-
Notifications
You must be signed in to change notification settings - Fork 7
/
BPR.py
267 lines (215 loc) · 10.6 KB
/
BPR.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
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
import tensorflow as tf
import time
import numpy as np
import os
import copy
import pickle
import argparse
import utility
import pandas as pd
from sklearn.metrics import *
class BPR:
def __init__(self, sess, args, train_df, vali_df
, key_genre, item_genre_list, user_genre_count):
self.dataname = args.dataname
self.key_genre = key_genre
self.item_genre_list = item_genre_list
self.user_genre_count = user_genre_count
self.sess = sess
self.args = args
self.num_cols = len(train_df['item_id'].unique())
self.num_rows = len(train_df['user_id'].unique())
self.hidden_neuron = args.hidden_neuron
self.neg = args.neg
self.batch_size = args.batch_size
self.train_df = train_df
self.vali_df = vali_df
self.num_train = len(self.train_df)
self.num_vali = len(self.vali_df)
self.train_epoch = args.train_epoch
self.lr = args.lr # learning rate
self.optimizer_method = args.optimizer_method
self.display_step = args.display_step
self.num_genre = args.num_genre
self.reg = args.reg # regularization term trade-off
print('**********BPR**********')
print(self.args)
self._prepare_model()
def run(self):
init = tf.global_variables_initializer()
self.sess.run(init)
for epoch_itr in range(1, self.train_epoch + 1):
self.train_model(epoch_itr)
if epoch_itr % self.display_step == 0:
self.test_model(epoch_itr)
return self.make_records()
def _prepare_model(self):
with tf.name_scope("input_data"):
self.user_input = tf.placeholder(tf.int32, shape=[None, 1], name="user_input")
self.item_input_pos = tf.placeholder(tf.int32, shape=[None, 1], name="item_input_pos")
self.item_input_neg = tf.placeholder(tf.int32, shape=[None, 1], name="item_input_neg")
with tf.variable_scope("BPR", reuse=tf.AUTO_REUSE):
self.P = tf.get_variable(name="P", initializer=tf.truncated_normal(shape=[self.num_rows, self.hidden_neuron],
mean=0, stddev=0.03), dtype=tf.float32)
self.Q = tf.get_variable(name="Q", initializer=tf.truncated_normal(shape=[self.num_cols, self.hidden_neuron],
mean=0, stddev=0.03), dtype=tf.float32)
self.saver = tf.train.Saver([self.P, self.Q])
p = tf.reduce_sum(tf.nn.embedding_lookup(self.P, self.user_input), 1)
q_neg = tf.reduce_sum(tf.nn.embedding_lookup(self.Q, self.item_input_neg), 1)
q_pos = tf.reduce_sum(tf.nn.embedding_lookup(self.Q, self.item_input_pos), 1)
predict_pos = (tf.reduce_sum(p * q_pos, 1))
predict_neg = (tf.reduce_sum(p * q_neg, 1))
cost1 = tf.reduce_sum(tf.nn.softplus(-(predict_pos - predict_neg)))
cost2 = self.reg * 0.5 * (self.l2_norm(self.P) + self.l2_norm(self.Q)) # regularization term
self.cost = cost1 + cost2 # the loss function
if self.optimizer_method == "Adam":
optimizer = tf.train.AdamOptimizer(self.lr)
elif self.optimizer_method == "Adadelta":
optimizer = tf.train.AdadeltaOptimizer(self.lr)
elif self.optimizer_method == "Adagrad":
optimizer = tf.train.AdadeltaOptimizer(self.lr)
elif self.optimizer_method == "RMSProp":
optimizer = tf.train.RMSPropOptimizer(self.lr)
elif self.optimizer_method == "GradientDescent":
optimizer = tf.train.GradientDescentOptimizer(self.lr)
elif self.optimizer_method == "Momentum":
optimizer = tf.train.MomentumOptimizer(self.lr, 0.9)
else:
raise ValueError("Optimizer Key ERROR")
with tf.variable_scope("Optimizer", reuse=tf.AUTO_REUSE):
self.optimizer = optimizer.minimize(self.cost)
def train_model(self, itr):
NS_start_time = time.time() * 1000.0
epoch_cost = 0
num_sample, user_list, item_pos_list, item_neg_list = utility.negative_sample(self.train_df, self.num_rows,
self.num_cols, self.neg)
NS_end_time = time.time() * 1000.0
start_time = time.time() * 1000.0
num_batch = int(len(user_list) / float(self.batch_size)) + 1
random_idx = np.random.permutation(len(user_list))
for i in range(num_batch):
# get the indices of the current batch
if i == num_batch - 1:
batch_idx = random_idx[i * self.batch_size:]
elif i < num_batch - 1:
batch_idx = random_idx[(i * self.batch_size):((i + 1) * self.batch_size)]
_, tmp_cost = self.sess.run( # do the optimization by the minibatch
[self.optimizer, self.cost],
feed_dict={self.user_input: user_list[batch_idx, :],
self.item_input_pos: item_pos_list[batch_idx, :],
self.item_input_neg: item_neg_list[batch_idx, :]})
epoch_cost += tmp_cost
if itr % self.display_step == 0:
print ("Training //", "Epoch %d //" % itr, " Total cost = {:.5f}".format(epoch_cost),
"Training time : %d ms" % (time.time() * 1000.0 - start_time),
"negative Sampling time : %d ms" % (NS_end_time - NS_start_time),
"negative samples : %d" % (num_sample))
ckpt_save_path = "./"+self.dataname+"/BPR_check_points"
if not os.path.exists(ckpt_save_path):
os.makedirs(ckpt_save_path)
self.saver.save(sess, ckpt_save_path + "/check_point.ckpt", global_step=itr)
def test_model(self, itr): # calculate the cost and rmse of testing set in each epoch
if itr % self.display_step == 0:
start_time = time.time() * 1000.0
P, Q = self.sess.run([self.P, self.Q])
Rec = np.matmul(P, Q.T)
[precision, recall, f_score, NDCG] = utility.test_model_all(Rec, self.vali_df, self.train_df)
utility.ranking_analysis(Rec, self.vali_df, self.train_df, self.key_genre, self.item_genre_list,
self.user_genre_count)
print (
"Testing //", "Epoch %d //" % itr,
"Testing time : %d ms" % (time.time() * 1000.0 - start_time))
print("=" * 200)
def make_records(self): # record all the results' details into files
P, Q = self.sess.run([self.P, self.Q])
Rec = np.matmul(P, Q.T)
[precision, recall, f_score, NDCG] = utility.test_model_all(Rec, self.vali_df, self.train_df)
return precision, recall, f_score, NDCG, Rec
@staticmethod
def l2_norm(tensor):
return tf.reduce_sum(tf.square(tensor))
parser = argparse.ArgumentParser(description='BPR')
parser.add_argument('--train_epoch', type=int, default=20)
parser.add_argument('--display_step', type=int, default=1)
parser.add_argument('--lr', type=float, default=0.01)
parser.add_argument('--reg', type=float, default=0.1)
parser.add_argument('--optimizer_method', choices=['Adam', 'Adadelta', 'Adagrad', 'RMSProp', 'GradientDescent',
'Momentum'], default='Adam')
parser.add_argument('--hidden_neuron', type=int, default=20)
parser.add_argument('--n', type=int, default=1)
parser.add_argument('--neg', type=int, default=5)
parser.add_argument('--batch_size', type=int, default=1024)
parser.add_argument('--dataname', nargs='?', default='ml1m-6')
args = parser.parse_args()
dataname = args.dataname
train_df = pickle.load(open('./' + dataname + '/training_df.pkl'))
vali_df = pickle.load(open('./' + dataname + '/valiing_df.pkl')) # for validation
# vali_df = pickle.load(open('./' + dataname + '/testing_df.pkl')) # for testing
key_genre = pickle.load(open('./' + dataname + '/key_genre.pkl'))
item_idd_genre_list = pickle.load(open('./' + dataname + '/item_idd_genre_list.pkl'))
genre_item_vector = pickle.load(open('./' + dataname + '/genre_item_vector.pkl'))
genre_count = pickle.load(open('./' + dataname + '/genre_count.pkl'))
user_genre_count = pickle.load(open('./' + dataname + '/user_genre_count.pkl'))
num_item = len(train_df['item_id'].unique())
num_user = len(train_df['user_id'].unique())
num_genre = len(key_genre)
args.num_genre = num_genre
item_genre_list = []
for u in range(num_item):
gl = item_idd_genre_list[u]
tmp = []
for g in gl:
if g in key_genre:
tmp.append(g)
item_genre_list.append(tmp)
print('!' * 100)
print('number of positive feedback: ' + str(len(train_df)))
print('estimated number of training samples: ' + str(args.neg * len(train_df)))
print('!' * 100)
# genreate item_genre matrix
genre_item_indicator = np.zeros((num_genre, num_item))
for k in range(num_genre):
genre_item_indicator[k, :] = genre_item_vector[key_genre[k]]
precision = np.zeros(4)
recall = np.zeros(4)
f1 = np.zeros(4)
ndcg = np.zeros(4)
RSP = np.zeros(4)
REO = np.zeros(4)
n = args.n
for i in range(n):
with tf.Session() as sess:
bpr = BPR(sess, args, train_df, vali_df, key_genre, item_genre_list, user_genre_count)
[prec_one, rec_one, f_one, ndcg_one, Rec] = bpr.run()
[RSP_one, REO_one] = utility.ranking_analysis(Rec, vali_df, train_df, key_genre, item_genre_list,
user_genre_count)
precision += prec_one
recall += rec_one
f1 += f_one
ndcg += ndcg_one
RSP += RSP_one
REO += REO_one
with open('Rec_' + dataname + '_BPR.mat', "wb") as f:
np.save(f, Rec)
precision /= n
recall /= n
f1 /= n
ndcg /= n
RSP /= n
REO /= n
print('')
print('*' * 100)
print('Averaged precision@1\t%.7f,\t||\tprecision@5\t%.7f,\t||\tprecision@10\t%.7f,\t||\tprecision@15\t%.7f' \
% (precision[0], precision[1], precision[2], precision[3]))
print('Averaged recall@1\t%.7f,\t||\trecall@5\t%.7f,\t||\trecall@10\t%.7f,\t||\trecall@15\t%.7f' \
% (recall[0], recall[1], recall[2], recall[3]))
print('Averaged f1@1\t\t%.7f,\t||\tf1@5\t\t%.7f,\t||\tf1@10\t\t%.7f,\t||\tf1@15\t\t%.7f' \
% (f1[0], f1[1], f1[2], f1[3]))
print('Averaged NDCG@1\t\t%.7f,\t||\tNDCG@5\t\t%.7f,\t||\tNDCG@10\t\t%.7f,\t||\tNDCG@15\t\t%.7f' \
% (ndcg[0], ndcg[1], ndcg[2], ndcg[3]))
print('*' * 100)
print('Averaged RSP @1\t%.7f\t||\t@5\t%.7f\t||\t@10\t%.7f\t||\t@15\t%.7f' \
% (RSP[0], RSP[1], RSP[2], RSP[3]))
print('Averaged REO @1\t%.7f\t||\t@5\t%.7f\t||\t@10\t%.7f\t||\t@15\t%.7f' \
% (REO[0], REO[1], REO[2], REO[3]))
print('*' * 100)