-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrun.py
140 lines (118 loc) · 6.12 KB
/
run.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
import os
import torch
import numpy as np
import torch.nn as nn
import argparse
import configparser
from datetime import datetime
from lib.TrainInits import init_seed
from lib.dataloader import get_dataloader
from lib.TrainInits import print_model_parameters
from lib.metrics import MAE_torch
from utils import normalized_network, get_adjacency_matrix
from train import Trainer
from model import SPA
def masked_mae_loss(scaler, mask_value):
def loss(preds, labels):
if scaler:
preds = scaler.inverse_transform(preds)
labels = scaler.inverse_transform(labels)
mae = MAE_torch(pred=preds, true=labels, mask_value=mask_value)
return mae
return loss
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--debug', default=True, type=eval)
parser.add_argument('--dataset', default='PeMSD4', type=str)
parser.add_argument('--model', default='SPA', type=str)
#train
parser.add_argument('--device',type=str,default='cuda:0',help='')
parser.add_argument('--loss_func', default='mae', type=str)
parser.add_argument('--seed', default=10, type=int)
parser.add_argument('--batch_size',type=int,default=64,help='batch size')
parser.add_argument('--epochs',type=int,default=200,help='')
parser.add_argument('--lr_init', default=0.001, type=float)
parser.add_argument('--lr_decay', default=False, type=eval)
parser.add_argument('--lr_decay_rate', default=0.3, type=float)
parser.add_argument('--lr_decay_step', default='5,20,40,70', type=str)
parser.add_argument('--early_stop', default=True, type=eval)
parser.add_argument('--early_stop_patience', default=15, type=int)
parser.add_argument('--grad_norm', default=False, type=eval)
parser.add_argument('--max_grad_norm', default=5, type=int)
parser.add_argument('--teacher_forcing', default=False, type=bool)
parser.add_argument('--real_value', default=True, type=eval, help = 'use real value for loss calculation')
parser.add_argument('--weight_decay',type=float,default=0.0001,help='weight decay rate')
#test
parser.add_argument('--mae_thresh', default=None, type=eval)
parser.add_argument('--mape_thresh', default=0., type=float)
#log
parser.add_argument('--log_dir', default='./', type=str)
parser.add_argument('--log_step', default=20, type=int)
parser.add_argument('--plot', default=False, type=eval)
# model
parser.add_argument('--in_channels',type=int,default=1,help='inputs dimension')
parser.add_argument('--embed_size',type=int,default=64,help='')
parser.add_argument('--time_num',type=int,default=288,help='')
parser.add_argument('--heads',type=int,default=2,help='')
parser.add_argument('--num_layers',type=int,default=4,help='')
parser.add_argument('--forward_expansion',type=int,default=4,help='')
parser.add_argument('--t_dropout',type=int,default=0,help='')
parser.add_argument('--kernel_size',type=int,default=3,help='')
parser.add_argument('--a_dropout',type=int,default=0.3,help='')
# data
parser.add_argument('--val_ratio', default=0.2, type=float)
parser.add_argument('--test_ratio', default=0.2, type=float)
parser.add_argument('--lag', default=12, type=int)
parser.add_argument('--horizon', default=12, type=int)
parser.add_argument('--num_nodes', default=307, type=int)
parser.add_argument('--tod', default=False, type=eval)
parser.add_argument('--normalizer', default='std', type=str)
parser.add_argument('--column_wise', default=False, type=eval)
parser.add_argument('--default_graph', default=False, type=eval)
parser.add_argument('--input_dim', default=1, type=int)
parser.add_argument('--output_dim', default=1, type=int)
parser.add_argument('--e_dim', default=10, type=int)
args = parser.parse_args()
init_seed(args.seed)
device = torch.device(args.device)
train_loader, val_loader, test_loader, scaler = get_dataloader(args,
normalizer=args.normalizer,
tod=args.tod, dow=False,
weather=False, single=False)
if args.default_graph:
adjinit, distance = get_adjacency_matrix(('./data/{}/distance.csv').format(args.dataset), args.num_nodes)
else:
adjinit = None
if args.model == 'SPA':
model = SPA(num_nodes=args.num_nodes, in_channels = args.in_channels, embed_size = args.embed_size, time_num = args.time_num, num_layers = args.num_layers,
T_dim = args.lag, output_T_dim = args.horizon, heads = args.heads, forward_expansion = args.forward_expansion,
t_dropout = args.t_dropout, kernel_size = args.kernel_size, a_dropout = args.a_dropout, e_dim=args.e_dim)
model = model.to(device)
print_model_parameters(model, only_num=False)
#init loss function, optimizer
if args.loss_func == 'mask_mae':
loss = masked_mae_loss(scaler, mask_value=0.0)
elif args.loss_func == 'mae':
loss = torch.nn.L1Loss().to(args.device)
elif args.loss_func == 'mse':
loss = torch.nn.MSELoss().to(args.device)
else:
raise ValueError
optimizer = torch.optim.Adam(params=model.parameters(), lr=args.lr_init,
weight_decay=args.weight_decay)
#learning rate decay
lr_scheduler = None
if args.lr_decay:
print('Applying learning rate decay.')
lr_decay_steps = [int(i) for i in list(args.lr_decay_step.split(','))]
lr_scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer=optimizer, milestones=lr_decay_steps, gamma=args.lr_decay_rate)
#config log path
current_time = datetime.now().strftime('%Y%m%d%H%M%S')
current_dir = os.path.join('./log')
log_dir = os.path.join(current_dir,'experiments', args.dataset, '{}_{}_{}_{}_{}_{}_{}'.format(
current_time, args.model, args.lr_init, args.input_dim, args.num_nodes, args.kernel_size, args.num_layers))
args.log_dir = log_dir
#start training
trainer = Trainer(model, loss, optimizer, train_loader, val_loader, test_loader, scaler,
args, lr_scheduler=lr_scheduler)
trainer.train()