-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathPMF.py
168 lines (143 loc) · 5.92 KB
/
PMF.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
# -*- coding: utf-8 -*-
from __future__ import print_function
import scipy.sparse as sp
import scipy.stats as stats
import heapq
import numpy as np
from numpy.random import RandomState
import math
import time
import os
import re
import copy
from Recommendation import Recommendation, get_dataset_path
class PMF(Recommendation):
def __init__(self, file, train_file, test_file, latent_size):
super(PMF, self).__init__(file, train_file, test_file)
# 一些参数
self.latent_size = latent_size
self.lambda_alpha = 0.01
self.lambda_beta = 0.01
self.iterations = 2000
# self.lr = 3e-5
self.lr = 7e-6
# self.lr = 5e-3
self.momentum = 0.9
self.random_state = RandomState(1)
self.I = copy.deepcopy(self.trainMatrix)
self.I[self.I != 0] = 1
# Initialize factor parameters U0 ,V0
# self.U = np.array([[1,0,0],
# [0,1,0],
# [0,0,1]])
# self.V = np.array([[1,1,1],
# [2,2,2],
# [3,3,3],
# [4,4,4],
# [5,5,5]])
self.U = 0.1 * self.random_state.rand(np.size(self.trainMatrix, 0), self.latent_size)
self.V = 0.1 * self.random_state.rand(np.size(self.trainMatrix, 1), self.latent_size)
# self.U = stats.norm(sp.(self.trainMatrix, 0), self.latent_size)
# self.V = stats.norm(np.size(self.trainMatrix, 1), self.latent_size)
self.train()
# 预测评分矩阵
print('预测评分矩阵：')
self.predict_matrix = np.dot(self.U, self.V.T)
np.save(os.getcwd() + '\\out_file\\PMF\\predictMatrix_' + os.path.basename(train_file) + '_PMF.npy',
self.predict_matrix)
self.predict_matrix = np.load(
os.getcwd() + '\\out_file\\PMF\\predictMatrix_' + os.path.basename(train_file) + '_PMF.npy')
print(self.predict_matrix)
def loss(self):
# the loss function of the model
loss = np.sum(self.I.toarray() * (
self.trainMatrix.toarray() - np.dot(self.U, self.V.T)) ** 2) + self.lambda_alpha * np.sum(
np.square(self.U)) + self.lambda_beta * np.sum(np.square(self.V))
return loss
def predict_test(self):
index_data = np.array(list(self.testMatrix.keys()))
u_features = self.U.take(index_data.take(0, axis=1), axis=0)
v_features = self.V.take(index_data.take(1, axis=1), axis=0)
predict_value = np.sum(u_features * v_features, 1)
'''
print(predict_value)
for i in range(len(predict_value)):
predict_value[i] = 1 / (1+math.exp(predict_value[i]))
print(predict_value)
'''
return predict_value
def train(self):
train_loss_list = []
test_rmse_list = []
last_test_rmse = None
# momentum
momentum_u = np.zeros(self.U.shape)
momentum_v = np.zeros(self.V.shape)
for it in range(self.iterations):
# derivate of Vi
grads_u = np.dot(self.I.toarray() * (self.trainMatrix.toarray() - np.dot(self.U, self.V.T)),
-self.V) + self.lambda_alpha * self.U
# derivate of Tj
grads_v = np.dot((self.I.toarray() * (self.trainMatrix.toarray() - np.dot(self.U, self.V.T))).T,
-self.U) + self.lambda_beta * self.V
# update the parameters
momentum_u = (self.momentum * momentum_u) + self.lr * grads_u
momentum_v = (self.momentum * momentum_v) + self.lr * grads_v
self.U = self.U - momentum_u
self.V = self.V - momentum_v
# training evaluation
train_loss = self.loss()
train_loss_list.append(train_loss)
test_preds = self.predict_test()
test_rmse = self.RMSE(test_preds, np.array(list(self.testMatrix.values())))
test_rmse_list.append(test_rmse)
print('traning iteration:{: d} ,loss:{: f}, test_rmse:{: f}'.format(it, train_loss, test_rmse))
if last_test_rmse and (last_test_rmse - test_rmse) <= 0:
print('convergence at iterations:{: d}'.format(it))
break
else:
last_test_rmse = test_rmse
return self.U, self.V, train_loss_list, test_rmse_list
def RMSE(self, preds, truth):
# return np.sqrt(np.mean(np.square(preds - truth)))
return np.mean(np.abs(preds - truth))
# tf.reduce_mean(tf.abs(y - y_))
def Evaluate_MAE_AND_NMAE(self):
matrix_sub = sp.dok_matrix.copy(self.testMatrix)
user_add = []
m = 0 # 用户数
num_item_each_user = np.zeros(self.num_users)
for (userid, itemid) in self.testMatrix.keys():
matrix_sub[userid, itemid] = math.fabs(matrix_sub[userid, itemid] - self.predict_matrix[userid][itemid])
if userid not in user_add:
user_add.append(userid)
m += 1
num_item_each_user[userid] += 1
sum = 0
sum_each_row = np.sum(matrix_sub.toarray(), axis=1)
for i in range(0, self.num_users):
if num_item_each_user[i] != 0:
sum += sum_each_row[i] / num_item_each_user[i]
MAE = sum / m
NMAE = sum / (4 * m)
return MAE, NMAE
if __name__ == '__main__':
# 0: Hybird_,
# 1: test_,
# 2: ml_100k_,
# 3: ml_1m_,
# 4: pcc_data,
# 5:ml_200_1000_
list_dataset = get_dataset_path(2)
recommemdation = PMF(list_dataset[0], list_dataset[1], list_dataset[2], 128)
# count=1
# mae,nmae=recommemdation.Evaluate_MAE_AND_NMAE()
# print(mae)
# print(nmae)
# for i in range(99):
# for j in range(400):
# if matrix[i,j] >=5.0:
# print(matrix[i,j])
# count+=1
# break
# print(count)