-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathdistill.py
159 lines (132 loc) · 6.34 KB
/
distill.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
import json
import argparse
from collections import Counter
import numpy as np
import torch
from utils import *
from dataset import get_dataset, get_largest_cc, load_eigen, Transd2Ind, DataGraphSAINT
import agent as agent
import importlib
importlib.reload(agent)
from agent import GraphAgent
def middle_train():
args = parser.parse_args([])
args.reduction_rate = args.reduction_rate * 0.5
section = f"{args.dataset}-{str(args.reduction_rate)}"
with open(args.config, "r") as config_file:
config = json.load(config_file)
if section in config:
config = config[section]
for key, value in config.items():
setattr(args, key, value)
print(args)
torch.cuda.set_device(args.gpu_id)
seed_everything(args.seed)
data_graphsaint = ['flickr', 'reddit', 'ogbn-arxiv']
if args.dataset in data_graphsaint:
data = DataGraphSAINT(args.dataset)
else:
data_full = get_dataset(args.dataset, args.normalize_features)
data = Transd2Ind(data_full)
dataset_dir = f"./data/{args.dataset}"
idx_lcc = np.load(f"{dataset_dir}/idx_lcc.npy")
idx_train_lcc = np.load(f"{dataset_dir}/idx_train_lcc.npy")
idx_map = np.load(f"{dataset_dir}/idx_map.npy")
eigenvals_lcc, eigenvecs_lcc = load_eigen(args.dataset)
eigenvals_lcc = torch.FloatTensor(eigenvals_lcc)
eigenvecs_lcc = torch.FloatTensor(eigenvecs_lcc)
n_syn = int(len(data.idx_train) * args.reduction_rate)
args.eigen_k = args.eigen_k if args.eigen_k < n_syn else n_syn
# 计算需要的特征值和特征基(真实图)
eigenvals, eigenvecs = get_syn_eigen(real_eigenvals=eigenvals_lcc, real_eigenvecs=eigenvecs_lcc,
eigen_k=args.eigen_k, ratio=args.ratio)
# 计算对应协方差矩阵
co_x_trans_real = get_subspace_covariance_matrix(eigenvecs, data.x_full[idx_lcc]) # kdd
# 计算嵌入结果
embed_sum = get_embed_sum(eigenvals=eigenvals, eigenvecs=eigenvecs, x=data.x_full[idx_lcc])
embed_sum = embed_sum[idx_map, :]
embed_mean_real = get_embed_mean(embed_sum=embed_sum, label=data.y_full[idx_train_lcc])
data = data.cuda()
eigenvals = eigenvals.cuda()
co_x_trans_real = co_x_trans_real.cuda()
embed_mean_real = embed_mean_real.cuda()
accs = []
for ep in range(args.runs):
args.expID = ep
agent = GraphAgent(args, data)
acc = agent.train_middle(eigenvals, co_x_trans_real, embed_mean_real)
accs.append(acc)
def mls_train():
args = parser.parse_args([])
args.reduction_rate = args.reduction_rate
print(args.reduction_rate)
section = f"{args.dataset}-{str(args.reduction_rate)}"
with open(args.config, "r") as config_file:
config = json.load(config_file)
if section in config:
config = config[section]
for key, value in config.items():
setattr(args, key, value)
print(args)
torch.cuda.set_device(args.gpu_id)
seed_everything(args.seed)
data_graphsaint = ['flickr', 'reddit', 'ogbn-arxiv']
if args.dataset in data_graphsaint:
data = DataGraphSAINT(args.dataset)
else:
data_full = get_dataset(args.dataset, args.normalize_features)
data = Transd2Ind(data_full)
dataset_dir = f"./data/{args.dataset}"
idx_lcc = np.load(f"{dataset_dir}/idx_lcc.npy")
idx_train_lcc = np.load(f"{dataset_dir}/idx_train_lcc.npy")
idx_map = np.load(f"{dataset_dir}/idx_map.npy")
eigenvals_lcc, eigenvecs_lcc = load_eigen(args.dataset)
eigenvals_lcc = torch.FloatTensor(eigenvals_lcc)
eigenvecs_lcc = torch.FloatTensor(eigenvecs_lcc)
n_syn = int(len(data.idx_train) * args.reduction_rate)
args.eigen_k = args.eigen_k if args.eigen_k < n_syn else n_syn
# 计算需要的特征值和特征基(真实图)
eigenvals, eigenvecs = get_syn_eigen(real_eigenvals=eigenvals_lcc, real_eigenvecs=eigenvecs_lcc,
eigen_k=args.eigen_k, ratio=args.ratio)
eigenvals_middle, eigenvecs_middle = get_syn_eigen(real_eigenvals=eigenvals_lcc, real_eigenvecs=eigenvecs_lcc,
eigen_k=args.eigen_k_middle, ratio=args.ratio_middle)
# 计算对应协方差矩阵
co_x_trans_real = get_subspace_covariance_matrix(eigenvecs, data.x_full[idx_lcc]) # kdd
co_x_trans_real_middle = get_subspace_covariance_matrix(eigenvecs_middle, data.x_full[idx_lcc])
# 计算嵌入结果
embed_sum = get_embed_sum(eigenvals=eigenvals, eigenvecs=eigenvecs, x=data.x_full[idx_lcc])
embed_sum = embed_sum[idx_map, :]
embed_mean_real = get_embed_mean(embed_sum=embed_sum, label=data.y_full[idx_train_lcc])
embed_sum_middle = get_embed_sum(eigenvals=eigenvals_middle, eigenvecs=eigenvecs_middle, x=data.x_full[idx_lcc])
embed_sum_middle = embed_sum_middle[idx_map, :]
embed_mean_real_middle = get_embed_mean(embed_sum=embed_sum_middle, label=data.y_full[idx_train_lcc])
data = data.cuda()
eigenvals = eigenvals.cuda()
co_x_trans_real = co_x_trans_real.cuda()
embed_mean_real = embed_mean_real.cuda()
eigenvals_middle = eigenvals_middle.cuda()
co_x_trans_real_middle = co_x_trans_real_middle.cuda()
embed_mean_real_middle = embed_mean_real_middle.cuda()
accs = []
for ep in range(args.runs):
args.expID = ep
agent = GraphAgent(args, data)
acc = agent.train_bottom(eigenvals_middle, co_x_trans_real_middle, embed_mean_real_middle)
acc = agent.train_top(eigenvals, co_x_trans_real, embed_mean_real)
accs.append(acc)
parser = argparse.ArgumentParser()
parser.add_argument("--gpu_id", type=int, default=0, help="gpu id")
parser.add_argument("--seed", type=int, default=15)
parser.add_argument("--config", type=str, default='./config/config_distill.json')
parser.add_argument("--runs", type=int, default=1)
parser.add_argument("--dataset", type=str, default="citeseer") # [citeseer, cora, ogbn-arxiv, flickr, reddit]
parser.add_argument("--normalize_features", type=bool, default=True)
parser.add_argument("--reduction_rate", type=float, default=0.5)
parser.add_argument("--k", type=int, default=2)
parser.add_argument("--evaluate_gnn", type=str, default="GCN")
parser.add_argument("--epoch_gnn", type=int, default=2000)
parser.add_argument("--nlayers", type=float, default=2)
parser.add_argument("--hidden_dim", type=int, default=256)
parser.add_argument("--use_mine",type=bool,default=False)
middle_train()
mls_train()