DuyTa's picture
Upload folder using huggingface_hub (#1)
44168e6 verified
raw
history blame
2.15 kB
import torch
import os
from monai.utils import set_determinism
from pytorch_lightning.callbacks import ModelCheckpoint, EarlyStopping
import os
from pytorch_lightning.loggers import TensorBoardLogger
from trainer import BRATS
from dataset.utils import get_loader
import pytorch_lightning as pl
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
set_determinism(seed=0)
os.system('cls||clear')
print("Training ...")
data_dir = "/app/brats_2021_task1"
json_list = "/app/info.json"
roi = (128, 128, 128)
batch_size = 1
fold = 1
max_epochs = 500
val_every = 10
train_loader, val_loader,test_loader = get_loader(batch_size, data_dir, json_list, fold, roi, volume=1, test_size=0.2)
print("Done initialize dataloader !! ")
model = BRATS(use_VAE = True, train_loader = train_loader,val_loader = val_loader, test_loader=test_loader )
checkpoint_callback = ModelCheckpoint(
monitor='val/MeanDiceScore',
dirpath='./checkpoints/{}'.format("SegTransVAE"),
filename='Epoch{epoch:3d}-MeanDiceScore{val/MeanDiceScore:.4f}',
save_top_k=3,
mode='max',
save_last= True,
auto_insert_metric_name=False
)
early_stop_callback = EarlyStopping(
monitor='val/MeanDiceScore',
min_delta=0.0001,
patience=15,
verbose=False,
mode='max'
)
tensorboardlogger = TensorBoardLogger(
'logs',
name = "SegTransVAE",
default_hp_metric = None
)
trainer = pl.Trainer(#fast_dev_run = 10,
# accelerator='ddp',
#overfit_batches=5,
devices = [0],
precision=16,
max_epochs = max_epochs,
enable_progress_bar=True,
callbacks=[checkpoint_callback, early_stop_callback],
# auto_lr_find=True,
num_sanity_val_steps=1,
logger = tensorboardlogger,
check_val_every_n_epoch = 10,
# limit_train_batches=0.01,
# limit_val_batches=0.01
)
# trainer.tune(model)
trainer.fit(model)