-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathmodel_my_extention.py
206 lines (172 loc) · 7.45 KB
/
model_my_extention.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
from __future__ import print_function
import argparse
import numpy as np
import torch
import torch.utils.data
from torch import nn, optim
import torch.nn.functional as F
# class Cluster_layer(nn.Module):
# def __init__(self, emb_size, num_cluster=2, iters=4, tau=1.0, **kwargs):
# super(Cluster_layer, self).__init__()
# self.n_cluster = num_cluster
# self.iters = iters
# self.tau = tau
# self.centers = nn.Parameter(torch.nn.init.xavier_uniform_(torch.empty(self.n_cluster, emb_size)))
#
# def forward(self, u_vecs):
# for i in range(self.iters):
# distance = torch.matmul(u_vecs.unsqueeze(1), F.normalize(self.centers, p=1, dim=1).T)
# assigns = F.softmax(distance*self.tau)
#
#
# return o, c
class CDAE(nn.Module):
def __init__(self, NUM_USER, NUM_MOVIE, NUM_BOOK, EMBED_SIZE, dropout, x_cluster_num=10, y_cluster_num=10, is_sparse=False):
super(CDAE, self).__init__()
self.NUM_MOVIE = NUM_MOVIE
self.NUM_BOOK = NUM_BOOK
self.NUM_USER = NUM_USER
self.emb_size = EMBED_SIZE
self.user_embeddings = nn.Embedding(self.NUM_USER, EMBED_SIZE, sparse=is_sparse)
self.user_embeddings.weight.data = torch.from_numpy(
np.random.normal(0, 0.01, size=[self.NUM_USER, EMBED_SIZE])).float()
self.encoder_x = nn.Sequential(
nn.Linear(self.NUM_MOVIE, EMBED_SIZE),
nn.ReLU(),
nn.Linear(EMBED_SIZE, EMBED_SIZE)
)
self.decoder_x = nn.Sequential(
nn.Linear(EMBED_SIZE, EMBED_SIZE),
nn.ReLU(),
nn.Linear(EMBED_SIZE, self.NUM_MOVIE)
)
self.encoder_y = nn.Sequential(
nn.Linear(self.NUM_BOOK, EMBED_SIZE),
nn.ReLU(),
nn.Linear(EMBED_SIZE, EMBED_SIZE)
)
self.decoder_y = nn.Sequential(
nn.Linear(EMBED_SIZE, EMBED_SIZE),
nn.ReLU(),
nn.Linear(EMBED_SIZE, self.NUM_BOOK)
)
self.x_clustering = Cluster_layer(num_cluster=x_cluster_num)
self.y_clustering = Cluster_layer(num_cluster=y_cluster_num)
self.epsilon = torch.tensor(1e-10).type(torch.FloatTensor) # .cuda()
self.dropout = nn.Dropout(dropout)
self.relu = nn.ReLU
def forward(self, batch_user, batch_user_x, batch_user_y):
# encoding for x and y domain
h_user_x = self.encoder_x(self.dropout(batch_user_x))
h_user_y = self.encoder_y(self.dropout(batch_user_y))
# disentangling for user representations in x and y domains
x_centers, x_assigns = self.x_clustering(h_user_x.unsqueeze(1))
y_centers, y_assigns = self.y_clustering(h_user_y.unsqueeze(1))
x_centers_ = torch.sum(x_centers ** 2, dim=2, keepdim=True)
y_centers_ = torch.sum(y_centers ** 2, dim=2, keepdim=True)
distance_ = torch.sqrt(torch.max(x_centers_ - 2 * torch.bmm(x_centers, y_centers.permute(0, 2, 1)) + y_centers_.permute(0, 2, 1),
self.epsilon))
distance = torch.min(distance_, dim=1)
distance = distance.values
# adding the ont-hot user encoding for user representations in x and y domains
# h_user_x_ =
h_user = self.user_embeddings(batch_user)
feature_x = torch.add(h_user_x, h_user)
feature_y = torch.add(h_user_y, h_user)
z_x = F.relu(feature_x)
z_y = F.relu(feature_y)
# decoding for x and y domain
preds_x = self.decoder_x(z_x)
preds_y = self.decoder_y(z_y)
return preds_x, preds_y, feature_x, feature_y, distance
def get_user_embedding(self, batch_user_x, batch_user_y):
# this is for SIGIR's experiment
h_user_x = self.encoder_x(self.dropout(batch_user_x))
h_user_y = self.encoder_y(self.dropout(batch_user_y))
return h_user_x, h_user_y
def get_latent_z(self, batch_user, batch_user_x, batch_user_y):
# this is for clustering visualization
h_user_x = self.encoder_x(self.dropout(batch_user_x))
h_user_y = self.encoder_y(self.dropout(batch_user_y))
h_user = self.user_embeddings(batch_user)
feature_x = torch.add(h_user_x, h_user)
feature_y = torch.add(h_user_y, h_user)
z_x = F.relu(feature_x)
z_y = F.relu(feature_y)
return z_x, z_y
class Discriminator(nn.Module):
def __init__(self, n_fts, dropout):
super(Discriminator, self).__init__()
self.dropout = dropout
self.training = True
self.disc = nn.Sequential(
nn.Linear(n_fts, int(n_fts / 2)),
nn.ReLU(),
nn.Dropout(dropout),
nn.Linear(int(n_fts / 2), 1))
def forward(self, x):
# make mlp for discriminator
h = self.disc(x)
return h
def save_embedding_process(model, save_loader, feed_data, is_cuda):
fts1 = feed_data['fts1']
fts2 = feed_data['fts2']
user_embedding1_list = []
user_embedding2_list = []
model.eval()
for batch_idx, data in enumerate(save_loader):
data = data.reshape([-1])
val_user_arr = data.numpy()
v_item1 = fts1[val_user_arr]
v_item2 = fts2[val_user_arr]
if is_cuda:
v_user = torch.LongTensor(val_user_arr).cuda()
v_item1 = torch.FloatTensor(v_item1).cuda()
v_item2 = torch.FloatTensor(v_item2).cuda()
else:
v_user = torch.LongTensor(val_user_arr)
v_item1 = torch.FloatTensor(v_item1)
v_item2 = torch.FloatTensor(v_item2)
res = model.get_user_embedding(v_item1, v_item2)
user_embedding1 = res[0]
user_embedding2 = res[1]
if is_cuda:
user_embedding1 = user_embedding1.detach().cpu().numpy()
user_embedding2 = user_embedding2.detach().cpu().numpy()
else:
user_embedding1 = user_embedding1.detach().numpy()
user_embedding2 = user_embedding2.detach().numpy()
user_embedding1_list.append(user_embedding1)
user_embedding2_list.append(user_embedding2)
return np.concatenate(user_embedding1_list, axis=0), np.concatenate(user_embedding2_list, axis=0)
def save_z_process(model, save_loader, feed_data, is_cuda):
fts1 = feed_data['fts1']
fts2 = feed_data['fts2']
user_embedding1_list = []
user_embedding2_list = []
model.eval()
for batch_idx, data in enumerate(save_loader):
data = data.reshape([-1])
val_user_arr = data.numpy()
v_item1 = fts1[val_user_arr]
v_item2 = fts2[val_user_arr]
if is_cuda:
v_user = torch.LongTensor(val_user_arr).cuda()
v_item1 = torch.FloatTensor(v_item1).cuda()
v_item2 = torch.FloatTensor(v_item2).cuda()
else:
v_user = torch.LongTensor(val_user_arr)
v_item1 = torch.FloatTensor(v_item1)
v_item2 = torch.FloatTensor(v_item2)
res = model.get_latent_z(v_user, v_item1, v_item2)
user_embedding1 = res[0]
user_embedding2 = res[1]
if is_cuda:
user_embedding1 = user_embedding1.detach().cpu().numpy()
user_embedding2 = user_embedding2.detach().cpu().numpy()
else:
user_embedding1 = user_embedding1.detach().numpy()
user_embedding2 = user_embedding2.detach().numpy()
user_embedding1_list.append(user_embedding1)
user_embedding2_list.append(user_embedding2)
return np.concatenate(user_embedding1_list, axis=0), np.concatenate(user_embedding2_list, axis=0)