-
Notifications
You must be signed in to change notification settings - Fork 78
/
Copy pathtrain_student.py
168 lines (134 loc) · 6.42 KB
/
train_student.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
# !/usr/bin/env python
# -*- encoding: utf-8 -*-
import os
# os.environ['CUDA_VISIBLE_DEVICES'] = '0'
# os.environ['TL_BACKEND'] = 'torch'
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
# 0:Output all; 1:Filter out INFO; 2:Filter out INFO and WARNING; 3:Filter out INFO, WARNING, and ERROR
import yaml
import argparse
import tensorlayerx as tlx
from gammagl.datasets import Planetoid, Amazon
from gammagl.models import MLP
from gammagl.utils import mask_to_index
from tensorlayerx.model import TrainOneStep, WithLoss
class SemiSpvzLoss(WithLoss):
def __init__(self, net, loss_fn):
super(SemiSpvzLoss, self).__init__(backbone=net, loss_fn=loss_fn)
def forward(self, data, teacher_logits):
student_logits = self.backbone_network(data['x'])
train_y = tlx.gather(data['y'], data['t_idx'])
train_teacher_logits = tlx.gather(teacher_logits, data['t_idx'])
train_student_logits = tlx.gather(student_logits, data['t_idx'])
loss = self._loss_fn(train_y, train_student_logits, train_teacher_logits, args.lamb)
return loss
def get_training_config(config_path, model_name, dataset):
with open(config_path, "r") as conf:
full_config = yaml.load(conf, Loader=yaml.FullLoader)
dataset_specific_config = full_config["global"]
model_specific_config = full_config[dataset][model_name]
if model_specific_config is not None:
specific_config = dict(dataset_specific_config, **model_specific_config)
else:
specific_config = dataset_specific_config
specific_config["model_name"] = model_name
return specific_config
def calculate_acc(logits, y, metrics):
metrics.update(logits, y)
rst = metrics.result()
metrics.reset()
return rst
def kl_divergence(teacher_logits, student_logits):
# convert logits to probabilities
teacher_probs = tlx.softmax(teacher_logits)
student_probs = tlx.softmax(student_logits)
# compute KL divergence
kl_div = tlx.reduce_sum(teacher_probs * (tlx.log(teacher_probs+1e-10) - tlx.log(student_probs+1e-10)), axis=-1)
return tlx.reduce_mean(kl_div)
def cal_mlp_loss(labels, student_logits, teacher_logits, lamb):
loss_l = tlx.losses.softmax_cross_entropy_with_logits(student_logits, labels)
loss_t = kl_divergence(teacher_logits, student_logits)
return lamb * loss_l + (1 - lamb) * loss_t
def train_student(args):
# load datasets
if str.lower(args.dataset) not in ['cora','pubmed','citeseer','computers','photo']:
raise ValueError('Unknown dataset: {}'.format(args.dataset))
if args.dataset in ['cora', 'pubmed', 'citeseer']:
dataset = Planetoid(args.dataset_path, args.dataset)
elif args.dataset == 'computers':
dataset = Amazon(args.dataset_path, args.dataset, train_ratio=200/13752, val_ratio=(200/13752)*1.5)
elif args.dataset == 'photo':
dataset = Amazon(args.dataset_path, args.dataset, train_ratio=160/7650, val_ratio=(160/7650)*1.5)
graph = dataset[0]
# load teacher_logits from .npy file
teacher_logits = tlx.files.load_npy_to_any(path = r'./', name = f'{args.dataset}_{args.teacher}_logits.npy')
teacher_logits = tlx.ops.convert_to_tensor(teacher_logits)
# for mindspore, it should be passed into node indices
train_idx = mask_to_index(graph.train_mask)
test_idx = mask_to_index(graph.test_mask)
val_idx = mask_to_index(graph.val_mask)
t_idx = tlx.concat([train_idx, test_idx, val_idx], axis=0)
net = MLP(in_channels=dataset.num_node_features,
hidden_channels=conf["hidden_dim"],
out_channels=dataset.num_classes,
num_layers=conf["num_layers"],
act=tlx.nn.ReLU(),
norm=None,
dropout=float(conf["dropout_ratio"]))
optimizer = tlx.optimizers.Adam(lr=conf["learning_rate"], weight_decay=conf["weight_decay"])
metrics = tlx.metrics.Accuracy()
train_weights = net.trainable_weights
loss_func = SemiSpvzLoss(net, cal_mlp_loss)
train_one_step = TrainOneStep(loss_func, optimizer, train_weights)
data = {
"x": graph.x,
"y": graph.y,
"train_idx": train_idx,
"test_idx": test_idx,
"val_idx": val_idx,
"t_idx": t_idx
}
best_val_acc = 0
for epoch in range(args.n_epoch):
net.set_train()
train_loss = train_one_step(data, teacher_logits)
net.set_eval()
logits = net(data['x'])
val_logits = tlx.gather(logits, data['val_idx'])
val_y = tlx.gather(data['y'], data['val_idx'])
val_acc = calculate_acc(val_logits, val_y, metrics)
print("Epoch [{:0>3d}] ".format(epoch+1)\
+ " train loss: {:.4f}".format(train_loss.item())\
+ " val acc: {:.4f}".format(val_acc))
# save best model on evaluation set
if val_acc > best_val_acc:
best_val_acc = val_acc
net.save_weights(args.best_model_path+args.dataset+"_"+args.teacher+"_MLP.npz", format='npz_dict')
net.load_weights(args.best_model_path+args.dataset+"_"+args.teacher+"_MLP.npz", format='npz_dict')
net.set_eval()
logits = net(data['x'])
test_logits = tlx.gather(logits, data['test_idx'])
test_y = tlx.gather(data['y'], data['test_idx'])
test_acc = calculate_acc(test_logits, test_y, metrics)
print("Test acc: {:.4f}".format(test_acc))
if __name__ == '__main__':
# parameters setting
parser = argparse.ArgumentParser()
parser.add_argument("--model_config_path",type=str,default="./train.conf.yaml",help="path to modelconfigeration")
parser.add_argument("--teacher", type=str, default="SAGE", help="teacher model")
parser.add_argument("--lamb", type=float, default=0, help="parameter balances loss from hard labels and teacher outputs")
parser.add_argument("--n_epoch", type=int, default=200, help="number of epoch")
parser.add_argument('--dataset', type=str, default="cora", help="dataset")
parser.add_argument("--dataset_path", type=str, default=r'./data', help="path to save dataset")
parser.add_argument("--best_model_path", type=str, default=r'./', help="path to save best model")
parser.add_argument("--gpu", type=int, default=0)
args = parser.parse_args()
conf = {}
if args.model_config_path is not None:
conf = get_training_config(args.model_config_path, args.teacher, args.dataset)
conf = dict(args.__dict__, **conf)
if args.gpu >= 0:
tlx.set_device("GPU", args.gpu)
else:
tlx.set_device("CPU")
train_student(args)