def train_epoch(epoch, model: BaseModel, dataset, device: str,
tracker: BaseTracker, checkpoint: ModelCheckpoint, log):
model.train()
tracker.reset("train")
train_loader = dataset.train_dataloader()
iter_data_time = time.time()
with Ctq(train_loader) as tq_train_loader:
for i, data in enumerate(tq_train_loader):
data = data.to(device) # This takes time
model.set_input(data)
t_data = time.time() - iter_data_time
iter_start_time = time.time()
model.optimize_parameters(dataset.batch_size)
if i % 10 == 0:
tracker.track(model)
tq_train_loader.set_postfix(**tracker.get_metrics(),
data_loading=float(t_data),
iteration=float(time.time() -
iter_start_time),
color=COLORS.TRAIN_COLOR)
iter_data_time = time.time()
metrics = tracker.publish()
checkpoint.save_best_models_under_current_metrics(model, metrics)
log.info("Learning rate = %f" % model.learning_rate)
def main(cfg):
OmegaConf.set_struct(cfg, False)
# Get device
device = torch.device("cuda" if (torch.cuda.is_available() and cfg.cuda) else "cpu")
log.info("DEVICE : {}".format(device))
# Enable CUDNN BACKEND
torch.backends.cudnn.enabled = cfg.enable_cudnn
# Checkpoint
checkpoint = ModelCheckpoint(cfg.checkpoint_dir, cfg.model_name, cfg.weight_name, strict=True)
# Create model and datasets
dataset = instantiate_dataset(checkpoint.data_config)
model = checkpoint.create_model(dataset, weight_name=cfg.weight_name)
log.info(model)
log.info("Model size = %i", sum(param.numel() for param in model.parameters() if param.requires_grad))
# Set dataloaders
dataset.create_dataloaders(
model, cfg.batch_size, cfg.shuffle, cfg.num_workers, cfg.precompute_multi_scale,
)
log.info(dataset)
model.eval()
if cfg.enable_dropout:
model.enable_dropout_in_eval()
model = model.to(device)
tracker: BaseTracker = dataset.get_tracker(model, dataset, False, False)
# Run training / evaluation
run(cfg, model, dataset, device, tracker, checkpoint)
def eval_epoch(
epoch: int,
model: BaseModel,
dataset,
device,
tracker: BaseTracker,
checkpoint: ModelCheckpoint,
visualizer: Visualizer,
early_break: bool,
):
model.eval()
tracker.reset("val")
visualizer.reset(epoch, "val")
loader = dataset.val_dataloader()
with Ctq(loader) as tq_val_loader:
for data in tq_val_loader:
data = data.to(device)
with torch.no_grad():
model.set_input(data)
model.forward()
tracker.track(model)
tq_val_loader.set_postfix(**tracker.get_metrics(),
color=COLORS.VAL_COLOR)
if visualizer.is_active:
visualizer.save_visuals(model.get_current_visuals())
if early_break:
break
metrics = tracker.publish(epoch)
tracker.print_summary()
checkpoint.save_best_models_under_current_metrics(model, metrics)
def main(cfg):
OmegaConf.set_struct(cfg, False)
# Get device
device = torch.device("cuda" if (
torch.cuda.is_available() and cfg.cuda) else "cpu")
log.info("DEVICE : {}".format(device))
# Enable CUDNN BACKEND
torch.backends.cudnn.enabled = cfg.enable_cudnn
# Checkpoint
checkpoint = ModelCheckpoint(cfg.checkpoint_dir,
cfg.model_name,
cfg.weight_name,
strict=True)
# Setup the dataset config
# Generic config
train_dataset_cls = get_dataset_class(checkpoint.data_config)
setattr(checkpoint.data_config, "class", train_dataset_cls.FORWARD_CLASS)
setattr(checkpoint.data_config, "dataroot", cfg.input_path)
# Datset specific configs
if cfg.data:
for key, value in cfg.data.items():
checkpoint.data_config.update(key, value)
# Create dataset and mdoel
dataset = instantiate_dataset(checkpoint.data_config)
model = checkpoint.create_model(dataset, weight_name=cfg.weight_name)
log.info(model)
log.info(
"Model size = %i",
sum(param.numel() for param in model.parameters()
if param.requires_grad))
# Set dataloaders
dataset.create_dataloaders(
model,
cfg.batch_size,
cfg.shuffle,
cfg.num_workers,
False,
)
log.info(dataset)
model.eval()
if cfg.enable_dropout:
model.enable_dropout_in_eval()
model = model.to(device)
# Run training / evaluation
if not os.path.exists(cfg.output_path):
os.makedirs(cfg.output_path)
run(model, dataset, device, cfg.output_path)
def test_best_metric(self):
self.run_path = os.path.join(DIR, "checkpt")
if not os.path.exists(self.run_path):
os.makedirs(self.run_path)
model_checkpoint = ModelCheckpoint(self.run_path,
self.model_name,
"test",
run_config=self.config,
resume=False)
model = MockModel()
optimal_state = model.state.item()
mock_metrics = {
"current_metrics": {
"acc": 12
},
"stage": "test",
"epoch": 10
}
model_checkpoint.save_best_models_under_current_metrics(
model, mock_metrics)
model.state[0] = 2
mock_metrics = {
"current_metrics": {
"acc": 0
},
"stage": "test",
"epoch": 11
}
model_checkpoint.save_best_models_under_current_metrics(
model, mock_metrics)
mock_metrics = {
"current_metrics": {
"acc": 10
},
"stage": "train",
"epoch": 11
}
model_checkpoint.save_best_models_under_current_metrics(
model, mock_metrics)
mock_metrics = {
"current_metrics": {
"acc": 15
},
"stage": "train",
"epoch": 11
}
model_checkpoint.save_best_models_under_current_metrics(
model, mock_metrics)
ckp = torch.load(os.path.join(self.run_path, self.model_name + ".pt"))
self.assertEqual(ckp["models"]["best_acc"]["state"].item(),
optimal_state)
self.assertEqual(ckp["models"]["latest"]["state"].item(),
model.state.item())
def test_model_ckpt_using_pointnet2ms(self,):
# Create a checkpt
name = "model"
self.run_path = os.path.join(DIR, "checkpt")
print(self.run_path)
if not os.path.exists(self.run_path):
os.makedirs(self.run_path)
model_checkpoint = ModelCheckpoint(self.run_path, name, "test", run_config=self.config, resume=False)
dataset = MockDatasetGeometric(5)
model = instantiate_model(self.config, dataset)
model.set_input(dataset[0], "cpu")
model.instantiate_optimizers(self.config)
mock_metrics = {"current_metrics": {"acc": 12}, "stage": "test", "epoch": 10}
model_checkpoint.save_best_models_under_current_metrics(model, mock_metrics)
# Load checkpoint and initialize model
model_checkpoint = ModelCheckpoint(self.run_path, name, "test", self.config, resume=True)
model2 = model_checkpoint.create_model(dataset, weight_name="acc")
self.assertEqual(str(model.optimizer.__class__.__name__), str(model2.optimizer.__class__.__name__))
self.assertEqual(model.optimizer.defaults, model2.optimizer.defaults)
self.assertEqual(model.schedulers["lr_scheduler"].state_dict(), model2.schedulers["lr_scheduler"].state_dict())
self.assertEqual(model.schedulers["bn_scheduler"].state_dict(), model2.schedulers["bn_scheduler"].state_dict())
shutil.rmtree(self.run_path)
remove(os.path.join(ROOT, "{}.pt".format(name)))
remove(os.path.join(DIR, "{}.pt".format(name)))
def test_epoch(model: BaseModel, dataset, device, tracker: BaseTracker,
checkpoint: ModelCheckpoint, log):
model.eval()
tracker.reset("test")
loader = dataset.test_dataloader()
with Ctq(loader) as tq_test_loader:
for data in tq_test_loader:
data = data.to(device)
with torch.no_grad():
model.set_input(data)
model.forward()
tracker.track(model)
tq_test_loader.set_postfix(**tracker.get_metrics(),
color=COLORS.TEST_COLOR)
metrics = tracker.publish()
tracker.print_summary()
checkpoint.save_best_models_under_current_metrics(model, metrics)
def train_epoch(
epoch: int,
model: BaseModel,
dataset,
device: str,
tracker: BaseTracker,
checkpoint: ModelCheckpoint,
visualizer: Visualizer,
early_break: bool,
):
model.train()
tracker.reset("train")
visualizer.reset(epoch, "train")
train_loader = dataset.train_dataloader
iter_data_time = time.time()
with Ctq(train_loader) as tq_train_loader:
for i, data in enumerate(tq_train_loader):
model.set_input(data, device)
t_data = time.time() - iter_data_time
iter_start_time = time.time()
model.optimize_parameters(epoch, dataset.batch_size)
if i % 10 == 0:
tracker.track(model)
tq_train_loader.set_postfix(**tracker.get_metrics(),
data_loading=float(t_data),
iteration=float(time.time() -
iter_start_time),
color=COLORS.TRAIN_COLOR)
if visualizer.is_active:
visualizer.save_visuals(model.get_current_visuals())
iter_data_time = time.time()
if early_break:
break
metrics = tracker.publish(epoch)
checkpoint.save_best_models_under_current_metrics(model, metrics)
log.info("Learning rate = %f" % model.learning_rate)
def main(cfg):
OmegaConf.set_struct(cfg, False)
# Get device
device = torch.device("cuda" if (
torch.cuda.is_available() and cfg.cuda) else "cpu")
log.info("DEVICE : {}".format(device))
# Enable CUDNN BACKEND
torch.backends.cudnn.enabled = cfg.enable_cudnn
# Checkpoint
checkpoint = ModelCheckpoint(cfg.checkpoint_dir,
cfg.model_name,
cfg.weight_name,
strict=True)
# Setup the dataset config
# Generic config
dataset = instantiate_dataset(cfg.data)
model = checkpoint.create_model(dataset, weight_name=cfg.weight_name)
log.info(model)
log.info(
"Model size = %i",
sum(param.numel() for param in model.parameters()
if param.requires_grad))
log.info(dataset)
model.eval()
if cfg.enable_dropout:
model.enable_dropout_in_eval()
model = model.to(device)
# Run training / evaluation
output_path = os.path.join(cfg.checkpoint_dir, cfg.data.name, "features")
if not os.path.exists(output_path):
os.makedirs(output_path, exist_ok=True)
run(model, dataset, device, output_path, cfg)
def test_epoch(
epoch: int,
model: BaseModel,
dataset,
device,
tracker: BaseTracker,
checkpoint: ModelCheckpoint,
visualizer: Visualizer,
early_break: bool,
):
model.eval()
loaders = dataset.test_dataloaders()
for idx, loader in enumerate(loaders):
stage_name = dataset.get_test_dataset_name(idx)
tracker.reset(stage_name)
visualizer.reset(epoch, stage_name)
with Ctq(loader) as tq_test_loader:
for data in tq_test_loader:
data = data.to(device)
with torch.no_grad():
model.set_input(data)
model.forward()
tracker.track(model)
tq_test_loader.set_postfix(**tracker.get_metrics(),
color=COLORS.TEST_COLOR)
if visualizer.is_active:
visualizer.save_visuals(model.get_current_visuals())
if early_break:
break
metrics = tracker.publish(epoch)
tracker.print_summary()
checkpoint.save_best_models_under_current_metrics(model, metrics)
def main(cfg):
OmegaConf.set_struct(
cfg,
False) # This allows getattr and hasattr methods to function correctly
if cfg.pretty_print:
print(cfg.pretty())
# Get device
device = torch.device("cuda" if (
torch.cuda.is_available() and cfg.training.cuda) else "cpu")
log.info("DEVICE : {}".format(device))
# Enable CUDNN BACKEND
torch.backends.cudnn.enabled = cfg.training.enable_cudnn
# Start Wandb if public
launch_wandb(cfg, cfg.wandb.public and cfg.wandb.log)
# Checkpoint
checkpoint = ModelCheckpoint(
cfg.training.checkpoint_dir,
cfg.model_name,
cfg.training.weight_name,
run_config=cfg,
resume=bool(cfg.training.checkpoint_dir),
)
# Create model and datasets
if not checkpoint.is_empty:
dataset = instantiate_dataset(checkpoint.data_config)
model = checkpoint.create_model(dataset,
weight_name=cfg.training.weight_name)
else:
dataset = instantiate_dataset(cfg.data)
model = instantiate_model(cfg, dataset)
model.instantiate_optimizers(cfg)
log.info(model)
model.log_optimizers()
log.info(
"Model size = %i",
sum(param.numel() for param in model.parameters()
if param.requires_grad))
# Set dataloaders
dataset.create_dataloaders(
model,
cfg.training.batch_size,
cfg.training.shuffle,
cfg.training.num_workers,
cfg.training.precompute_multi_scale,
)
log.info(dataset)
# Choose selection stage
selection_stage = getattr(cfg, "selection_stage", "")
checkpoint.selection_stage = dataset.resolve_saving_stage(selection_stage)
tracker: BaseTracker = dataset.get_tracker(model, dataset, cfg.wandb.log,
cfg.tensorboard.log)
launch_wandb(cfg, not cfg.wandb.public and cfg.wandb.log)
# Run training / evaluation
model = model.to(device)
visualizer = Visualizer(cfg.visualization, dataset.num_batches,
dataset.batch_size, os.getcwd())
run(cfg, model, dataset, device, tracker, checkpoint, visualizer)