-
Notifications
You must be signed in to change notification settings - Fork 0
/
main_RBM_100k.py
134 lines (106 loc) · 5.26 KB
/
main_RBM_100k.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
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Fri Mar 9 21:13:47 2018
@author: baifrank
"""
import numpy as np
import pandas as pd
import torch
import torch.nn.parallel
from sklearn import linear_model
import matplotlib.pyplot as plt
from RBM_torch import dataset
from RBM_torch import utils
from RBM_torch import rbm
from RBM_torch import RBM_symm_utils
###########################################################################
if __name__ == "__main__":
path = '/Users/baifrank/Desktop/ml-100k'
train_path, test_path = path+'/ua.base', path+'/ua.test'
##导入用户数量，电影数量，torch形式的原始评分矩阵(train+test)
nb_users,nb_movies,training_set,test_set = dataset.data_input(train_path,test_path)
k = 5
##实值矩阵Binary化，列数*5
train_tensor_u = utils.expand(training_set, k, neg=0)
train_tensor_i = utils.expand(training_set.t(), k, neg=0)
train_tensor_neg_u = utils.expand(training_set, k, neg=1)
train_tensor_neg_i = utils.expand(training_set.t(), k, neg=1)
##user/item based##
def rbm_run(nh,nb_epoch,k_gibbs,batch_size,decay,momentum,item_based):
l_mae_train, l_mse_train, l_mae_test, l_mse_test =[], [], [], []
##初始化一个rbm的class
if item_based == True:
train_tensor, train_tensor_neg = train_tensor_i, train_tensor_neg_i
nb_rows, nv = nb_movies, nb_users
else:
train_tensor, train_tensor_neg = train_tensor_u, train_tensor_neg_u
nb_rows, nv = nb_users, nb_movies
rbm_model = rbm.RBM(nv,nh)
rbm_model.params_init()
##开始迭代之前先初始化前一次迭代的梯度
prev_gw = torch.randn(rbm_model.dim)
prev_gbv = torch.randn(1, rbm_model.k*rbm_model.num_visible)
prev_gbh = torch.randn(1, rbm_model.num_hidden)
##Training the RBM
for epoch in range(1, nb_epoch + 1):
print("Calculating:"+str(epoch)+'/'+str(nb_epoch))
for id_start in range(0, nb_rows - batch_size, batch_size):
v0 = train_tensor[id_start:id_start+batch_size]
vk = train_tensor[id_start:id_start+batch_size]
v0_neg = train_tensor_neg[id_start:id_start+batch_size]
##Gibbs Sampling
ph0,h0 = rbm_model.sample_hidden(v0)
for k in range(k_gibbs):
_,hk = rbm_model.sample_hidden(vk)
_,vk = rbm_model.sample_visible(hk)
vk[v0_neg == -1] = 0
phk,_ = rbm_model.sample_hidden(vk)
rbm_model.train(v0, vk, ph0, phk,prev_gw, prev_gbv, prev_gbh, w_lr=0.01,v_lr=0.01,h_lr=0.01,decay=decay,momentum=momentum)
prev_gw, prev_gvb, prev_gvh = rbm_model.gradient(v0, vk, ph0, phk)
##重构之后计算误差
data_recons = utils.predict(train_tensor, rbm_model, do_round = True, range_15 = True)
if item_based == True:
data_recons = data_recons.t()
mae_train, mse_train = utils.calculate_error(training_set,data_recons)
mae_test, mse_test = utils.calculate_error(test_set,data_recons)
print(mae_train, mse_train, mae_test, mse_test)
l_mae_train.append(mae_train)
l_mse_train.append(mse_train)
l_mae_test.append(mae_test)
l_mse_test.append(mse_test)
return rbm_model, l_mae_train, l_mse_train, l_mae_test,l_mse_test
##输出
"l_mae_train, l_mse_train, l_mae_test,l_mse_test"
##user_based
rbm_model3_u, l_mae_train3_u, l_mse_train3_u, l_mae_test3_u, l_mse_test3_u = rbm_run(nh=100,nb_epoch=400,k_gibbs=10,batch_size=100,decay=0,momentum=0,item_based=False)
utils.show_min_loss(l_mae_train3_u, l_mae_test3_u)
##item_based
rbm_model3_i, l_mae_train3_i, l_mse_train3_i, l_mae_test3_i, l_mse_test3_i = rbm_run(nh=100,nb_epoch=400,k_gibbs=10,batch_size=100,decay=0,momentum=0,item_based=True)
utils.show_min_loss(l_mae_train3_i, l_mae_test3_i)
##作图观察误差走势
x = range(0,400)
y = l_mae_train3_u
plt.figure()
plt.plot(x,y,"-",label="nh=100,k_gibbs=10",color="red")
plt.xlabel("iteration")
plt.ylabel("mse")
plt.title("User based RBM(Train set)")
plt.legend()
plt.show()
##基于uRBM和iRBM构建sRBM，并输出误差
RBM_symm_utils.RBM_symm(rbm_model3_u,rbm_model3_i,train_tensor_u,train_tensor_i,training_set,test_set)
##将表现最好的RBM模型重构的打分结果储存在本地
rbm_model_best_100k = rbm_model3_u
recons_100k = utils.predict(train_tensor_u, rbm_model_best_100k, do_round = True, range_15 = True)
l_recons_100k = recons_100k.numpy().flatten().tolist()
##定义对应的行与列的id
##将rbm的打分转化成dataframe输出到本地储存
iid_list = list(range(1,1683))*943
uid_list = []
for i in range(1,944):
for j in range(1682):
uid_list.append(i)
data_pre_100k={'iid':iid_list,'uid':uid_list,'pre':l_recons_100k}
df_pre_100k = pd.DataFrame(data_pre_100k)
df_pre_100k.to_csv("/Users/baifrank/Desktop/recomm_output/pre_best_100k.csv",index=False)