-
Notifications
You must be signed in to change notification settings - Fork 5
/
loader.py
216 lines (204 loc) · 7.24 KB
/
loader.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
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
#!/usr/bin/env python
# -*- coding: utf-8 -*-
import os
import sys
import numpy as np
import pandas as pd
import networkx as nx
from collections import Counter
from scipy.sparse import csr_matrix
from data_helper import train_tiedAE
from sklearn import preprocessing
def load_data(data):
if data == "Alibaba-s":
data = np.loadtxt('data/Alibaba_s/alibaba_s.txt', dtype=int)
return data, None
elif data == "Alibaba":
data = np.loadtxt('data/Alibaba/alibaba_edges.txt', dtype=int)
feats_u = np.loadtxt('data/Alibaba/alibaba_featu.txt', dtype=float)
feats_v = np.loadtxt('data/Alibaba/alibaba_featv.txt', dtype=float)
feats = np.concatenate((feats_u,feats_v),axis=0)
return data, feats
elif data == "DTI":
data = np.loadtxt('data/DTI/DTI.txt', dtype=int)
return data, None
elif data == "Amazon":
data = np.loadtxt('data/Amazon/amazon_edges.txt', dtype=int)
feats_u = np.loadtxt('data/Amazon/amazon_featu.txt', dtype=float)
feats_v = np.loadtxt('data/Amazon/amazon_featv.txt', dtype=float)
feats = np.concatenate((feats_u,feats_v),axis=0)
return data, feats
def load_attributes(data, feats):
# print(feats.shape)
num_btype = len(np.unique(data[:,2]))
btypes = [str(i) for i in range(num_btype)]
# btypes = []
btypes.append('base')
# print(btypes)
init_feats = dict()
for btype in btypes:
init_feats[btype] = feats #preprocessing.normalize(feats)
return init_feats
def split_train_test(data, feats, flag, ratio):
n_samples = data.shape[0]
n_test = int(n_samples*ratio)
ridx = np.random.choice(n_samples, n_test, replace=False)
test = data[ridx]
train = np.delete(data, ridx, axis=0)
print(test.shape,train.shape)
train_nodes_u = [i for i in list(set(train[:,0]))]
train_nodes_i = [i for i in list(set(train[:,1]))]
# print(len(train_nodes_u),len(train_nodes_i))
train_ui = train_nodes_u
for i in train_nodes_i:
train_ui.append(i)
# if feats != None:
# feats_train = np.array([feats[i] for i in train_ui])
# print(feats.shape, feats_train.shape)
u_train,i_train = np.unique(train[:,0]),np.unique(train[:,1])
f_test = []
for line in test:
if line[0] in u_train and line[1] in i_train:
f_test.append(line)
f_test = np.array(f_test)
# print("### The train contains %d edges(%d drugs and %d targets), and test contains %d edges(%d drugs and %d targets)."
# % (train.shape[0],len(u_train),len(i_train),f_test.shape[0],len(np.unique(f_test[:,0])),len(np.unique(f_test[:,1]))))
idx_u_map = {j:i for i,j in enumerate(np.unique(train[:,0]))}
idx_i_map = {j:len(np.unique(train[:,0]))+i for i,j in enumerate(np.unique(train[:,1]))}
new_train,new_test = [],[]
for line in train:
tmp = []
tmp.append(idx_u_map[line[0]])
tmp.append(idx_i_map[line[1]])
tmp.append(line[2])
# tmp.append(line[3]) # label
new_train.append(tmp)
for line in f_test:
tmp = []
tmp.append(idx_u_map[line[0]])
tmp.append(idx_i_map[line[1]])
tmp.append(line[2])
# tmp.append(line[3])
new_test.append(tmp)
new_train,new_test = np.array(new_train),np.array(new_test)
u_train_new,i_train_new = np.unique(new_train[:,0]),np.unique(new_train[:,1])
# u_test_new,i_test_new = np.unique(new_test[:,0]),np.unique(new_test[:,1])
# print("### The train contains %d edges(%d drugs and %d targets), and test contains %d edges(%d drugs and %d targets)."
# % (new_train.shape[0],len(u_train_new),len(i_train_new),new_test.shape[0],len(u_test_new),len(i_test_new)))
print(len(np.unique(new_train[:,0])),len(np.unique(new_train[:,1])))
if flag == True:
feats_train = np.array([feats[i] for i in train_ui])
return new_train, new_test, feats_train
elif flag == False:
return new_train, new_test, None
def extract_hyedges_types(data, btype):
new_data = []
for line in data:
if line[2] == btype:
new_data.append(line)
return np.array(new_data)
def construct_hierarchical_hypergraph(data, nodes):
hygraphs = dict()
btypes = list(set(data[:,2])) #[0,1,2,3,4]
hygraph_base = construct_hypergraph(data, nodes)
hygraphs['base'] = hygraph_base
for btype in btypes:
data_type = extract_hyedges_types(data, btype)
hygraphs[str(btype)] = construct_hypergraph(data_type, nodes)
return hygraphs
def construct_hypergraph(data, nodes):
# construct Bigraph
nodes_u = [i for i in list(set(data[:,0]))]
nodes_i = [i for i in list(set(data[:,1]))]
all_nodes_u,all_nodes_i = nodes['u'],nodes['i']
print(len(nodes_u),len(nodes_i),len(all_nodes_u),len(all_nodes_i))
Bigraph = nx.Graph()
for line in data:
node_u,node_i = line[0],line[1]
Bigraph.add_node(node_u, bipartite=0)
Bigraph.add_node(node_i, bipartite=1)
Bigraph.add_edge(node_u, node_i, btype=line[2])
# construct Hygraph
Hygraph = dict()
n_neigs_u = 0
n_neigs_i = 0
for u in all_nodes_u:
if u in nodes_u:
neighbors = Bigraph.edges(u)
neigs_u = [i for u,i in neighbors]
Hygraph[u] = neigs_u
n_neigs_u += len(neigs_u)
else:
Hygraph[u] = []
for i in all_nodes_i:
if i in nodes_i:
neighbors = Bigraph.edges(i)
neigs_i = [u for i,u in neighbors]
Hygraph[i] = neigs_i
n_neigs_i += len(neigs_i)
else:
Hygraph[i] = []
print('### The number of hyper-edges: %d' %(len(nodes_u)+len(nodes_i)))
print('### The average nodes in each hyper-edge: %0.2f (%0.2f for drugs and %0.2f for targets)'
% ((n_neigs_u+n_neigs_i)/(len(nodes_u)+len(nodes_i)),n_neigs_u/len(nodes_u),(n_neigs_i/len(nodes_i))))
return Hygraph
def generate_adj(data,nodes,btype):
N_u = nodes['n_u']#len(np.unique(data[:,0]))
N_i = nodes['n_i']#len(np.unique(data[:,1]))
N = N_u + N_i
adj = np.zeros((N, N), dtype=int)
if btype == 'base':
for line in data:
adj[line[0],line[1]] = 1
else:
for line in data:
if line[2] == int(btype):
adj[line[0],line[1]] = 1
# print(np.sum(adj == 1))
return csr_matrix(adj).astype('float32')
def initialize_features(args, data, nodes, dim=32):
# Encoder Based Approach
print('### Generating initial features by Encoder-Based-Approach...')
num_btype = len(np.unique(data[:,2]))
btypes = [str(i) for i in range(num_btype)]
# btypes = []
btypes.append('base')
initial_feats = dict()
for btype in btypes:
A = generate_adj(data,nodes,btype).todense()
initial_feat = train_tiedAE(A,dim=args.dim_f,lr=args.lr_eba,weight_decay=args.weight_decay_eba,n_epochs=args.epoch_eba)
initial_feats[btype] = preprocessing.normalize(initial_feat)
return initial_feats
def generate_incidence_matrix_multiple(hygraphs):
Hs = dict()
btypes = hygraphs.keys()
n_smp = len(hygraphs['base'])
for btype in btypes:
H = generate_incidence_matrix(hygraphs[btype],n_smp)
Hs[btype] = H
return Hs
def generate_incidence_matrix(hyedges, n_smp):
H = np.zeros((n_smp,n_smp))
for key,val in hyedges.items():
for v in val:
H[v,key] = 1
return H
def generate_negative_samples(pos_edges, hygraph, num_neg_samples):
nodes_u = list(set([u for u,i in pos_edges]))
nodes_i = list(set([i for u,i in pos_edges]))
neg_edges = []
for u in nodes_u:
candidates = list(set(nodes_i) - set(hygraph[u]))
neg_nodes = np.random.choice(candidates, num_neg_samples, replace=False)
for neg in neg_nodes:
tmp = [u, neg]
neg_edges.append(tmp)
for i in nodes_i:
candidates = list(set(nodes_u) - set(hygraph[i]))
neg_nodes = np.random.choice(candidates, num_neg_samples, replace=False)
for neg in neg_nodes:
tmp = [i, neg]
neg_edges.append(tmp)
neg_edges = np.array(neg_edges)
np.random.shuffle(neg_edges)
return neg_edges