-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathtrain_distil.py
174 lines (159 loc) · 8.41 KB
/
train_distil.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
169
170
171
172
173
174
from util import *
import torch
import torch.nn as nn
import numpy as np
import argparse
from torch.utils.data import DataLoader
from model import TSFormer, DistilTSFormer
from engine import distil_trainer
import time
import os
import random
parser = argparse.ArgumentParser()
parser.add_argument("--device", type=str, default='cuda:0')
parser.add_argument('--sdata', type=str, default='../data/METR-LA')
parser.add_argument('--tdata', type=str, default='../data/PEMSD7M')
parser.add_argument("--long_his", type=int, default=2016)
parser.add_argument('--short_his', type=int, default=288*3) # 3 days as a temporary setting
parser.add_argument('--batch_size', type=int, default=12)
parser.add_argument('--output_len', type=int, default=12)
parser.add_argument('--mask_ratio', type=float, default=0.75)
parser.add_argument('--embed_dim', type=int, default=96)
parser.add_argument('--patch_size', type=int, default=12)
parser.add_argument("--encoder_depth", type=int, default=4, help='Number of transformer blocks in encoder')
parser.add_argument("--decoder_depth", type=int, default=1, help='Number of transformer blocks in decoder')
parser.add_argument("--num_heads", type=int, default=4, help='Number of attention heads')
parser.add_argument('--dropout', type=float, default=0.15)
parser.add_argument('--num_token', type=int, default=168, help='Number of tokens in a sequence')
parser.add_argument('--in_channel', type=int, default=1)
parser.add_argument('--mlp_ratio', type=int, default=4, help='Width of MLP w.r.t embed_dim')
parser.add_argument('--learning_rate', type=float, default=5e-4)
parser.add_argument('--weight_decay', type=float, default=0)
parser.add_argument('--num_epoch', type=int, default=50)
parser.add_argument('--print_every', type=int, default=50)
parser.add_argument('--expid', type=str, default='test')
parser.add_argument('--data_number', type=int, default=0, help='Target Data Number. 0 for all data')
parser.add_argument('--teacher_model_path', type=str, default=None, help='Path to the teacher model')
parser.add_argument("--student_model", type=str, help='[TSFormer, DistilFormer]')
parser.add_argument('--lambda_d', type=float, default=1, help='Weight of distilation loss')
args = parser.parse_args()
def main(args):
train_dataset_s = TimeSeriesForecastingDataset(args.sdata+'/data_in2016_out12.pkl',
args.sdata+'/index_in2016_out12.pkl', mode='train')
train_dataset_t = TimeSeriesForecastingDataset(args.tdata+'/data_in2016_out12.pkl',
args.tdata+'/index_in2016_out12.pkl', mode='train', data_number=args.data_number)
val_dataset_s = TimeSeriesForecastingDataset(args.sdata+'/data_in2016_out12.pkl',
args.sdata+'/index_in2016_out12.pkl', mode='valid')
val_dataset_t = TimeSeriesForecastingDataset(args.tdata+'/data_in2016_out12.pkl',
args.tdata+'/index_in2016_out12.pkl', mode='valid')
test_dataset_s = TimeSeriesForecastingDataset(args.sdata+'/data_in2016_out12.pkl',
args.sdata+'/index_in2016_out12.pkl', mode='test')
test_dataset_t = TimeSeriesForecastingDataset(args.tdata+'/data_in2016_out12.pkl',
args.tdata+'/index_in2016_out12.pkl', mode='test')
print("Source data length: train %d, valid %d, test %d; Num node %d" % (len(train_dataset_s),
len(val_dataset_s), len(test_dataset_s), train_dataset_s.data.shape[1]))
print("Target data length: train %d, valid %d, test %d; Num node %d" % (len(train_dataset_t),
len(val_dataset_t), len(test_dataset_t), train_dataset_t.data.shape[1]))
scaler_pkl_s = load_pkl(args.sdata+'/scaler_in2016_out12.pkl')
scaler_pkl_t = load_pkl(args.tdata+'/scaler_in2016_out12.pkl')
mean_s, std_s = scaler_pkl_s['args']['mean'], scaler_pkl_s['args']['std']
scaler_s = Scaler(mean_s, std_s)
mean_t, std_t = scaler_pkl_t['args']['mean'], scaler_pkl_t['args']['std']
scaler_t = Scaler(mean_t, std_t)
train_loader_s = DataLoader(train_dataset_s, batch_size=args.batch_size, shuffle=True)
train_loader_t = DataLoader(train_dataset_t, batch_size=args.batch_size, shuffle=True)
val_loader_s = DataLoader(val_dataset_s, batch_size=args.batch_size)
val_loader_t = DataLoader(val_dataset_t, batch_size=args.batch_size)
test_loader_s = DataLoader(test_dataset_s, batch_size=args.batch_size)
test_loader_t = DataLoader(test_dataset_t, batch_size=args.batch_size)
engine = distil_trainer(args, scaler_s, scaler_t)
print("Distilling TSFormer from %s to %s with %d target data" % (args.sdata, args.tdata, args.data_number))
if 'METR-LA' in args.sdata:
sdata_prefix = 'METR-LA'
elif 'PEMS-BAY' in args.sdata:
sdata_prefix = 'PEMS-BAY'
if 'PEMSD7M' in args.tdata:
save_distilformer = 'garage_nf/PEMSD7M/' + sdata_prefix+'_'+args.expid + '/'
if not os.path.exists(save_distilformer):
os.makedirs(save_distilformer)
elif 'HKTSM' in args.tdata:
save_distilformer = 'garage_nf/HKTSM/' + sdata_prefix+'_'+args.expid + '/'
if not os.path.exists(save_distilformer):
os.makedirs(save_distilformer)
val_distil_t = []
val_mae = []
val_rmse = []
val_mape = []
train_distil_s = []
train_distil_t = []
train_mae = []
train_time = []
val_time = []
for ep in range(1, args.num_epoch+1):
epoch_distil_s = []
epoch_distil_t = []
epoch_mae = []
epoch_valdistil_t = []
epoch_valmae = []
epoch_valrmse = []
epoch_valmape = []
s1 = time.time()
for i, (_, xt) in enumerate(train_loader_t):
# print('xt', xt.shape)
batch_size = xt.shape[0]
xs = sample_batch(train_dataset_s, batch_size)
xt = xt.to(args.device)
xs = xs.to(args.device)
metrics = engine.train_distil(xs, xt)
epoch_distil_s.append(metrics[0])
epoch_distil_t.append(metrics[1])
epoch_mae.append(metrics[2])
if i % args.print_every == 0:
print("Distil epoch %d, Iter %d, train distil loss source %.4f, target %.4f, mae %.4f, time spent %.4fs" %\
(ep, i, epoch_distil_s[-1], epoch_distil_t[-1], epoch_mae[-1], time.time() - s1))
# print('x', x.shape) (B, len, node, feat)
engine.scheduler.step()
s2 = time.time()
print('Distil epoch %d, train distil loss source %.4f, target %.4f, mae %.4f, time %.4fs' % \
(ep, np.mean(epoch_distil_s), np.mean(epoch_distil_t), np.mean(epoch_mae), s2-s1))
train_time.append(s2-s1)
train_distil_s.append(np.mean(epoch_distil_s))
train_distil_t.append(np.mean(epoch_distil_t))
train_mae.append(np.mean(epoch_mae))
t1 = time.time()
for y, x in val_loader_t:
x = x.to(args.device)
metrics = engine.eval_distil(x)
epoch_valdistil_t.append(metrics[0])
epoch_valmae.append(metrics[1])
epoch_valrmse.append(metrics[2])
epoch_valmape.append(metrics[3])
# print(metrics[3])
t2 = time.time()
val_time.append(t2-t1)
val_mae.append(np.mean(epoch_valmae))
val_rmse.append(np.mean(epoch_valrmse))
val_mape.append(np.mean(epoch_valmape))
val_distil_t.append(np.mean(epoch_valdistil_t))
print("Distil epoch %d, val distil loss %.4f, mae %.4f, rmse %.4f, mape %.4f, time %.4fs" % \
(ep, val_distil_t[-1], val_mae[-1], val_rmse[-1], val_mape[-1], t2-t1))
torch.save(engine.student_model.state_dict(), save_distilformer+'epoch_%d_%.4f.pth' % (ep, val_mae[-1]))
print("Distillation training finish")
print("Average epoch training time is %.4fs, validation time is %.4fs" % (np.mean(train_time), np.mean(val_time)))
np.save(save_distilformer+"val_mae.npy", arr = np.array(val_mae))
np.save(save_distilformer+'val_rmse.npy', arr=np.array(val_rmse))
np.save(save_distilformer+'val_mape.npy', arr=np.array(val_mape))
np.save(save_distilformer+'val_distil_t.npy', arr=np.array(val_distil_t))
np.save(save_distilformer+'train_distil_t.npy', arr = np.array(train_distil_t))
np.save(save_distilformer+'train_loss.npy', arr = np.array(train_mae))
def sample_batch(dataset, batch_size):
# sample a batch of data from a dataset
data_size = len(dataset)
data_idx = np.random.randint(0, data_size, batch_size)
sampled_x = []
for idx in data_idx:
_, x = dataset[idx]
sampled_x.append(x)
return torch.stack(sampled_x, dim=0)
if __name__ == '__main__':
main(args)