original
self.SVD_symmetric_sub_graph = None # n x m，symmetric norm，svd
self.SVD_origin_sub_graph = None # n x m，original，svd
self.U_mul_S = None
self.V_mul_S = None
self.train_csr_record = sp.load_npz(os.path.join(self.dataset_path, 'train_csr_record.npz'))
self.train_csr_record = self.train_csr_record.astype(np.bool).astype(np.int)
self.test_csr_record = sp.load_npz(os.path.join(self.dataset_path, 'test_csr_record.npz'))
self.test_csr_record = self.test_csr_record.astype(np.bool).astype(np.int)
self.test_coo_record = self.test_csr_record.tocoo(copy=True)
self.test_labels = [[] for _ in range(self.test_coo_record.shape[0])]
self.intermediate_items = {}
self.structure_weight = {}
self.distant_items = {}
self.init()
def init(self):
self.num_users = self.train_csr_record.shape[0]
self.num_items = self.train_csr_record.shape[1]
# self.lil_train_record = (self.train_coo_record + self.train_skew_coo_record).tolil(copy=True)
self.lil_train_record = self.train_csr_record.tolil(copy=True)
self.lil_test_record = self.test_csr_record.tolil(copy=True)
# self.activity = np.array(self.lil_train_record.sum(axis=1)).flatten()
# self.popularity = np.array(self.lil_train_record.sum(axis=0)).flatten()
self.activity = np.array(self.lil_train_record.sum(axis=1))
self.popularity = np.array(self.lil_train_record.sum(axis=0))
self.initSparseGraph()
for i in range(len(self.test_coo_record.data)):
row = self.test_coo_record.row[i]
col = self.test_coo_record.col[i]
self.test_labels[row].append(col)
if self.flags_obj.model_name in ['AED']:
self.khops_walk()
def khops_walk(self):
try:
with open(os.path.join(self.dataset_path, 'intermediate_items.txt'), 'r') as f:
for line in f.readlines():
line = line.strip('\n').split('::')
index = line[0].strip(' ').split(' ')
weight = line[1].strip(' ').split(' ')
if len(index) > 1:
user = int(index[0])
user_intermediate_items = [int(i) for i in index[1:]]
self.intermediate_items[user] = user_intermediate_items
user_intermediate_weight = [int(i) for i in weight]
self.structure_weight[user] = user_intermediate_weight
else:
self.intermediate_items[user] = []
self.structure_weight[user] = []
print("Successfully loaded users intermediate items...")
except:
user_1_hops = self.train_csr_record
item_1_hops = self.train_csr_record.T
user_2_hops = user_1_hops.dot(item_1_hops)
diags = sp.diags(np.ones(self.num_users), format='csr', dtype=np.int32)
user_2_hops = user_2_hops - user_2_hops.multiply(diags)
weight_user_1_3_hops = user_2_hops.dot(user_1_hops)
weight_user_3_hops = weight_user_1_3_hops - weight_user_1_3_hops.multiply(user_1_hops)
weight_user_3_hops = weight_user_3_hops.toarray().astype(np.int32)
weight_act_pop = (1 - self.flags_obj.temp) * np.log(1 + self.activity * self.popularity).astype(np.float32)
weight_3_hop = self.flags_obj.temp * np.log(1 + weight_user_3_hops).astype(np.float32)
self.structure_weight = weight_3_hop + weight_act_pop
# self.structure_weight = weight_user_3_hops
def initSparseGraph(self):
try:
# pop_adj = sp.load_npz(os.path.join(self.dataset_path, 's_pre_pop_adj_matrix.npz'))
# print('Successfully loaded popular adjacency matrix...')
symmetric_matrix = sp.load_npz(os.path.join(self.dataset_path, 'symmetric_matrix_csr.npz'))
print('Successfully loaded symmetric matrix...')
original_matrix = sp.load_npz(os.path.join(self.dataset_path, 'original_matrix_csr.npz'))
print('Successfully loaded original matrix...')
symmetric_sub_matrix = sp.load_npz(os.path.join(self.dataset_path, 'symmetric_sub_matrix_csr.npz'))
print('Successfully loaded symmetric sub-matrix...')
except:
empty_matrix = sp.dok_matrix((self.num_users + self.num_items, self.num_users + self.num_items),
dtype=np.float32)
original_matrix = empty_matrix.tolil(copy=True)
# recon_svd_matrix = empty_matrix.tolil(copy=True)
original_matrix[:self.num_users, self.num_users:] = self.lil_train_record
original_matrix[self.num_users:, :self.num_users] = self.lil_train_record.T
original_matrix = original_matrix.todok()
row_sum = np.array(original_matrix.sum(axis=1))
d_inv = np.power(row_sum, -0.5).flatten()
d_inv[np.isinf(d_inv)] = 0.
d_matrix = sp.diags(d_inv)
symmetric_matrix = d_matrix.dot(original_matrix).dot(d_matrix)
# row_sum = np.array(norm_adj.sum(axis=1))
symmetric_matrix = symmetric_matrix.tocsr()
sp.save_npz(os.path.join(self.dataset_path, 'symmetric_matrix_csr.npz'), symmetric_matrix)
print("Successfully generating symmetric matrix")
original_matrix = original_matrix.tocsr()
sp.save_npz(os.path.join(self.dataset_path, 'original_matrix_csr.npz'), original_matrix)
print("Successfully generating original matrix")
train_dok = self.train_csr_record.todok(copy=True)
activity_sum = np.array(train_dok.sum(axis=-1))
d_activity = np.power(activity_sum, -0.5).flatten()
d_activity = sp.diags(d_activity)
popular_sum = np.array(train_dok.sum(axis=0))
d_popular = np.power(popular_sum, -0.5).flatten()
d_popular = sp.diags(d_popular)
symmetric_sub_matrix = d_activity.dot(train_dok).dot(d_popular)
symmetric_sub_matrix = symmetric_sub_matrix.tocsr()
sp.save_npz(os.path.join(self.dataset_path, 'symmetric_sub_matrix_csr.npz'), symmetric_sub_matrix)
print("Successfully generating symmetric sub-matrix")
if self.flags_obj.adj_split:
self.symmetric_Graph = self._split_A_hat(symmetric_matrix)
else:
# self.popular_Graph = self._convert_sp_matrix_to_tensor(pop_adj)
# self.popular_Graph = self.popular_Graph.coalesce().to(self.flags_obj.device)
self.symmetric_Graph = self.convert_sp_matrix_to_tensor(symmetric_matrix)
self.symmetric_Graph = self.symmetric_Graph.coalesce().to(self.flags_obj.device)
self.symmetric_sub_graph = self.convert_sp_matrix_to_tensor(symmetric_sub_matrix)
self.symmetric_sub_graph = self.symmetric_sub_graph.coalesce().to(self.flags_obj.device)
self.origin_Graph = self.convert_sp_matrix_to_tensor(original_matrix)
self.origin_Graph = self.origin_Graph.coalesce().to(self.flags_obj.device)
self.origin_sub_graph = self.convert_sp_matrix_to_tensor(self.train_csr_record)
self.origin_sub_graph = self.origin_sub_graph.coalesce().to(self.flags_obj.device)
if self.flags_obj.model_name in ['lightGCL']:
svd_u, s, svd_v = torch.svd_lowrank(self.symmetric_sub_graph, q=self.flags_obj.q)
self.SVD_symmetric_sub_graph = svd_u @ torch.diag(s) @ svd_v.T
self.U_mul_S = svd_u @ torch.diag(s)
self.V_mul_S = svd_v @ torch.diag(s)
elif self.flags_obj.model_name in ['AED']:
svd_u, s, svd_v = torch.svd_lowrank(self.origin_sub_graph, q=self.flags_obj.q)
self.SVD_origin_sub_graph = svd_u @ torch.diag(s) @ svd_v.T
self.U_mul_S = svd_u @ torch.diag(s)
self.V_mul_S = svd_v @ torch.diag(s)
def _split_A_hat(self, adj):
adj_fold = []
num_folds = 5
fold_len = (self.num_users + self.num_items) // num_folds
for fold in range(num_folds):
start = fold * fold_len
if fold == num_folds - 1:
end = self.num_users + self.num_items
else:
end = (fold + 1) * fold_len
adj_fold.append(self.convert_sp_matrix_to_tensor(adj[start:end]).coalesce().to('cuda:0'))
return adj_fold
@staticmethod
def convert_sp_matrix_to_tensor(x):
coo = x.tocoo().astype(np.float32)
row = torch.Tensor(coo.row).long()
col = torch.Tensor(coo.col).long()
index = torch.stack([row, col])
data = torch.FloatTensor(coo.data)
return torch.sparse.FloatTensor(index, data, torch.Size(coo.shape))
class UniformTrainDataset(Dataset):
def __init__(self, dataset: GraphDataset):
super(UniformTrainDataset, self).__init__()
self.lil_train_record = dataset.lil_train_record
self.num_users = dataset.num_users
self.num_items = dataset.num_items
def __len__(self):
return self.lil_train_record.nnz
def __getitem__(self, index):
while True:
user = np.random.randint(0, self.num_users)
user_positive_items = self.lil_train_record.rows[user]
if user_positive_items:
positive_item = random.choice(user_positive_items)
break
while True:
negative_item = np.random.randint(0, self.num_items)
if negative_item in user_positive_items:
continue
else:
break
return user, positive_item, negative_item
class GCNRSTrainDataset(Dataset):
def __init__(self, dataset: GraphDataset):
super(GCNRSTrainDataset, self).__init__()
self.lil_train_record = dataset.lil_train_record
self.dok_train_record = dataset.lil_train_record.todok(copy=True)
self.train_record = list(self.dok_train_record.keys())
self.num_items = dataset.num_items
def __len__(self):
return self.lil_train_record.nnz
def __getitem__(self, index):
user = self.train_record[index][0]
positive_item = self.train_record[index][1]
user_positive_items = self.lil_train_record.rows[user]
while True:
negative_item = np.random.randint(0, self.num_items)
if negative_item in user_positive_items:
continue
else:
break
return user, positive_item, negative_item
def shuffle(*arrays):
shuffle_indices = np.arange(len(arrays[0]))
np.random.shuffle(shuffle_indices)
if len(arrays) == 1:
result = arrays[0][shuffle_indices]
else:
result = tuple(array[shuffle_indices] for array in arrays)
return result
def minibatch(*tensors, batch_size):
if len(tensors) == 1:
tensor = tensors[0]
for i in range(0, len(tensor), batch_size):
yield tensor[i:i + batch_size]
else:
for i in range(0, len(tensors[0]), batch_size):
yield tuple(tensor[i:i + batch_size] for tensor in tensors)
def set_seed(seed):
np.random.seed(seed)
if torch.cuda.is_available():
torch.cuda.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
torch.manual_seed(seed)