-
Notifications
You must be signed in to change notification settings - Fork 9
/
Copy pathgenerate_data.py
182 lines (175 loc) · 8.17 KB
/
generate_data.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
#!/usr/bin/env python
# -*- coding: utf-8 -*-
# @Time : 2021/7/5 3:19
# @Author : ZM7
# @File : generate_data
# @Software: PyCharm
import dgl
import os
import numpy as np
import torch
from TKG.utils import get_data_with_t
import argparse
from dgl import save_graphs
from collections import defaultdict
from TKG.load_data import load_data_list
def get_total_number(inPath, fileName):
with open(os.path.join(inPath, fileName), 'r') as fr:
for line in fr:
line_split = line.split()
return int(line_split[0]), int(line_split[1])
def generate_graph(data, time, time_num, nodes_num=None, rel_nums=None, name='Sample'):
"""
:param data:
:param time:
:param time_num: 每个时刻下entity的数量
:param nodes_num:
:param rel_nums:
:param name:
:return:
"""
u = []
v = []
rel = []
rel_1 = []
u_id = [] # 记录节点的原始编号
t_id = [] # 记录节点每次出现的时间
rel_r = [] #时间差
rel_t = []
rel_s = []
rel_s_ = []
rel_h = []
rel_o = []
L = len(time) # 总的时间长度
time_num = np.roll(time_num, 1)
time_num[0] = 0
idx = np.cumsum(time_num)
num_t = np.zeros(L)
for i, tim in enumerate(time):
print(i)
if name == 'Sample':
data_1 = get_data_with_t(data, tim)
else:
data_1 = data[i]
src1, rel1, dst1 = data_1[:,[0,1,2]].transpose()
uniq_v, edges = np.unique((src1, dst1), return_inverse=True)
n_src1, n_dst1 = np.reshape(edges, (2, -1)) + idx[i] # 给节点在新图中重新编码
u.append(np.concatenate((n_src1, n_dst1)))
u_id.append(uniq_v)
t_id.append(i*np.ones(len(uniq_v), dtype=int))
v.append(np.concatenate((n_dst1, n_src1)))
rel.append(np.concatenate((rel1, rel1+rel_nums)))
rel_1.append(np.concatenate((rel1, rel1+rel_nums)))
rel_r.append(0 * np.ones(len(rel1) * 2, dtype=int))
rel_t.append(i*np.ones(len(rel1)*2, dtype=int))
rel_s.append(i*np.ones(len(rel1)*2, dtype=int))
rel_s_.append(i*np.ones(len(rel1)*2, dtype=int))
rel_h.append(np.concatenate((n_src1, n_dst1)))
rel_o.append(np.concatenate((n_dst1, n_src1)))
if i == len(time)-1:
break
for j, tim in enumerate(time[i+1:]):
if name == 'Sample':
data_2 = get_data_with_t(data, time[i+j+1])
else:
data_2 = data[i+j+1]
src2, rel2, dst2 = data_2[:,[0,1,2]].transpose()
uniq_v2, edges2 = np.unique((src2, dst2), return_inverse=True)
un_entity = np.intersect1d(uniq_v, uniq_v2) # 两个时刻子图重复的节点
if len(un_entity) == 0:
print('hhh', i, tim)
continue
u1 = np.where(np.in1d(uniq_v, un_entity))[0] # 判断公共节点在不同子图内的索引
u2 = np.where(np.in1d(uniq_v2, un_entity))[0] # 判断公共节点在不同子图内的索引
u.append(np.concatenate((u1 + idx[i], u2 + idx[i+j+1])))
v.append(np.concatenate((u2 + idx[i+j+1], u1 + idx[i])))
rel.append(2*rel_nums*np.ones(2*len(un_entity), dtype=int)) # 时间类型的边
rel_1.append((2 * rel_nums + j) * np.ones(2 * len(un_entity), dtype=int)) # 给不同时间类型的边定义大小
#rel.append((2 * rel_nums + j) * np.ones(2 * len(un_entity), dtype=int)) # 时间差
rel_r.append((j+1) * np.ones(2 * len(un_entity), dtype=int)) # 时间差
rel_t.append(L*np.ones(2*len(un_entity), dtype=int))
rel_s.append((i+j+1)*np.ones(2*len(un_entity), dtype=int)) # 头实体发生时间
rel_s_.append(i*np.ones(2*len(un_entity), dtype=int)) # 尾实体发生时间
rel_h.append(np.concatenate((u1 + idx[i], u2 + idx[i+j+1])))
rel_o.append(np.concatenate((u2 + idx[i+j+1], u1 + idx[i])))
u = np.concatenate(u)
u_id = np.concatenate(u_id)
t_id = np.concatenate(t_id)
v = np.concatenate(v)
rel = np.concatenate(rel)
rel_1 = np.concatenate(rel_1)
rel_r = np.concatenate(rel_r)
rel_t = np.concatenate(rel_t)
rel_s = np.concatenate(rel_s)
rel_s_ = np.concatenate(rel_s_)
rel_h = np.concatenate(rel_h)
rel_o = np.concatenate(rel_o)
graph = dgl.graph((u, v))
graph.edata['etype'] = torch.LongTensor(rel) # edge relation
graph.edata['etype1'] = torch.LongTensor(rel_1)
graph.edata['e_r'] = torch.LongTensor(rel_r) # 相对时间
graph.edata['e_t'] = torch.LongTensor(rel_t) # edge time (T is L)
graph.edata['e_s'] = torch.LongTensor(rel_s) # edge time (头实体发生的时间)
graph.edata['e_s_'] = torch.LongTensor(rel_s_) # edge time (尾实体发生的时间)
graph.edata['e_rel_h'] = torch.LongTensor(rel_h)
graph.edata['e_rel_o'] = torch.LongTensor(rel_o)
graph.ndata['id'] = torch.from_numpy(u_id).long()
graph.ndata['t_id'] = torch.from_numpy(t_id).long()
# ----计算每个时刻的s前一时刻在大图中的位置
s_his = defaultdict(int) #记录前一次的索引
s_his_t = defaultdict(int) #记录前一次发生的时间
s_his_f = defaultdict(int) #记录前一次的索引
s_his_l = defaultdict(int) #记录历史序列的长度
s_last_index = np.zeros((nodes_num, L), dtype=int) #记录每个时刻的索引
s_last_t = L*np.ones((nodes_num, L), dtype=int) # 记录每个时刻上次交互的时间,上次没有发生记为L
s_last_f = graph.num_nodes() * np.ones((nodes_num, L), dtype=int) #记录每个时刻的索引，没有发生过的节点值为nodes_num
s_last_l = np.zeros((nodes_num, L), dtype=int)
node_id = graph.ndata['id'].numpy()
time_id = graph.ndata['t_id'].numpy()
# node_id_new = len(node_id) - 1 - np.unique(node_id[::-1], return_index=True)[1]
# 初始化
id, node_id_f = np.unique(node_id, return_index=True)
for n_i, ei in enumerate(id):
s_his[ei] = node_id_f[n_i]
s_his_t[ei] = L # T+1时间
s_his_f[ei] = graph.num_nodes()
s_his_l[ei] = 0 # 0
s_last_index[id, 0] = node_id_f
for i, tim in enumerate(time):
if name == 'Sample':
data_1 = get_data_with_t(data, tim)
else:
data_1 = data[i]
src1, rel1, dst1 = data_1[:,[0,1,2]].transpose()
# -----找一个数组记录每个元素这一时刻的索引------
en = np.unique((src1, dst1))
if i > 0:
s_last_index[:, i] = s_last_index[:, i-1]
s_last_t[:, i] = s_last_t[:, i-1]
s_last_f[:, i] = s_last_f[:, i-1]
s_last_l[:, i] = s_last_l[:, i-1]
for e_i, e in enumerate(en):
s_last_index[e, i] = s_his[e]
s_last_t[e, i] = s_his_t[e]
s_last_f[e, i] = s_his_f[e]
s_last_l[e, i] = s_his_l[e]
s_his[e] = e_i + idx[i] # 计算s上一次交互的编号
s_his_f[e] = e_i + idx[i] # 计算s上一次交互的编号
s_his_t[e] = i # 计算s上一次发生交互的时间
s_his_l[e] = s_his_l[e] + 1 # 计算s历史序列的长度
# -----------------------------------------
return graph, torch.from_numpy(s_last_index).long(), torch.from_numpy(s_last_t).long(), \
torch.from_numpy(s_last_f).long(), torch.from_numpy(s_last_l).long()
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Data')
parser.add_argument('--data', default='Sample')
args = parser.parse_args()
print(args)
num_nodes, num_rels, train_list, valid_list, test_list = load_data_list(args.data, args.space)
total_data = train_list + valid_list + test_list
time_num = [len(np.unique(da[:, [0, 2]])) for da in total_data]
total_times = range(len(total_data))
save_path = './' + 'graph_' + args.data
graph, s_index, s_t, s_f, s_l = generate_graph(total_data, total_times, time_num=time_num, nodes_num=num_nodes,
rel_nums=num_rels, name=args.data)
save_graphs(save_path, graph, {'s_index': s_index, 's_t': s_t, 's_f': s_f, 's_l': s_l})