#!usr/bin/python+# -*- coding:UTF-8 -*-
#-------------------------FUNCTION---------------------------#
from pylab import *
import numpy as np
import random
import math
def SGD(train,test,N,M,eta,K,lambda_1,lambda_2,Step):
# train: train data
# test: test data
# N:the number of user
# M:the number of item
# eta: the learning rata ，eta=0.005
# K: the number of latent factor ，K=10
# lambda_1,lambda_2: regularization parameters ，lambda_1=0.1，lambda_2=0.1
# Step: the max iteration ，Step=50
U = np.random.normal(0, 0.1, (N, K))
V = np.random.normal(0, 0.1, (M, K))
L=1000.0
rmse=[]
loss=[]
for ste in range(Step):
los=0.0
for data in train:
u=data[0]
i=data[1]
r=data[2]
e=r-np.dot(U[u],V[i].T)
U[u]=U[u]+eta*(e*V[i]-lambda_1*U[u])
V[i]=V[i]+eta*(e*U[u]-lambda_2*V[i])
los=los+0.5*(e**2+lambda_1*np.square(U[u]).sum()+lambda_2*np.square(V[i]).sum())
loss.append(los)
rms=RMSE(U,V,test)
rmse.append(rms)
if loss < L:
break
if ste%10==0:
print (ste/10)
return los,rmse,U,V
def RMSE(U,V,test):
count=len(test)
sum_rmse=0.0
for t in test:
u=t[0]
i=t[1]
r=t[2]
pr=np.dot(U[u],V[i].T)
sum_rmse+=np.square(r-pr)
rmse=np.sqrt(sum_rmse/count)
return rmse
def Load_data(filedir,ratio):
user_set={}
item_set={}
N=0;#the number of user
M=0;#the number of item
u_idx=0
i_idx=0
data=[]
f = open(filedir)
for line in f.readlines():
u,i,r,t=line.split()
if int(u) not in user_set:
user_set[int(u)]=u_idx
u_idx+=1
if int(i) not in item_set:
item_set[int(i)]=i_idx
i_idx+=1
data.append([user_set[int(u)],item_set[int(i)],int(r)])
f.close()
N=u_idx;
M=i_idx;
np.random.shuffle(data)
train=data[0:int(len(data)*ratio)]
test=data[int(len(data)*ratio):]
return N,M,train,test
def Figure(loss,rmse):
fig1=plt.figure('LOSS')
x = range(len(loss))
plot(x, loss, color='g',linewidth=3)
plt.title('Convergence curve')
plt.xlabel('Iterations')
plt.ylabel('Loss')
fig2=plt.figure('RMSE')
x = range(len(rmse))
plot(x, rmse, color='r',linewidth=3)
plt.title('Convergence curve')
plt.xlabel('Iterations')
plt.ylabel('RMSE')
show()
#----------------------------SELF TEST----------------------------#
def main():
dir_data="F:/Machine learning Code/PMF/ml-100k (1)/ml-100k/u.data"
ratio=0.8
N,M,train,test=Load_data(dir_data,ratio)
eta=0.005
K=10
lambda_1=0.1
lambda_2=0.1
Step=50
loss,rmse,U,V=SGD(train,test,N,M,eta,K,lambda_1,lambda_2,Step)
print (rmse[-1]);
Figure(loss,rmse)
if __name__ == '__main__':
main()
