-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathtrain.py
200 lines (176 loc) · 9.64 KB
/
train.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
import torch
import torch.nn as nn
import sys
import data_loader
import MetaProperty
import math
import numpy as np
import pandas as pd
import pickle
import gzip
import time
from datetime import datetime
from torch.autograd import Variable
from sklearn.metrics import mean_absolute_error
from sklearn.metrics import mean_squared_error
from math import sqrt
from utils.loss import bpr_loss
from utils.evaluation import evaluate
from utils.sample import sample_items
def to_var(x, volatile=False):
if torch.cuda.is_available():
x = x.cuda()
return Variable(x, volatile=volatile)
def limit(x):
return (x - (torch.min(x))) / (torch.max(x) - torch.min(x)) + 1e-7
# Hyper Parameters
num_epochs = 30
batch_size = 512
learning_rate = 1e-3
negative_sample_size = 20
n_sample = 10
rank_task_weight = 0.9
print("batch_size: ", batch_size)
print("learning_rate: ", learning_rate)
data_name = 'amazon'
# in ['age', 'length', 'rate', 'polar_sentiment', 'helpful', 'polar_helpful']
review_property = 'all'
print("Loading data...", data_name)
print("Review property: ", review_property)
if data_name == 'yelp':
num_users = 47154
num_items = 16531
train_loader, train_dataset = data_loader.get_loader('yelp_train.p', data_name, review_property, batch_size=batch_size)
val_loader, val_dataset = data_loader.get_loader('yelp_valid.p', data_name, review_property, batch_size=batch_size, shuffle=False)
test_loader, test_dataset = data_loader.get_loader('yelp_test.p', data_name, review_property, batch_size=batch_size, shuffle=False)
elif data_name == 'amazon':
num_users = 26010
num_items = 16514
train_loader, train_dataset = data_loader.get_loader('amazon_train.p', data_name, review_property, batch_size=batch_size)
val_loader, val_dataset = data_loader.get_loader('amazon_valid.p', data_name, review_property, batch_size=batch_size, shuffle=False)
test_loader, test_dataset = data_loader.get_loader('amazon_test.p', data_name, review_property, batch_size=batch_size, shuffle=False)
print("train/val/test/: {:d}/{:d}/{:d}".format(len(train_loader), len(val_loader),len(test_loader)))
print("==================================================================================")
meta_property = MetaProperty.MetaProperty(data_name)
if torch.cuda.is_available():
meta_property.cuda()
optimizer = torch.optim.Adam(meta_property.parameters(), lr=learning_rate)
print("==================================================================================")
print("Training Start..", datetime.now())
batch_loss = 0
batch_rat_loss = 0
batch_rank_loss = 0
# Train the Model
total_step = len(train_loader)
best_p1 = 0.0
best_p5 = 0.0
best_p10 = 0.0
best_r1 = 0.0
best_r5 = 0.0
best_r10 = 0.0
best_map = 0.0
# KL_loss = nn.KLDivLoss(reduction='batchmean')
for epoch in range(num_epochs):
for idx, (user_id, item_id, user_review, user_review_scores, item_review, item_review_scores) in enumerate(train_loader):
user_id = to_var(user_id)
item_id = to_var(item_id)
user_review = to_var(user_review)
user_review_scores = to_var(user_review_scores)
item_review = to_var(item_review)
item_review_scores = to_var(item_review_scores)
rank_pos_outputs = meta_property(user_id, item_id, user_review, user_review_scores,
item_review, item_review_scores)
for i in range(n_sample):
sample_item_id, item_review, item_review_scores = train_dataset.get_neg_items(len(user_id))
sample_item_id = to_var(torch.from_numpy(np.array(sample_item_id)))
item_review = to_var(item_review)
item_review_scores = to_var(item_review_scores)
rank_neg_outputs = meta_property(user_id, sample_item_id, user_review, user_review_scores,
item_review, item_review_scores)
# Cos-UU loss
# rank_loss = rank_loss + bpr_loss(rank_pos_outputs, rank_neg_outputs) + (1.0 / n_sample) *
# torch.mean(cos(limit(meta_property.user_prop_pref(user_id)),
# limit(meta_property.item_prop_pref(sample_item_id))))
# Cos-UI loss
# rank_loss = rank_loss + bpr_loss(rank_pos_outputs, rank_neg_outputs) + (1.0 / n_sample) * torch.mean(
# cos(limit(meta_property.item_prop_pref(item_id)), limit(meta_property.item_prop_pref(sample_item_id))))
# KL-UU Loss
# rank_loss = rank_loss + bpr_loss(rank_pos_outputs, rank_neg_outputs) - (1.0 / n_sample) * KL_loss(
# limit(meta_property.user_prop_pref(user_id)), limit(meta_property.item_prop_pref(sample_item_id)))
# KL-UI Loss
# rank_loss = rank_loss + bpr_loss(rank_pos_outputs, rank_neg_outputs) -
# (1.0 / n_sample) * KL_loss(limit(meta_property.item_prop_pref(item_id)),
# limit(meta_property.item_prop_pref(sample_item_id)))
if i == 0:
rank_loss = bpr_loss(rank_pos_outputs, rank_neg_outputs)
else:
rank_loss += bpr_loss(rank_pos_outputs, rank_neg_outputs)
loss = rank_loss
optimizer.zero_grad()
loss.backward()
optimizer.step()
if idx % 100 == 0:
print(" loss: {}, time: {}".format(loss, datetime.now()))
print("epoch: ", epoch, datetime.now())
print("==================================================================================")
print("Begin Ranking Prediction..")
with torch.no_grad():
k = [1, 5, 10]
precision = [0.0] * len(k)
recall = [0.0] * len(k)
apks = 0.0
precision_u = [[],[],[]]
recall_u = [[],[],[]]
for i, user in enumerate(range(num_users)):
true_results = test_dataset.get_user_items(user)
user_torch = to_var(torch.LongTensor([user]))
user_emb = meta_property.user_embedding(user_torch)
user_emb = user_emb.squeeze()
user_lv = meta_property.user_latent_vector(user_torch).squeeze()
user_emb = torch.cat((user_emb, user_lv))
item_emb = meta_property.item_embedding.weight.data
item_lv = meta_property.item_latent_vector.weight.data
item_emb = torch.cat((item_emb, item_lv), 1)
pred_long_term_results = torch.mv(item_emb, user_emb)
item_bias = meta_property.item_bias(to_var(torch.LongTensor(range(num_items + 1)))).squeeze()
pred_results = pred_long_term_results + meta_property.user_bias(user_torch).squeeze() + item_bias
# remove visited items
visited_items = train_dataset.get_user_items(user)
pred_results[visited_items] = -np.inf
pred_results[num_items] = -np.inf
sub_precision, sub_recall, apk, sub_precision_u, sub_recall_u = evaluate(true_results, pred_results, ks=k)
precision = [precision[i] + sub_precision[i] for i in range(len(k))]
recall = [recall[i] + sub_recall[i] for i in range(len(k))]
apks = apks + apk
if i % 10000 == 0:
print('Tested Users {0}: test result precision@1 {1:.5f}, recall@1 {2:.5f}, precision@5 {3:.5f},'
'recall@5 {4:.5f}, precision@10 {5:.5f}, recall@10 {6:.5f}, MAP {7:.5f}, time: {8}'.format(i, precision[0] / num_users,
recall[0] / num_users, precision[1] / num_users,
recall[1] / num_users, precision[2] / num_users,
recall[2] / num_users, apks / num_users, datetime.now()))
print('Tested Users {0}: test result precision@1 {1:.5f}, recall@1 {2:.5f}, precision@5 {3:.5f},'
'recall@5 {4:.5f}, precision@10 {5:.5f}, recall@10 {6:.5f}, MAP {7:.5f}, time: {8}'.format(i, precision[0] / num_users,
recall[0] / num_users, precision[1] / num_users,
recall[1] / num_users, precision[2] / num_users,
recall[2] / num_users, apks / num_users, datetime.now()))
# if epoch < 5:
# pickle.dump([precision, recall], gzip.open("epoch_{0}_results.p",'wb'))
if (apks / num_users) > best_map:
best_p1 = precision[0] / num_users
best_p5 = precision[1] / num_users
best_p10 = precision[2] / num_users
best_r1 = recall[0] / num_users
best_r5 = recall[1] / num_users
best_r10 = recall[2] / num_users
best_map = apks / num_users
print("==================================================================================")
print("Testing End..")
print('Best Results after {0} epochs: test result precision@1 {1:.5f}, recall@1 {2:.5f}, precision@5 {3:.5f},'
'recall@5 {4:.5f}, precision@10 {5:.5f}, recall@10 {6:.5f}, MAP {7:.5f}, time: {8}'.format(num_epochs, best_p1, best_r1, best_p5, best_r5,
best_p10, best_r10, best_map, datetime.now()))
print("==================================================================================")
print("Training End..")
print("batch_size: ", batch_size)
print("learning_rate: ", learning_rate)
print("Loading data...", data_name)
print("Property: ", review_property)