-
Notifications
You must be signed in to change notification settings - Fork 1
/
utility.py
182 lines (133 loc) · 7.77 KB
/
utility.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
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
import os
import random
import numpy as np
import scipy.sparse as sp
import torch
from torch.utils.data import Dataset, DataLoader
def print_statistics(X, string):
print('>'*10 + string + '>'*10 )
print('Average interactions', X.sum(1).mean(0).item())
nonzero_row_indice, nonzero_col_indice = X.nonzero()
unique_nonzero_row_indice = np.unique(nonzero_row_indice)
unique_nonzero_col_indice = np.unique(nonzero_col_indice)
print('Non-zero rows', len(unique_nonzero_row_indice)/X.shape[0])
print('Non-zero columns', len(unique_nonzero_col_indice)/X.shape[1])
print('Matrix density', len(nonzero_row_indice)/(X.shape[0]*X.shape[1]))
class BundleTrainDataset(Dataset):
def __init__(self, conf, u_b_pairs, u_b_graph, num_bundles, u_b_for_neg_sample, b_b_for_neg_sample, neg_sample=1,
u_i_pairs=None, u_i_graph=None, num_items=0 # for BPR
):
self.conf = conf
self.u_b_pairs = u_b_pairs
self.u_b_graph = u_b_graph
self.num_bundles = num_bundles
self.neg_sample = neg_sample
self.u_i_pairs = u_i_pairs
self.u_i_graph = u_i_graph
self.num_items = num_items
self.u_b_for_neg_sample = u_b_for_neg_sample
self.b_b_for_neg_sample = b_b_for_neg_sample
def __getitem_ui__(self,index):
user_i, pos_item = self.u_i_pairs[index]
all_items = [pos_item]
while True:
i = np.random.randint(self.num_items)
if self.u_i_graph[user_i, i] == 0 and not i in all_items:
all_items.append(i)
if len(all_items) == self.neg_sample+1:
break
return torch.LongTensor([user_i]), torch.LongTensor(all_items)
def __getitem__(self, index):
user_b, pos_bundle = self.u_b_pairs[index]
all_bundles = [pos_bundle]
while True:
i = np.random.randint(self.num_bundles)
if self.u_b_graph[user_b, i] == 0 and not i in all_bundles:
all_bundles.append(i)
if len(all_bundles) == self.neg_sample+1:
break
user_b = torch.LongTensor([user_b])
bundles = torch.LongTensor(all_bundles)
return user_b, bundles
def __len__(self):
return len(self.u_b_pairs)
class BundleTestDataset(Dataset):
def __init__(self, u_b_pairs, u_b_graph, u_b_graph_train, num_users, num_bundles):
self.u_b_pairs = u_b_pairs
self.u_b_graph = u_b_graph
self.train_mask_u_b = u_b_graph_train
self.num_users = num_users
self.num_bundles = num_bundles
self.users = torch.arange(num_users, dtype=torch.long).unsqueeze(dim=1)
self.bundles = torch.arange(num_bundles, dtype=torch.long)
def __getitem__(self, index):
u_b_grd = torch.from_numpy(self.u_b_graph[index].toarray()).squeeze()
u_b_mask = torch.from_numpy(self.train_mask_u_b[index].toarray()).squeeze()
return index, u_b_grd, u_b_mask
def __len__(self):
return self.u_b_graph.shape[0]
class Datasets():
def __init__(self, conf):
self.path = conf['data_path']
self.name = conf['dataset']
batch_size_train = conf['batch_size_train']
batch_size_test = conf['batch_size_test']
self.num_users, self.num_bundles, self.num_items = self.get_data_size()
b_i_graph = self.get_bi()
u_i_pairs, u_i_graph = self.get_ui()
u_b_pairs_train, u_b_graph_train = self.get_ub("train")
u_b_pairs_val, u_b_graph_val = self.get_ub("tune")
u_b_pairs_test, u_b_graph_test = self.get_ub("test")
u_b_for_neg_sample, b_b_for_neg_sample = None, None
self.bundle_train_data = BundleTrainDataset(conf, u_b_pairs_train, u_b_graph_train, self.num_bundles, u_b_for_neg_sample, b_b_for_neg_sample, conf["neg_num"],u_i_pairs, u_i_graph, self.num_items)
self.bundle_val_data = BundleTestDataset(u_b_pairs_val, u_b_graph_val, u_b_graph_train, self.num_users, self.num_bundles)
self.bundle_test_data = BundleTestDataset(u_b_pairs_test, u_b_graph_test, u_b_graph_train, self.num_users, self.num_bundles)
self.graphs = [u_b_graph_train, u_i_graph, b_i_graph]
self.train_loader = DataLoader(self.bundle_train_data, batch_size=batch_size_train, shuffle=True, num_workers=10, drop_last=True)
self.val_loader = DataLoader(self.bundle_val_data, batch_size=batch_size_test, shuffle=False, num_workers=20)
self.test_loader = DataLoader(self.bundle_test_data, batch_size=batch_size_test, shuffle=False, num_workers=20)
def get_data_size(self):
name = self.name
if "_" in name:
name = name.split("_")[0]
with open(os.path.join(self.path, self.name, '{}_data_size.txt'.format(name)), 'r') as f:
return [int(s) for s in f.readline().split('\t')][:3]
def get_aux_graph(self, u_i_graph, b_i_graph, conf):
u_b_from_i = u_i_graph @ b_i_graph.T
u_b_from_i = u_b_from_i.todense()
bn1_window = [int(i*self.num_bundles) for i in conf['hard_window']]
u_b_for_neg_sample = np.argsort(u_b_from_i, axis=1)[:, bn1_window[0]:bn1_window[1]]
b_b_from_i = b_i_graph @ b_i_graph.T
b_b_from_i = b_b_from_i.todense()
bn2_window = [int(i*self.num_bundles) for i in conf['hard_window']]
b_b_for_neg_sample = np.argsort(b_b_from_i, axis=1)[:, bn2_window[0]:bn2_window[1]]
return u_b_for_neg_sample, b_b_for_neg_sample
def get_bi(self):
with open(os.path.join(self.path, self.name, 'bundle_item.txt'), 'r') as f:
b_i_pairs = list(map(lambda s: tuple(int(i) for i in s[:-1].split('\t')), f.readlines()))
indice = np.array(b_i_pairs, dtype=np.int32)
values = np.ones(len(b_i_pairs), dtype=np.float32)
b_i_graph = sp.coo_matrix(
(values, (indice[:, 0], indice[:, 1])), shape=(self.num_bundles, self.num_items)).tocsr()
print_statistics(b_i_graph, 'B-I statistics')
return b_i_graph
def get_ui(self):
with open(os.path.join(self.path, self.name, 'user_item.txt'), 'r') as f:
u_i_pairs = list(map(lambda s: tuple(int(i) for i in s[:-1].split('\t')), f.readlines()))
indice = np.array(u_i_pairs, dtype=np.int32)
values = np.ones(len(u_i_pairs), dtype=np.float32)
u_i_graph = sp.coo_matrix(
(values, (indice[:, 0], indice[:, 1])), shape=(self.num_users, self.num_items)).tocsr()
print_statistics(u_i_graph, 'U-I statistics')
return u_i_pairs, u_i_graph
def get_ub(self, task):
with open(os.path.join(self.path, self.name, 'user_bundle_{}.txt'.format(task)), 'r') as f:
u_b_pairs = list(map(lambda s: tuple(int(i) for i in s[:-1].split('\t')), f.readlines()))
indice = np.array(u_b_pairs, dtype=np.int32)
values = np.ones(len(u_b_pairs), dtype=np.float32)
u_b_graph = sp.coo_matrix(
(values, (indice[:, 0], indice[:, 1])), shape=(self.num_users, self.num_bundles)).tocsr()
print_statistics(u_b_graph, "U-B statistics in %s" %(task))
return u_b_pairs, u_b_graph