-
Notifications
You must be signed in to change notification settings - Fork 4
/
itemGraph.py
167 lines (138 loc) · 4.51 KB
/
itemGraph.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
import pandas as pd
import numpy as np
import math
import collections
#Read the data: user_item_matrix, sij, suv
user_item_matrix = pd.read_pickle('user_item_matrix.pickle')
# print(user_item_matrix.shape)
# user_item_matrix = np.array([[2,2,4],[np.nan,5,2],[2,2,5],[4,4,1]])
# user_item_matrix = pd.DataFrame(user_item_matrix)
items, users = user_item_matrix.shape
s_uv = pd.read_pickle('s_uv.pickle')
s_ij = pd.read_pickle('s_ij.pickle')
max_rating = 5
maxIter = 10
class itemNode:
def __init__(self,itemId, y,neighbors):
self.itemId = itemId
self.y = y
self.neighbors = list(neighbors)
#The neighbors list contains self, which should not be considered.
if itemId in self.neighbors:
self.neighbors.remove(itemId)
self.phi = []
self.msgs = {}
self.psy = collections.defaultdict(list)
self.set_phi()
self.set_psy()
self.set_msgs()
def set_msgs(self):
for n in self.neighbors:
self.msgs[n] = np.array([0.2,0.2,0.2,0.2,0.2])
def set_phi(self):
if self.y < 1:
self.phi = [0.95,0,0,0,0]
elif self.y > 5:
self.phi = [0,0,0,0,0.95]
else:
for i in range(1,max_rating+1):
flag = 0
if i==math.ceil(self.y):
self.phi.append(0.95 - (math.ceil(self.y)-self.y))
flag = 1
if i==math.floor(self.y) and flag==0:
self.phi.append(math.ceil(self.y)-self.y)
flag = 1
if flag==0:
self.phi.append(0)
zero_indices = np.where(np.array(self.phi)==0)
self.phi = np.array(self.phi)
self.phi.put(zero_indices[0],(1-sum(self.phi))/len(zero_indices[0]))
def set_psy(self):
global s_ij
global max_rating
for n in self.neighbors:
p = []
sigma = (((1-s_ij.iloc[self.itemId,n])/(1-0.35)) + 1)/math.sqrt(2)
for i in range(1,max_rating+1):
for j in range(1,max_rating+1):
p.append(-((i-j)**2)/sigma)
self.psy[n] = np.exp(p)
self.psy[n] = self.psy[n]/sum(self.psy[n])
self.psy[n] = self.psy[n].reshape([5,5])
def compute_yi(mask, user, mean_normalized_user_item_matrix, r_cap):
deno = pd.DataFrame(mask.values * s_uv.iloc[user].values)
deno = (deno.abs()).sum(axis=1)
numerator = pd.DataFrame(mean_normalized_user_item_matrix.values*s_uv.iloc[user].values).sum(axis=1)
y_i = r_cap.iloc[user] + numerator/deno
y_i = y_i.fillna(r_cap.iloc[user])
y_i = y_i.as_matrix()
return y_i
def graph_traversal(maxIter, itemNodes):
for i in range(maxIter):
visited = []
stack = [itemNodes[0]]
#DFS Graph traversal
while stack:
node = stack.pop()
visited.append(node.itemId)
for n in node.neighbors:
local_msgs = np.array([1,1,1,1,1])
for k in node.neighbors:
if k != n:
local_msgs = local_msgs*itemNodes[k].msgs[node.itemId]
factor = node.psy[n]*node.phi
for i in range(max_rating):
factor[i] = factor[i]*local_msgs[i]
node.msgs[n] = factor.sum(axis=0)
node.msgs[n] = node.msgs[n]/float(node.msgs[n].sum())
if n not in visited:
stack.append(itemNodes[n])
def inference(itemNodes):
final_ratings = []
for i in range(len(itemNodes)):
m = np.array([1,1,1,1,1])
for n in itemNodes[i].neighbors:
m = m*itemNodes[i].msgs[n]
Pz = itemNodes[i].phi*m
Pz = Pz/Pz.sum()
r = [1,2,3,4,5]
final_ratings.append(sum(r*Pz))
return final_ratings
#Calculate the mean rating for all user
r_cap = user_item_matrix.mean(axis=0)
#Mean normalize all ratings, (rui - ru)
mean_normalized_user_item_matrix = user_item_matrix - r_cap
#This is a mask, since we only have to consider those items which are being rated
item_rated = user_item_matrix.notnull()
user_ratings = []
for user in range(users):
print('Computing for User:',user)
#Compute y_i's
y_i = compute_yi(item_rated, user, mean_normalized_user_item_matrix, r_cap)
top_K = 2
kmax_value = s_ij.fillna(-2).as_matrix()
kmax_value = np.sort(kmax_value)[:,::-1]
kmax_value = kmax_value[:,top_K]
rows,cols = s_ij.shape
neighbors = []
s_ij_np_matrix = s_ij.as_matrix()
final_neighbors = np.zeros([cols,cols])
for i in range(cols):
neighbors.append(np.where(s_ij_np_matrix[i,:]>=kmax_value[i])[0][:top_K+1])
final_neighbors[i,neighbors[i]] = 1
final_neighbors = final_neighbors + final_neighbors.transpose()
neighbors = []
for i in range(cols):
neighbors.append(np.where(final_neighbors[i,:]>0)[0])
#Creating item Nodes of the graph
items = len(neighbors)
itemNodes = []
for i in range(items):
itemNodes.append(itemNode(i, y_i[i], neighbors[i]))
graph_traversal(maxIter, itemNodes)
final_ratings = inference(itemNodes)
user_ratings.append(final_ratings)
print(user_ratings)
user_ratings = np.array(user_ratings)
np.save('user_ratings.npy',user_ratings)