-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
166 lines (139 loc) · 7.97 KB
/
main.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
import argparse
import time
import random
import numpy as np
import torch
# import os
# os.environ["CUDA_VISIBLE_DEVICES"] = "7"
import rlctr.trainer as trainer
import rlctr.utils.tensor_utils as utils
def build_args():
parser = argparse.ArgumentParser(description='GraphNAS')
register_default_args(parser)
args = parser.parse_args()
return args
def register_default_args(parser):
parser.add_argument('--mode', type=str, default='train',
choices=['train', 'derive', 'random', 'bayes', 'finetuning'],
help='train: Training GraphNAS, derive: Deriving Architectures')
parser.add_argument('--random_seed', type=int, default=123)
parser.add_argument("--cuda", type=bool, default=True, required=False,
help="run in cuda mode")
parser.add_argument('--save_epoch', type=int, default=2)
parser.add_argument('--max_save_num', type=int, default=5)
# controller
parser.add_argument('--time_budget', type=float, default=5.0,help='time budget(h) for training controller.')
parser.add_argument('--layers_of_child_model', type=int, default=3)
parser.add_argument('--shared_initial_step', type=int, default=10)
parser.add_argument('--batch_size', type=int, default=64)
parser.add_argument('--entropy_mode', type=str, default='reward', choices=['reward', 'regularizer'])
parser.add_argument('--entropy_coeff', type=float, default=1e-4)
parser.add_argument('--shared_rnn_max_length', type=int, default=35)
parser.add_argument('--load_path', type=str, default='')
parser.add_argument('--search_mode', type=str, default='macro')
parser.add_argument('--search_space', type=str, default='CAD')
parser.add_argument('--format', type=str, default='two')
# parser.add_argument('--format', type=str, default='two')
parser.add_argument('--max_epoch', type=int, default=30)
parser.add_argument('--gnn_hidden', type=int, default=64)
parser.add_argument('--ema_baseline_decay', type=float, default=0.95)
parser.add_argument('--discount', type=float, default=1.0)
parser.add_argument('--controller_max_step', type=int, default=10,
help='step for controller parameters')
parser.add_argument('--controller_optim', type=str, default='adam')
parser.add_argument('--controller_lr', type=float, default=3.5e-4,
help="will be ignored if --controller_lr_cosine=True")
parser.add_argument('--controller_grad_clip', type=float, default=0)
parser.add_argument('--tanh_c', type=float, default=2.5)
parser.add_argument('--softmax_temperature', type=float, default=5.0)
parser.add_argument('--derive_num_sample', type=int, default=10)
parser.add_argument('--hyper_eval_inters', type=int, default=30)
parser.add_argument('--derive_finally', type=bool, default=True)
parser.add_argument('--derive_from_history', type=bool, default=True)
parser.add_argument('--controller_hid', type=int, default=100)
# child model
parser.add_argument("--early_stop_epoch", type=int, default=50,
help="epoch that valid loss bot decrease.")
parser.add_argument("--only_one_act_funtion", type=bool, default=False,
help="epoch that valid loss bot decrease.")
parser.add_argument("--shared_params", action='store_true', default=False,
help="shared_params between child model.")
parser.add_argument("--cos_lr", action='store_true', default=False,
help="use cos lr in training stage")
parser.add_argument("--ln", action='store_true', default=False,
help="layer norm")
parser.add_argument("--gpu", type=int, default=1,
help="gpu number")
parser.add_argument("--without_jk", type=bool, default=False,
help="without_jk: remove jk in snag.")
parser.add_argument("--dataset", type=str, default="catA", required=False,
help="The input dataset.")
parser.add_argument("--epochs", type=int, default=800,
help="number of training epochs")
parser.add_argument("--train_epochs", type=int, default=800,
help="number of controller training epoch.")
parser.add_argument("--gnn_epochs", type=int, default=50,
help="number of training gnn epochs")
parser.add_argument("--retrain_epochs", type=int, default=800,
help="number of training epochs")
parser.add_argument("--multi_label", type=bool, default=False,
help="multi_label or single_label task")
parser.add_argument("--residual", action="store_false",
help="use residual connection")
parser.add_argument("--in_drop", type=float, default=0.6,
help="input feature dropout")
parser.add_argument("--lr", type=float, default=0.005,
help="learning rate")
parser.add_argument("--param_file", type=str, default="cora_test.pkl",
help="learning rate")
parser.add_argument("--optim_file", type=str, default="opt_cora_test.pkl",
help="optimizer save path")
parser.add_argument('--weight_decay', type=float, default=5e-4)
parser.add_argument('--max_param', type=float, default=5E6)
parser.add_argument('--supervised', type=bool, default=False)
parser.add_argument('--submanager_log_file', type=str, default=f"sub_manager_logger_file_{time.time()}.txt")
parser.add_argument("--show_gnn_training_info", action='store_true', default=True,
help="show gnn training details")
parser.add_argument('--gnn_hidden_dim', type=int, default=64)
parser.add_argument('--task', type=str, default='rating_pred')
parser.add_argument('--out_path', type=str, default='output/')
parser.add_argument('--out_file', type=str, default='default.csv')
parser.add_argument('--finetune_num', type=int, default=10)
parser.add_argument("--best_test_from_val", action='store_true', default=True,
help="best searched test performance according to val results")
parser.add_argument("--load_ckpt", action='store_true', default=False)
parser.add_argument('--ckpt_path', type=str)
parser.add_argument('--catalog_name', type=str, default='catA', help='catalog name')
parser.add_argument('--data_location', type=str, default='./data/', help='data path')
parser.add_argument('--embedding_size', type=str, default='one_hot', help='embedding type')
parser.add_argument('--train_batch_size', type=int, default=1024, help='eval batch size')
parser.add_argument('--eval_batch_size', type=int, default=1024, help='eval batch size')
parser.add_argument('--dropout', type=float, default=0.3, help='dropout for model')
parser.add_argument('--hidden_size', type=int, default=2048, help='default hidden_size in supernet')
parser.add_argument('--embedding_keyword', type=str, default='embedding', help='data path')
parser.add_argument('--num_layers', type=int, default=3, help='num of layers of GNN method.')
def main(args): # pylint:disable=redefined-outer-name
if args.cuda and not torch.cuda.is_available(): # cuda is not available
args.cuda = False
torch.cuda.set_device(args.gpu)
random.seed(args.random_seed)
np.random.seed(args.random_seed)
torch.manual_seed(args.random_seed)
# if args.cuda:
# torch.cuda.manual_seed(args.random_seed)
utils.makedirs(args.dataset)
trnr = trainer.Trainer(args)
if args.mode =='train':
print(args)
trnr.train()
elif args.mode in ['random', 'bayes']:
print(args)
trnr.random_bayes_search(mode=args.mode, max_evals=args.train_epochs * args.controller_max_step)
elif args.mode == 'derive':
trnr.derive()
else:
raise Exception(f"[!] Mode not found: {args.mode}")
trnr.save_training_dict()
if __name__ == "__main__":
args = build_args()
main(args)