-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
94 lines (74 loc) · 2.94 KB
/
train.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
import argparse
import sys
import numpy as np
import matplotlib.pyplot as plt
import torch; torch.manual_seed(0)
import torch.nn as nn
import torchvision.transforms as transforms
import asdstm.training as training
import asdstm.preprocessing as preprocessing
from asdstm.models import AttentionUNet2d
def preprocess_fn(batch, testing=False):
X, Y, xyz = batch
transform = transforms.Compose([
preprocessing.Cutout(),
preprocessing.AddNoise(c=0.1),
preprocessing.Normalize()
])
X = transform(X)
X = X.unsqueeze(1)
transform_Y = transforms.Compose([
preprocessing.MinimumToZero()
])
Y = transform_Y(Y[:, 0]) # 0: Atomic Disks, 1: vdW spheres, 2: HeightMap
Y = Y.unsqueeze(1)
if testing:
out = X, Y, xyz
else:
out = X, Y
return out
if __name__=='__main__':
device = 'cuda'
parser = argparse.ArgumentParser()
parser.add_argument('--fname', type=str, default='data/data.hdf5', help="Data location")
parser.add_argument('--savepath', type=str, default='./', help='Where to store model+predictions')
parser.add_argument('--ncpu', type=int, default=8, help="Number of workers")
parser.add_argument('--epoch', type=int, help="Number of epochs to train")
parser.add_argument('--pretrained', action=argparse.BooleanOptionalAction, help="Load pretrained model")
args = parser.parse_args()
train_set = training.SPMDataset(args.fname, mode='train', scan='stm', height='random')
val_set = training.SPMDataset(args.fname, mode='val', scan='stm', height='random')
test_set = training.SPMDataset(args.fname, mode='test', scan='stm', height='random')
dataloader_args={'batch_size': 30,
'num_workers': args.ncpu,
'pin_memory': True,
'shuffle': True}
model = AttentionUNet2d(act=nn.ReLU())
if args.pretrained:
model_state = torch.load('./pretrained/disks.pt')
model.load_state_dict(model_state)
model = model.to(device)
optimizer = torch.optim.Adam(model.parameters())
criterion = nn.MSELoss()
n_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
print('Number of trainable parameters: {}'.format(n_params))
print(f'Training data set size: {len(train_set)}')
print(f'Validation data set size: {len(val_set)}')
print(f'Test data set size: {len(test_set)}')
trainer = training.Trainer(model=model,
optimizer=optimizer,
criterion=criterion,
train_set=train_set,
validation_set=val_set,
test_set=test_set,
preprocess_fn=preprocess_fn,
model_path=args.savepath,
dataloader_args=dataloader_args,
print_interval=100,
device=device,
timing=False)
if not args.pretrained:
trainer.train(epochs=args.epoch)
trainer.save_final_model()
trainer.get_test_loss()
trainer.make_predictions(batches=3)