-
Notifications
You must be signed in to change notification settings - Fork 2
/
cur.py
194 lines (175 loc) · 6.39 KB
/
cur.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
import pandas as pd
import numpy as np
import math
import random
import time
precision_k = 5000
num_of_users = 6040 + 1
num_of_movies= 3952 + 1
num_of_ratings = 1000209
def preprocess():
'''
preprocessing the data by loading data into user_movie_matrix
returns :user_movie_matrix
'''
#Reading ratings file:
r_cols = ['user_id', 'movie_id', 'rating', 'unix_timestamp']
ratings = pd.read_csv('ml-1m/ratings.dat', sep="::", names=r_cols,encoding='latin-1',engine='python')
ratings= ratings.to_numpy()
indices = list(range(ratings.shape[0]))
random.shuffle(indices)
ratings = ratings[indices]
ratings= pd.DataFrame(ratings)
ratings = ratings.rename(columns={0: 'user_id',1 : 'movie_id',2 : 'rating', 3: 'unix_timestamp'},inplace= False)
ratings = ratings[['user_id', 'movie_id', 'rating']]
ratings_list = ratings.values.tolist()
user_movie_matrix = np.zeros((num_of_users,num_of_movies))
#computing utlity matrix
for i in range(num_of_ratings):
user_id = ratings_list[i][0]
movie_id = ratings_list[i][1]
rating = ratings_list[i][2]
user_movie_matrix[user_id][movie_id] = rating
return user_movie_matrix
def center(user_movie_matrix):
'''
centering the matrix around mean
parameters : user_movie_matrix
returns : matrix_centered_zero
'''
matrix_centered_zero = np.copy(user_movie_matrix)
#center the test data set about the mean
mean = 0.0
for i in range(1,num_of_users):
sum = 0.0
count = 0.0
#compute mean
for j in range(1,num_of_movies):
if(user_movie_matrix[i][j] != 0):
sum = sum + user_movie_matrix[i][j]
count = count + 1.0
mean = sum / count
#update data about mean
for j in range(1,num_of_movies):
if(user_movie_matrix[i][j] == 0.0):
matrix_centered_zero[i][j] = mean
else:
matrix_centered_zero[i][j] = matrix_centered_zero[i][j] - mean
global test
#compute the training data by removing the values in the 1000*1000 matrix
test = np.copy(matrix_centered_zero)
for i in range(1,1001):
for j in range(1,1001):
if(matrix_centered_zero[i][j] != 0):
test[i][j] = -1
#recenter the training data about mean
mean = 0.0
for i in range(1,num_of_users):
sum = 0.0
count = 0.0
#compute mean
for j in range(1,num_of_movies):
if(test[i][j] == -1):
sum = sum + 0.0
count = count + 1.0
elif(test[i][j] > 0):
sum = sum + test[i][j]
count = count + 1.0
mean = sum / count
#recenter the data about mean
for j in range(1,num_of_movies):
if(test[i][j] == -1 or test[i][j] == 0):
test[i][j] = mean
else:
test[i][j] = test[i][j] - mean
return matrix_centered_zero
def main():
user_movie_matrix = preprocess()
matrix_centered_zero = center(user_movie_matrix)
#k factor of CUR decomposition
k = 250
start = time.time()
#calculate the total sum of the sqaures of the elements
total_sum_sq = 0.0
for i in range(1,num_of_users):
for j in range(1,num_of_movies):
total_sum_sq = total_sum_sq + (test[i][j])**2
#calculating the probabilty distribution for all the columns
col_dis_pr = []
col_dis_pr.append(0.0)
for i in range(1,num_of_movies):
col_sum_sq = 0.0
for j in range(1,num_of_users):
col_sum_sq = col_sum_sq + (test[j][i])**2
col_dis_pr.append(col_sum_sq / total_sum_sq)
#calculating the probabilty distribution for all the rows
row_dis_pr = []
row_dis_pr.append(0.0)
for i in range(1,num_of_users):
row_sum_sq = 0.0
for j in range(1,num_of_movies):
row_sum_sq = row_sum_sq + (test[i][j])**2
row_dis_pr.append(row_sum_sq / total_sum_sq)
#making a list with the indices of all columns
cols_index = []
cols_index.append(-1)
for i in range(0,3952):
cols_index.append(i+1)
#making a list with the indices of all rows
rows_index = []
rows_index.append(-1)
for i in range(0,6040):
rows_index.append(i+1)
#computing random values with given probability distribution for the rows and columns
cols = np.random.choice(cols_index, 4 * k,replace=False, p = col_dis_pr)
rows = np.random.choice(rows_index, 4 * k,replace=False, p = row_dis_pr)
#c = 4 * k
c_attr = 1000.0
C = np.zeros((num_of_users, 4*k + 1))
#computing the C matrix
for i in range(1,4*k+1):
C[:,i] = np.divide(test[:,cols[i-1]], np.sqrt(np.multiply(c_attr,col_dis_pr[cols[i-1]])))
R = np.zeros((4*k+1, num_of_movies))
#Computing the R matrix
for i in range(1,4*k+1):
R[i,:] = np.divide(test[rows[i-1],:], np.sqrt(np.multiply(c_attr,row_dis_pr[rows[i-1]])))
#Computing the inverses
C_inv = np.linalg.pinv(C)
R_inv = np.linalg.pinv(R)
#Computing U
U = np.matmul(np.matmul(C_inv, test), R_inv)
#Computing the original matrix from C , U , R
answer = np.matmul(np.matmul(C, U), R)
squares_sum = 0.0
count_sq = 0.0
precision_rating = []
for i in range(1,1001):
for j in range(1,1001):
if(user_movie_matrix[i][j] != 0):
precision_rating.append(answer[i][j])
print("Actual rating")
print(user_movie_matrix[i][j])
print("Predicted rating")
print(answer[i][j])
#compute the rmse
squares_sum = squares_sum + (answer[i][j] - user_movie_matrix[i][j])**2
count_sq = count_sq + 1.0
print("")
print("Root mean squared error")
print(math.sqrt(squares_sum / count_sq))
print("Spearman's correlation")
correlation = 1 - ((6 * squares_sum) / (count_sq**3 - count_sq))
print(correlation)
#calculation of precision at top k
precision_rating.sort(reverse=True)
countk = 0.0
for i in range(0, precision_k):
if(precision_rating[i] >= 3.0):
countk = countk + 1
precision_at_topk = countk / precision_k
print("Precision at top k")
print(precision_at_topk)
print("Time required for CUR ")
print("--- %s seconds ---" % (time.time() - start))
if __name__== "__main__":
main()