-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrun.py
209 lines (168 loc) · 9.57 KB
/
run.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
import torch
import torch.optim as optim
from progressbar import *
import sys
import time
from OurModel import OurModel
from utils.parser import *
from utils.helpers import *
from utils.load_data import *
def train():
print("************************* Run with following settings ***************************")
print(args)
print("************************************************************************************")
device = torch.device("cuda:1" if torch.cuda.is_available() else "cpu")
filepath = args.data_path + args.dataset + '/{}_{}'.format(args.domain_1, args.domain_2)
n_users_1, n_items_1, user_ratings_1, n_interactions_1 = load_data(filepath=filepath+'/{}_all_item_list.dat'.format(args.domain_1))
n_users_2, n_items_2, user_ratings_2, n_interactions_2 = load_data(filepath=filepath+'/{}_all_item_list.dat'.format(args.domain_2))
print(n_users_1, n_items_1, n_interactions_1)
print(n_users_2, n_items_2, n_interactions_2)
user_ratings_test_1 = generate_test(user_ratings_1)
user_ratings_test_2 = generate_test(user_ratings_2)
config = dict()
config['n_users_1'] = n_users_1
config['n_items_1'] = n_items_1
config['n_users_2'] = n_users_2
config['n_items_2'] = n_items_2
"""
*********************************************************
Generate the Laplacian matrix, where each entry defines the decay factor (e.g., p_ui) between two connected nodes.
"""
plain_adj_1, norm_adj_1, mean_ad_1, pre_adj_1 = get_adj_mat(filepath=filepath, dataset=args.domain_1, \
n_users=n_users_1, n_items=n_items_1, user_ratings=user_ratings_1,\
user_ratings_test=user_ratings_test_1)
plain_adj_2, norm_adj_2, mean_adj_2, pre_adj_2 = get_adj_mat(filepath=filepath, dataset=args.domain_2, \
n_users=n_users_2, n_items=n_items_2, user_ratings=user_ratings_2,\
user_ratings_test=user_ratings_test_2)
all_h_list_1, all_t_list_1, all_v_list_1 = load_adjacency_list_data(plain_adj_1)
all_h_list_2, all_t_list_2, all_v_list_2 = load_adjacency_list_data(plain_adj_2)
A_values_init_1 = create_initial_A_values(args.n_factors, all_v_list_1)
A_values_init_2 = create_initial_A_values(args.n_factors, all_v_list_2)
config['norm_adj_1'] = plain_adj_1
config['norm_adj_2'] = plain_adj_2
config['all_h_list_1'] = all_h_list_1
config['all_h_list_2'] = all_h_list_2
config['all_t_list_1'] = all_t_list_1
config['all_t_list_2'] = all_t_list_2
t0 = time.time()
"""
***********************************************************
pretrain = 1: load embeddings with name such as embedding_xxx(.npz), l2_best_model(.npz)
pretrain = 0: default value, no pretrained embeddings.
"""
if args.pretrain == 1:
print("Try to load pretrain: ", args.embed_name)
pretrain_data = load_best(name=args.embed_name)
if pretrain_data == None:
print("Load pretrained model(%s)fail!" % (args.embed_name))
else:
pretrain_data = None
model = OurModel(data_config=config, pretrain_data=pretrain_data, args=args, device=device).to(device)
optimizer = optim.Adam(model.parameters(), lr=args.lr)
"""
*********************************************************
Train
"""
best_ret_1 = np.array([0] * 6)
best_ret_2 = np.array([0] * 6)
for epoch in range(args.epoch):
t1 = time.time()
loss, mf_loss, emb_loss, cor_loss = 0., 0., 0., 0.
bar_length = 300
widgets = ['Train: ', Percentage(), ' ', Bar('#'), ' ', ETA()]
pbar = ProgressBar(widgets=widgets, maxval=bar_length).start()
for idx in range(bar_length):
uij_1, uij_2 = generate_train_batch_for_all_overlap(user_ratings_1, user_ratings_test_1, n_items_1,
user_ratings_2, user_ratings_test_2, n_items_2,
batch_size=args.batch_size)
batch_loss = model(uij_1[:, 0], uij_1[:, 1], uij_1[:, 2], uij_2[:, 0], uij_2[:, 1], uij_2[:, 2])
optimizer.zero_grad()
batch_loss.backward()
optimizer.step()
loss += batch_loss.item() / bar_length
pbar.update(idx)
if np.isnan(loss) == True:
print('ERROR: loss is nan.')
sys.exit()
perf_str = 'Epoch %d [%.1fs]: train==[%.5f]' % (epoch, time.time() - t1, loss)
print(perf_str)
user_count_1 = 0
user_count_2 = 0
ret_1 = np.array([0.0] * 6)
ret_2 = np.array([0.0] * 6)
test_widgets = ['Test: ', Percentage(), ' ', Bar('#'), ' ', ETA()]
test_pbar = ProgressBar(widgets=test_widgets, maxval=n_users_1).start()
test_index = 0
for t_uij_1, t_uij_2 in generate_test_batch_for_all_overlap(user_ratings_1, user_ratings_test_1,
n_items_1, user_ratings_2,
user_ratings_test_2, n_items_2):
model.get_embeddings(t_uij_1[:, 0], t_uij_1[:, 1], t_uij_1[:, 2], t_uij_2[:, 0], t_uij_2[:, 1], t_uij_2[:, 2])
u_g_embeddings_1, u_g_embeddings_2 = model.u_g_embeddings_1, model.u_g_embeddings_2
pos_i_g_embeddings_1, pos_i_g_embeddings_2 = model.pos_i_g_embeddings_1, model.pos_i_g_embeddings_2
neg_i_g_embeddings_1, neg_i_g_embeddings_2 = model.neg_i_g_embeddings_1, model.neg_i_g_embeddings_2
pos_s_1 = torch.squeeze(model.rating(u_g_embeddings_1, pos_i_g_embeddings_1)).detach().cpu()
neg_s_1 = torch.squeeze(model.rating(u_g_embeddings_1, neg_i_g_embeddings_1)).detach().cpu()
pos_s_2 = torch.squeeze(model.rating(u_g_embeddings_2, pos_i_g_embeddings_2)).detach().cpu()
neg_s_2 = torch.squeeze(model.rating(u_g_embeddings_2, neg_i_g_embeddings_2)).detach().cpu()
user_count_1 += 1
predictions_1 = [pos_s_1[0]]
predictions_1 += neg_s_1
predictions_1 = [-1 * i for i in predictions_1]
rank_1 = np.array(predictions_1).argsort().argsort()[0]
if rank_1 < 2:
ret_1[0] += 1
ret_1[3] += 1 / np.log2(rank_1 + 2)
if rank_1 < 5:
ret_1[1] += 1
ret_1[4] += 1 / np.log2(rank_1 + 2)
if rank_1 < 10:
ret_1[2] += 1
ret_1[5] += 1 / np.log2(rank_1 + 2)
user_count_2 += 1
predictions_2 = [pos_s_2[0]]
predictions_2 += neg_s_2
predictions_2 = [-1 * i for i in predictions_2]
rank_2 = np.array(predictions_2).argsort().argsort()[0]
if rank_2 < 2:
ret_2[0] += 1
ret_2[3] += 1 / np.log2(rank_2 + 2)
if rank_2 < 5:
ret_2[1] += 1
ret_2[4] += 1 / np.log2(rank_2 + 2)
if rank_2 < 10:
ret_2[2] += 1
ret_2[5] += 1 / np.log2(rank_2 + 2)
test_pbar.update(test_index)
test_index += 1
best_ret_1 = best_result(best_ret_1, ret_1)
best_ret_2 = best_result(best_ret_2, ret_2)
print('%s: HR_2 %f HR_5 %f HR_10 %f'
% (args.domain_1, ret_1[0]/user_count_1, ret_1[1]/user_count_1, ret_1[2]/user_count_1))
print('%s: NDCG_2 %f NDCG_5 %f NDCG_10 %f'
% (args.domain_1, ret_1[3] / user_count_1, ret_1[4] / user_count_1, ret_1[5] / user_count_1))
print('Best HitRatio for %s: HR_2 %f HR_5 %f HR_10 %f'
% (args.domain_1, best_ret_1[0]/user_count_1, best_ret_1[1]/user_count_1,
best_ret_1[2]/user_count_1))
print('Best NDCG for %s: NDCG_2 %f NDCG_5 %f NDCG_10 %f'
% (args.domain_1, best_ret_1[3]/user_count_1, best_ret_1[4]/user_count_1,
best_ret_1[5]/user_count_1))
if ret_1[0] == best_ret_1[0] or ret_1[1] == best_ret_1[1] or ret_1[2] == best_ret_1[2] \
or ret_1[3] == best_ret_1[3] or ret_1[4] == best_ret_1[4] or ret_1[5] == best_ret_1[5]:
model_save(model.embedding_dict, args.weights_path, args, savename='best_model')
print('save the weights in path: ', args.weights_path)
print('%s: HR_2 %f HR_5 %f HR_10 %f'
% (args.domain_2, ret_2[0]/user_count_2, ret_2[1]/user_count_2, ret_2[2]/user_count_2))
print('%s: NDCG_2 %f NDCG_5 %f NDCG_10 %f'
% (args.domain_2, ret_2[3] / user_count_2, ret_2[4] / user_count_2, ret_2[5] / user_count_2))
print('Best HitRatio for %s: HR_2 %f HR_5 %f HR_10 %f'
% (args.domain_2, best_ret_2[0]/user_count_2, best_ret_2[1]/user_count_2,
best_ret_2[2]/user_count_2))
print('Best NDCG for %s: NDCG_2 %f NDCG_5 %f NDCG_10 %f'
% (args.domain_2, best_ret_2[3]/user_count_2, best_ret_2[4]/user_count_2,
best_ret_2[5]/user_count_2))
if ret_2[0] == best_ret_2[0] or ret_2[1] == best_ret_2[1] or ret_2[2] == best_ret_2[2] \
or ret_2[3] == best_ret_2[3] or ret_2[4] == best_ret_2[4] or ret_2[5] == best_ret_2[5]:
model_save(model.embedding_dict, args.weights_path, args, savename='best_model')
print('save the weights in path: ', args.weights_path)
if __name__ == '__main__':
train()