-
Notifications
You must be signed in to change notification settings - Fork 17
/
train.py
51 lines (44 loc) · 1.58 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
from multiply_model import MultiplyModel
from lib.datasets import create_dataset
import hydra
import pytorch_lightning as pl
from pytorch_lightning.loggers import WandbLogger
import os
import glob
from omegaconf import OmegaConf
@hydra.main(config_path="confs", config_name="taichi01_base")
def main(opt):
pl.seed_everything(42)
print("Working dir:", os.getcwd())
print(OmegaConf.to_yaml(opt))
checkpoint_callback = pl.callbacks.ModelCheckpoint(
dirpath="checkpoints/",
filename="{epoch:04d}-{loss}",
save_on_train_epoch_end=True,
every_n_epochs=100,
save_top_k=-1,
save_last=True)
logger = WandbLogger(project=opt.project_name, name=f"{opt.exp}/{opt.run}")
trainer = pl.Trainer(
# gpus=1,
devices=1,
accelerator="gpu",
callbacks=[checkpoint_callback],
max_epochs=10000,
check_val_every_n_epoch=50,
logger=logger,
log_every_n_steps=1,
num_sanity_val_steps=0
)
betas_path = os.path.join(hydra.utils.to_absolute_path('..'), 'data', opt.dataset.train.data_dir, 'mean_shape.npy')
model = MultiplyModel(opt, betas_path)
trainset = create_dataset(opt.dataset.train)
validset = create_dataset(opt.dataset.valid)
if opt.model.is_continue == True:
# checkpoint = sorted(glob.glob("checkpoints/*.ckpt"))[-1]
checkpoint = sorted(glob.glob("checkpoints/epoch=*.ckpt"))[-1]
trainer.fit(model, trainset, validset, ckpt_path=checkpoint)
else:
trainer.fit(model, trainset, validset)
if __name__ == '__main__':
main()