-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathrun_stode.py
162 lines (129 loc) · 6.01 KB
/
run_stode.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
import os
import random
import numpy as np
import torch
import torch.nn as nn
from torch.optim.lr_scheduler import StepLR, OneCycleLR, MultiStepLR
import time
from tqdm import tqdm
from loguru import logger
from args import args
from model import ODEGCN
from utils import generate_dataset, read_data, get_normalized_adj
from eval import masked_mae_np, masked_mape_np, masked_rmse_np
def logcosh(true, pred):
loss = torch.log(torch.cosh(pred - true))
return torch.mean(loss)
def train(loader, model, optimizer, criterion, std, mean, device):
batch_rmse_loss = 0
batch_mae_loss = 0
batch_mape_loss = 0
batch_loss = 0
for idx, (inputs, targets) in enumerate(tqdm(loader)):
model.train()
optimizer.zero_grad()
inputs = inputs.to(device)
targets = targets.to(device)
outputs = model(inputs) * std + mean
loss = criterion(outputs, targets)
loss.backward()
optimizer.step()
out_unnorm = outputs.detach().cpu().numpy()
target_unnorm = targets.detach().cpu().numpy()
mae_loss = masked_mae_np(target_unnorm, out_unnorm, 0.0)
rmse_loss = masked_rmse_np(target_unnorm, out_unnorm, 0.0)
mape_loss = masked_mape_np(target_unnorm, out_unnorm, 0.0)
batch_rmse_loss += rmse_loss
batch_mae_loss += mae_loss
batch_mape_loss += mape_loss
batch_loss += loss.detach().cpu().item()
return batch_loss / (idx + 1), batch_rmse_loss / (idx + 1), batch_mae_loss / (idx + 1), batch_mape_loss / (idx + 1)
@torch.no_grad()
def eval(loader, model, std, mean, device):
batch_rmse_loss = 0
batch_mae_loss = 0
batch_mape_loss = 0
for idx, (inputs, targets) in enumerate(tqdm(loader)):
model.eval()
inputs = inputs.to(device)
targets = targets.to(device)
output = model(inputs)
out_unnorm = output.detach().cpu().numpy() * std + mean
target_unnorm = targets.detach().cpu().numpy()
mae_loss = masked_mae_np(target_unnorm, out_unnorm, 0.0)
rmse_loss = masked_rmse_np(target_unnorm, out_unnorm, 0.0)
mape_loss = masked_mape_np(target_unnorm, out_unnorm, 0.0)
batch_rmse_loss += rmse_loss
batch_mae_loss += mae_loss
batch_mape_loss += mape_loss
return batch_rmse_loss / (idx + 1), batch_mae_loss / (idx + 1), batch_mape_loss / (idx + 1)
def main(args):
# random seed
# seed = 2
np.random.seed(args.seed)
torch.manual_seed(args.seed)
torch.cuda.manual_seed(args.seed)
random.seed(args.seed)
device = torch.device('cuda:' + str(args.num_gpu)) if torch.cuda.is_available() else torch.device('cpu')
if args.log:
logger.add('log_{time}.log')
options = vars(args)
if args.log:
logger.info(options)
else:
print(options)
data, mean, std, dtw_matrix, sp_matrix = read_data(args)
train_loader, valid_loader, test_loader, train_mean, train_std, val_mean, val_std, test_mean, test_std = generate_dataset(
data, args)
print('mean,std: ', train_mean, train_std, val_mean, val_std)
A_sp_wave = get_normalized_adj(sp_matrix).to(device)
A_se_wave = get_normalized_adj(dtw_matrix).to(device)
net = ODEGCN(num_nodes=data.shape[1],
num_features=data.shape[2],
num_timesteps_input=args.his_length,
num_timesteps_output=args.pred_length,
A_sp_hat=A_sp_wave,
A_se_hat=A_se_wave)
net = net.to(device)
lr = args.lr
optimizer = torch.optim.AdamW(net.parameters(), lr=lr)
criterion = nn.SmoothL1Loss()
scheduler = MultiStepLR(optimizer=optimizer,
milestones=args.lr_decay_steps,
gamma=args.lr_decay_rate)
val_mape_min = float('inf')
wait=0
if not os.path.exists(args.save):
os.makedirs(args.save)
for epoch in range(1, args.epochs + 1):
print("=====Epoch {}=====".format(epoch))
print('Training...')
loss, train_rmse, train_mae, train_mape = train(train_loader, net, optimizer, criterion, train_std, train_mean,
device)
print('Evaluating...')
valid_rmse, valid_mae, valid_mape = eval(valid_loader, net, val_std, val_mean, device)
if valid_mape <= val_mape_min:
print(f'\n##on train data## loss: {loss}, \n' +
f'##on train data## rmse loss: {train_rmse}, mae loss: {train_mae}, mape loss: {train_mape}\n' +
f'##on valid data## rmse loss: {valid_rmse}, mae loss: {valid_mae}, mape loss: {valid_mape}\n')
print(f'save model to {args.save + "epoch_" + str(epoch) + "_" + str(round(valid_mape.item(), 2)) + "_best_model.pth"}\n')
wait = 0
val_mape_min = valid_mape
best_model_wts = net.state_dict()
torch.save(best_model_wts,
args.save + "epoch_" + str(epoch) + "_" + str(round(val_mape_min.item(), 2)) + "_best_model.pth")
last_weight_add=args.save + "epoch_" + str(epoch) + "_" + str(round(val_mape_min.item(), 2)) + "_best_model.pth"
else:
print(f'\n##on train data## loss: {loss}, \n' +
f'##on train data## rmse loss: {train_rmse}, mae loss: {train_mae}, mape loss: {train_mape}\n' +
f'##on valid data## rmse loss: {valid_rmse}, mae loss: {valid_mae}, mape loss: {valid_mape}\n')
wait += 1
if wait==30:
print(f'can not have better mape, so ends now\n')
break
scheduler.step()
net.load_state_dict(torch.load(last_weight_add))
test_rmse, test_mae, test_mape = eval(test_loader, net, test_std, test_mean, device)
print(f'##on test data## rmse loss: {test_rmse}, mae loss: {test_mae}, mape loss: {test_mape}')
if __name__ == '__main__':
main(args)