-
Notifications
You must be signed in to change notification settings - Fork 4
/
dataset.py
173 lines (154 loc) · 7.47 KB
/
dataset.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
from dgl.data import FraudDataset
from dgl.data.utils import load_graphs
import dgl
import torch
import warnings
import pickle as pkl
import torch
import matplotlib.pyplot as plt
import pandas as pd
import numpy as np
plt.rcParams['axes.unicode_minus']=False
import seaborn as sns
from dgl.nn.pytorch.conv import EdgeWeightNorm
import pickle as pkl
import dgl
import dgl.function as fn
warnings.filterwarnings("ignore")
class Dataset:
def __init__(self, load_epoch, name='tfinance', del_ratio=0., homo=True, data_path='', adj_type='sym'):
self.name = name
graph = None
prefix = data_path
if name == 'tfinance':
graph, label_dict = load_graphs(f'{prefix}/tfinance')
graph = graph[0]
graph.ndata['label'] = graph.ndata['label'].argmax(1)
if del_ratio != 0.:
graph = graph.add_self_loop()
with open(f'probs_tfinance_BWGNN_{load_epoch}_{homo}.pkl', 'rb') as f:
pred_y = pkl.load(f)
graph.ndata['pred_y'] = pred_y
graph = random_walk_update(graph, del_ratio, adj_type)
graph = dgl.remove_self_loop(graph)
elif name == 'tsocial':
graph, label_dict = load_graphs(f'{prefix}/tsocial')
graph = graph[0]
if del_ratio != 0.:
graph = graph.add_self_loop()
with open(f'probs_tsocial_BWGNN_{load_epoch}_{homo}.pkl', 'rb') as f:
pred_y = pkl.load(f)
graph.ndata['pred_y'] = pred_y
graph = random_walk_update(graph, del_ratio, adj_type)
graph = dgl.remove_self_loop(graph)
elif name == 'yelp':
dataset = FraudDataset(name, train_size=0.4, val_size=0.2)
graph = dataset[0]
if homo:
graph = dgl.to_homogeneous(dataset[0], ndata=['feature', 'label', 'train_mask', 'val_mask', 'test_mask'])
graph = dgl.add_self_loop(graph)
if del_ratio != 0.:
with open(f'probs_yelp_BWGNN_{load_epoch}_{homo}.pkl', 'rb') as f:
graph.ndata['pred_y'] = pkl.load(f)
graph = random_walk_update(graph, del_ratio, adj_type)
graph = dgl.add_self_loop(dgl.remove_self_loop(graph))
else:
if del_ratio != 0.:
with open(f'probs_yelp_BWGNN_{load_epoch}_{homo}.pkl', 'rb') as f:
pred_y = pkl.load(f)
data_dict = {}
flag = 1
for relation in graph.canonical_etypes:
graph_r = dgl.to_homogeneous(graph[relation], ndata=['feature', 'label', 'train_mask', 'val_mask', 'test_mask'])
graph_r = dgl.add_self_loop(graph_r)
graph_r.ndata['pred_y'] = pred_y
graph_r = random_walk_update(graph_r, del_ratio, adj_type)
graph_r = dgl.remove_self_loop(graph_r)
data_dict[('review', str(flag), 'review')] = graph_r.edges()
flag += 1
graph_new = dgl.heterograph(data_dict)
graph_new.ndata['label'] = graph.ndata['label']
graph_new.ndata['feature'] = graph.ndata['feature']
graph_new.ndata['train_mask'] = graph.ndata['train_mask']
graph_new.ndata['val_mask'] = graph.ndata['val_mask']
graph_new.ndata['test_mask'] = graph.ndata['test_mask']
graph = graph_new
elif name == 'amazon':
dataset = FraudDataset(name, train_size=0.4, val_size=0.2)
graph = dataset[0]
if homo:
graph = dgl.to_homogeneous(dataset[0], ndata=['feature', 'label', 'train_mask', 'val_mask', 'test_mask'])
graph = dgl.add_self_loop(graph)
if del_ratio != 0.:
with open(f'probs_amazon_BWGNN_{load_epoch}_{homo}.pkl', 'rb') as f:
graph.ndata['pred_y'] = pkl.load(f)
graph = random_walk_update(graph, del_ratio, adj_type)
graph = dgl.add_self_loop(dgl.remove_self_loop(graph))
else:
if del_ratio != 0.:
with open(f'probs_amazon_BWGNN_{load_epoch}_{homo}.pkl', 'rb') as f:
pred_y = pkl.load(f)
data_dict = {}
flag = 1
for relation in graph.canonical_etypes:
graph[relation].ndata['pred_y'] = pred_y
graph_r = dgl.add_self_loop(graph[relation])
graph_r = random_walk_update(graph_r, del_ratio, adj_type)
graph_r = dgl.remove_self_loop(graph_r)
data_dict[('review', str(flag), 'review')] = graph_r.edges()
flag += 1
graph_new = dgl.heterograph(data_dict)
graph_new.ndata['label'] = graph.ndata['label']
graph_new.ndata['feature'] = graph.ndata['feature']
graph_new.ndata['train_mask'] = graph.ndata['train_mask']
graph_new.ndata['val_mask'] = graph.ndata['val_mask']
graph_new.ndata['test_mask'] = graph.ndata['test_mask']
graph = graph_new
else:
print('no such dataset')
exit(1)
graph.ndata['label'] = graph.ndata['label'].long().squeeze(-1)
graph.ndata['feature'] = graph.ndata['feature'].float()
print(graph)
self.graph = graph
def random_walk_update(graph, delete_ratio, adj_type):
edge_weight = torch.ones(graph.num_edges())
if adj_type == 'sym':
norm = EdgeWeightNorm(norm='both')
else:
norm = EdgeWeightNorm(norm='left')
graph.edata['w'] = norm(graph, edge_weight)
# functions
aggregate_fn = fn.u_mul_e('h', 'w', 'm')
reduce_fn = fn.sum(msg='m', out='ay')
graph.ndata['h'] = graph.ndata['pred_y']
graph.update_all(aggregate_fn, reduce_fn)
graph.ndata['ly'] = graph.ndata['pred_y'] - graph.ndata['ay']
# graph.ndata['lyyl'] = torch.matmul(graph.ndata['ly'], graph.ndata['ly'].T)
graph.apply_edges(inner_product_black)
# graph.apply_edges(inner_product_white)
black = graph.edata['inner_black']
# white = graph.edata['inner_white']
# delete
threshold = int(delete_ratio * graph.num_edges())
edge_to_move = set(black.sort()[1][:threshold].tolist())
# edge_to_protect = set(white.sort()[1][-threshold:].tolist())
edge_to_protect = set()
graph_new = dgl.remove_edges(graph, list(edge_to_move.difference(edge_to_protect)))
return graph_new
def inner_product_black(edges):
return {'inner_black': (edges.src['ly'] * edges.dst['ly']).sum(axis=1)}
def inner_product_white(edges):
return {'inner_white': (edges.src['ay'] * edges.dst['ay']).sum(axis=1)}
def find_inter(edges):
return edges.src['label'] != edges.dst['label']
def cal_hetero(edges):
return {'same': edges.src['label'] != edges.dst['label']}
def cal_hetero_normal(edges):
return {'same_normal': (edges.src['label'] != edges.dst['label']) & (edges.src['label'] == 0)}
def cal_normal(edges):
return {'normal': edges.src['label'] == 0}
def cal_hetero_anomal(edges):
return {'same_anomal': (edges.src['label'] != edges.dst['label']) & (edges.src['label'] == 1)}
def cal_anomal(edges):
return {'anomal': edges.src['label'] == 1}