-
Notifications
You must be signed in to change notification settings - Fork 1
/
utils.py
268 lines (228 loc) · 9.64 KB
/
utils.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
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
from name import *
import os
import json
import torch
import random
import numpy as np
from torch.utils.data import Dataset, DataLoader
from torch_geometric.data import Data
def readjson(path, name):
filename = name + "." + Json
filepath = os.path.join(path, filename)
f = open(filepath, "r", encoding="utf-8")
text = json.load(f)
f.close()
return text
def seq_cut(sequence_list, cut_percent=0.5):
leftseq_list = []
rightseq_list = []
for i, seq in enumerate(sequence_list):
position = int(len(seq) * cut_percent)
left_seq = seq[:position]
right_seq = seq[position:]
right_seq.reverse()
leftseq_list.append(left_seq)
rightseq_list.append(right_seq)
return leftseq_list, rightseq_list
def pad_sequences(vectorized_seqs, seq_lengths):
# print(len(vectorized_seqs), len(vectorized_seqs[0][0]))
seq_tensor = torch.zeros((len(vectorized_seqs), seq_lengths.max(), len(vectorized_seqs[0][0]))).long()
# tmp_seq_tensor = torch.ones((len(vectorized_seqs), seq_lengths.max(), len(vectorized_seqs[0][0]))).long()
# seq_tensor = seq_tensor - tmp_seq_tensor
for idx, (seq, seqlen) in enumerate(zip(vectorized_seqs, seq_lengths)):
# seq_tensor[idx, :seqlen] = torch.LongTensor(seq)
seq_tensor[idx, len(seq_tensor[idx]) - seqlen:] = torch.FloatTensor(seq)
return seq_tensor
def random_sample_creation(sequence_list, cut_percent=0.5, neg_percent=1):
dataset = []
neg_num = int(len(sequence_list) * neg_percent)
leftseq_list, rightseq_list = seq_cut(sequence_list, cut_percent)
for i in range(len(sequence_list)):
dataset.append((leftseq_list[i], rightseq_list[i], 1))
for i in range(neg_num):
sample_result = random.sample(range(0, neg_num), 2)
left_index = sample_result[0]
right_index = sample_result[1]
dataset.append((leftseq_list[left_index], rightseq_list[right_index], 0))
dataset = np.array(dataset)
return dataset
def create_dataset(seq_list, test_percent=0.2, cut_percent=0.5, neg_percent=1):
train_num = int(len(seq_list) * (1 - test_percent))
train_seq_list = seq_list[:train_num]
test_seq_list = seq_list[train_num:]
trainset = random_sample_creation(train_seq_list, cut_percent, neg_percent)
testset = random_sample_creation(test_seq_list, cut_percent, neg_percent)
return trainset, testset
class PaddedTensorDataset(Dataset):
def __init__(self, left_tensor, right_tensor, label_tensor, leftlength_tensor, rightlength_tensor):
assert left_tensor.size(0) == right_tensor.size(0) == label_tensor.size(0)
# self.data_tensor = data_tensor
self.left_tensor = left_tensor
self.leftlength_tensor = leftlength_tensor
self.right_tensor = right_tensor
self.rightlength_tensor = rightlength_tensor
self.label_tensor = label_tensor
def __getitem__(self, index):
return self.left_tensor[index], self.right_tensor[index], self.label_tensor[index], self.leftlength_tensor[
index], self.rightlength_tensor[index]
def __len__(self):
return self.left_tensor.size(0)
def create_dataloader(seqs, shuffle=True, num_workers=2, batch_size=4):
leftseqs = seqs[:, 0]
rightseqs = seqs[:, 1]
labels = seqs[:, 2]
leftseq_lengths = torch.LongTensor([len(s) for s in leftseqs])
rightseq_lengths = torch.LongTensor([len(s) for s in rightseqs])
left_tensor = pad_sequences(leftseqs, leftseq_lengths)
right_tensor = pad_sequences(rightseqs, rightseq_lengths)
label_tensor = torch.LongTensor([label for label in labels])
return DataLoader(PaddedTensorDataset(left_tensor, right_tensor, label_tensor, leftseq_lengths, rightseq_lengths),
shuffle=shuffle, num_workers=num_workers, batch_size=batch_size, drop_last=True)
def batchfirst2second(fea):
f = torch.transpose(fea, 0, 1)
return f
def changelabel(label):
assert len(label.shape) == 1
newlabel = label.clone().detach()
for i in range(0, newlabel.shape[0]):
if newlabel[i] == 0:
newlabel[i] = -1
return newlabel
def sep_lis(lis, c):
return [lis[i:i+c] for i in range(len(lis)) if i % c == 0]
def get_train_paper(authorpaper_dict, author_list):
train_paper = []
for author in author_list:
papers = authorpaper_dict[author]
tmp = []
for paper in papers:
tmp.append([paper])
train_paper.append(tmp)
return train_paper
def get_train_nodes(leftseqs, rightseqs):
train_nodes = set()
leftseqs = leftseqs.numpy().tolist()
rightseqs = rightseqs.numpy().tolist()
for author in leftseqs:
for paper in author:
train_nodes.add(paper[0])
# if paper[0] != 0:
# train_nodes.add(paper[0])
for author in rightseqs:
for paper in author:
train_nodes.add(paper[0])
# if paper[0] != 0:
# train_nodes.add(paper[0])
return list(train_nodes)
def get_tensor(vectors, seqs):
out_tensor = torch.stack([torch.stack([vectors[paper[0].item()]
for paper in author], dim=0)
for author in seqs], dim=0)
return out_tensor
def get_train_neigh_list(neighbours):
train_neigh_list = []
source = []
target = []
for node in neighbours:
for neigh in neighbours[node]:
source.append(int(node))
target.append(neigh)
# print(len(source), len(target))
train_neigh_list.append(source)
train_neigh_list.append(target)
return train_neigh_list
def get_1hopsubgraph(feature_embedding, neighbours, train_nodes):
train_nodeset = set(train_nodes)
node_set = set(train_nodes)
source_list = []
target_list = []
for source in train_nodeset:
if source != 0 and str(source) in neighbours.keys():
neighbor_list = neighbours[str(source)]
for target in neighbor_list:
node_set.add(target)
source_list.append(source)
target_list.append(target)
source_list.append(target)
target_list.append(source)
assert len(source_list) == len(target_list)
# print(len(train_nodeset), len(node_set))
node_list = list(node_set)
node_indexmap = {}
for index, node in enumerate(node_list):
node_indexmap[node] = index
for i in range(len(source_list)):
source_list[i] = node_indexmap[source_list[i]]
target_list[i] = node_indexmap[target_list[i]]
x = feature_embedding[node_list]
edge_index = [source_list, target_list]
subgraph_data = Data(x=torch.FloatTensor(x),
edge_index=torch.LongTensor(edge_index))
return subgraph_data, node_indexmap
def get_2hopsubgraph(feature_embedding, neighbours, train_nodes):
train_nodeset = set(train_nodes)
hop1_nodeset = set()
source_list = []
target_list = []
for node in train_nodeset:
if node != 0 and str(node) in neighbours.keys():
hop1_neighborlist = neighbours[str(node)]
# random.shuffle(hop1_neighborlist)
# hop1_neighborlist = hop1_neighborlist[:25]
for hop1_neighbor in hop1_neighborlist:
hop1_nodeset.add(hop1_neighbor)
source_list.append(node)
target_list.append(hop1_neighbor)
source_list.append(hop1_neighbor)
target_list.append(node)
hop2_nodeset = set()
for node in hop1_nodeset:
if node != 0 and str(node) in neighbours.keys():
hop2_neighborlist = neighbours[str(node)]
# random.shuffle(hop2_neighborlist)
# hop2_neighborlist = hop2_neighborlist[:10]
for hop2_neighbor in hop2_neighborlist:
hop2_nodeset.add(hop2_neighbor)
source_list.append(node)
target_list.append(hop2_neighbor)
source_list.append(hop2_neighbor)
target_list.append(node)
# print(len(train_nodeset), len(hop1_nodeset), len(hop2_nodeset))
assert len(source_list) == len(target_list)
node_list = list(set.union(train_nodeset, hop1_nodeset, hop2_nodeset))
node_indexmap = {}
for index, node in enumerate(node_list):
node_indexmap[node] = index
for i in range(len(source_list)):
source_list[i] = node_indexmap[source_list[i]]
target_list[i] = node_indexmap[target_list[i]]
x = feature_embedding[node_list]
edge_index = [source_list, target_list]
subgraph_data = Data(x=torch.FloatTensor(x),
edge_index=torch.LongTensor(edge_index))
return subgraph_data, node_indexmap
def load_data(args):
neighbours = readjson(Jsondir, "neighbours")
feature_embedding = readjson(Jsondir, "topic_embedding")
all_neighbours = readjson(Jsondir, "meta_neighbours")
dataset_name = 'dataset' + args.train_test_ratio
dataset = readjson(Jsondir, dataset_name)
trainset = dataset["trainset"]
trainset = np.array(trainset)
testset = dataset["testset"]
testset = np.array(testset)
print("Data Loaded")
return feature_embedding, neighbours, all_neighbours, trainset, testset
def write_data(results, args):
import csv
results.sort(key=lambda x: (-x[1], -x[2], -x[3], -x[4], -x[5]))
if not os.path.exists(Csvdir):
os.makedirs(Csvdir)
# dataset_name = 'dataset' + args.train_test_ratio
f = open(os.path.join(Csvdir, args.model_name + ".csv"), "w")
writer = csv.writer(f)
writer.writerow(["epoch", "acc", "f1", "auc", "p", "r"])
for i in results:
tmp = (str(i[0]), str(i[1]), str(i[2]), str(i[3]), str(i[4]), str(i[5]))
writer.writerow(tmp)
f.close()