/
directau.py
183 lines (149 loc) · 6.94 KB
/
directau.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
# -*- coding: utf-8 -*-
import numpy as np
import scipy.sparse as sp
import torch
import torch.nn as nn
import torch.nn.functional as F
from recbole.model.abstract_recommender import GeneralRecommender
from recbole.model.init import xavier_normal_initialization
from recbole.utils import InputType
class DirectAU(GeneralRecommender):
input_type = InputType.POINTWISE
def __init__(self, config, dataset):
super(DirectAU, self).__init__(config, dataset)
# load parameters info
self.embedding_size = config['embedding_size']
self.gamma = config['gamma']
self.encoder_name = config['encoder']
# define layers and loss
if self.encoder_name == 'MF':
self.encoder = MFEncoder(self.n_users, self.n_items, self.embedding_size)
elif self.encoder_name == 'LightGCN':
self.n_layers = config['n_layers']
self.interaction_matrix = dataset.inter_matrix(form='coo').astype(np.float32)
self.norm_adj = self.get_norm_adj_mat().to(self.device)
self.encoder = LGCNEncoder(self.n_users, self.n_items, self.embedding_size, self.norm_adj, self.n_layers)
else:
raise ValueError('Non-implemented Encoder.')
# storage variables for full sort evaluation acceleration
self.restore_user_e = None
self.restore_item_e = None
# parameters initialization
self.apply(xavier_normal_initialization)
def get_norm_adj_mat(self):
# build adj matrix
A = sp.dok_matrix((self.n_users + self.n_items, self.n_users + self.n_items), dtype=np.float32)
inter_M = self.interaction_matrix
inter_M_t = self.interaction_matrix.transpose()
data_dict = dict(zip(zip(inter_M.row, inter_M.col + self.n_users), [1] * inter_M.nnz))
data_dict.update(dict(zip(zip(inter_M_t.row + self.n_users, inter_M_t.col), [1] * inter_M_t.nnz)))
A._update(data_dict)
# norm adj matrix
sumArr = (A > 0).sum(axis=1)
# add epsilon to avoid divide by zero Warning
diag = np.array(sumArr.flatten())[0] + 1e-7
diag = np.power(diag, -0.5)
D = sp.diags(diag)
L = D * A * D
# covert norm_adj matrix to tensor
L = sp.coo_matrix(L)
row = L.row
col = L.col
i = torch.LongTensor([row, col])
data = torch.FloatTensor(L.data)
SparseL = torch.sparse.FloatTensor(i, data, torch.Size(L.shape))
return SparseL
def forward(self, user, item):
user_e, item_e = self.encoder(user, item)
return F.normalize(user_e, dim=-1), F.normalize(item_e, dim=-1)
@staticmethod
def alignment(x, y, alpha=2):
return (x - y).norm(p=2, dim=1).pow(alpha).mean()
@staticmethod
def uniformity(x, t=2):
return torch.pdist(x, p=2).pow(2).mul(-t).exp().mean().log()
def calculate_loss(self, interaction):
if self.restore_user_e is not None or self.restore_item_e is not None:
self.restore_user_e, self.restore_item_e = None, None
user = interaction[self.USER_ID]
item = interaction[self.ITEM_ID]
user_e, item_e = self.forward(user, item)
align = self.alignment(user_e, item_e)
uniform = self.gamma * (self.uniformity(user_e) + self.uniformity(item_e)) / 2
return align + uniform
def predict(self, interaction):
user = interaction[self.USER_ID]
item = interaction[self.ITEM_ID]
user_e = self.user_embedding(user)
item_e = self.item_embedding(item)
return torch.mul(user_e, item_e).sum(dim=1)
def full_sort_predict(self, interaction):
user = interaction[self.USER_ID]
if self.encoder_name == 'LightGCN':
if self.restore_user_e is None or self.restore_item_e is None:
self.restore_user_e, self.restore_item_e = self.encoder.get_all_embeddings()
user_e = self.restore_user_e[user]
all_item_e = self.restore_item_e
else:
user_e = self.encoder.user_embedding(user)
all_item_e = self.encoder.item_embedding.weight
score = torch.matmul(user_e, all_item_e.transpose(0, 1))
return score.view(-1)
# def save_params(self):
# user_embeddings, item_embeddings = self.encoder.get_all_embeddings()
# np.save('user-DirectAU.npy', user_embeddings.data.cpu().numpy())
# np.save('item-DirectAU.npy', item_embeddings.data.cpu().numpy())
# def check(self, interaction):
# user = interaction[self.USER_ID]
# item = interaction[self.ITEM_ID]
# user_e, item_e = self.forward(user, item)
#
# user_e = user_e.detach()
# item_e = item_e.detach()
#
# alignment_loss = self.alignment(user_e, item_e)
# uniform_loss = (self.uniformity(user_e) + self.uniformity(item_e)) / 2
#
# return alignment_loss, uniform_loss
class MFEncoder(nn.Module):
def __init__(self, user_num, item_num, emb_size):
super(MFEncoder, self).__init__()
self.user_embedding = nn.Embedding(user_num, emb_size)
self.item_embedding = nn.Embedding(item_num, emb_size)
def forward(self, user_id, item_id):
u_embed = self.user_embedding(user_id)
i_embed = self.item_embedding(item_id)
return u_embed, i_embed
def get_all_embeddings(self):
user_embeddings = self.user_embedding.weight
item_embeddings = self.item_embedding.weight
return user_embeddings, item_embeddings
class LGCNEncoder(nn.Module):
def __init__(self, user_num, item_num, emb_size, norm_adj, n_layers=3):
super(LGCNEncoder, self).__init__()
self.n_users = user_num
self.n_items = item_num
self.n_layers = n_layers
self.norm_adj = norm_adj
self.user_embedding = torch.nn.Embedding(user_num, emb_size)
self.item_embedding = torch.nn.Embedding(item_num, emb_size)
def get_ego_embeddings(self):
user_embeddings = self.user_embedding.weight
item_embeddings = self.item_embedding.weight
ego_embeddings = torch.cat([user_embeddings, item_embeddings], dim=0)
return ego_embeddings
def get_all_embeddings(self):
all_embeddings = self.get_ego_embeddings()
embeddings_list = [all_embeddings]
for layer_idx in range(self.n_layers):
all_embeddings = torch.sparse.mm(self.norm_adj, all_embeddings)
embeddings_list.append(all_embeddings)
lightgcn_all_embeddings = torch.stack(embeddings_list, dim=1)
lightgcn_all_embeddings = torch.mean(lightgcn_all_embeddings, dim=1)
user_all_embeddings, item_all_embeddings = torch.split(lightgcn_all_embeddings, [self.n_users, self.n_items])
return user_all_embeddings, item_all_embeddings
def forward(self, user_id, item_id):
user_all_embeddings, item_all_embeddings = self.get_all_embeddings()
u_embed = user_all_embeddings[user_id]
i_embed = item_all_embeddings[item_id]
return u_embed, i_embed