def main():
config = load_config(config_file_name="lgb_titanic_config.ini")
data = load_data(data_file_name="titanic_training.csv")
set_randomseed()
transformer = TransformerCsv(data_df=data)
lgb_model = LgbClassifier(model_params_path=config.get('model_params', 'lgb_params_path'),
target_col=config.get('data_columns', 'target_col'))
train_data, test_data = titanic_data_transform(transformer)
lgb_trainer = Trainer()
lgb_model = lgb_trainer.fit(model_class=lgb_model, train_data_df=train_data)
eval_dict = lgb_model.eval(test_data=test_data)
print("test acc:", eval_dict["acc"])
print("conf_matrix", eval_dict["conf_matrix"])
result_ploter = PlotClassificationResult()
result_ploter.plot_roc_curve(roc_tapple=eval_dict["roc_curve"], plot_show=False)
def main(args):
# region -----------------------------------------记录训练日志-----------------------------------------
if 0 == len(args.resume):
logger = Logger('./logs/' + args.model + '.log')
else:
logger = Logger('./logs/' + args.model + '.log', True)
logger.append(vars(args))
# if args.display:
# writer = SummaryWriter()
# else:
# writer = None
writer = SummaryWriter()
# endregion
# region ----------------------------------------数据预处理配置----------------------------------------
data_transforms = {
'train':
transforms.Compose([
transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
'val':
transforms.Compose([
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
}
# endregion
# region -------------------------------------------数据加载-------------------------------------------
train_datasets = datasets.CIFAR10(root="data",
train=True,
download=True,
transform=transforms.ToTensor())
val_datasets = datasets.CIFAR10(root="data",
train=False,
download=True,
transform=transforms.ToTensor())
# train_datasets = datasets.ImageFolder(os.path.join(args.data_root, 't256'), data_transforms['train'])
# val_datasets = datasets.ImageFolder(os.path.join(args.data_root, 'v256'), data_transforms['val'])
train_dataloaders = torch.utils.data.DataLoader(train_datasets,
batch_size=args.batch_size,
shuffle=True,
num_workers=4)
val_dataloaders = torch.utils.data.DataLoader(val_datasets,
batch_size=1024,
shuffle=False,
num_workers=4)
# endregion
# region --------------------------------------网络无关配置设置----------------------------------------
# 记录日志
if args.debug:
x, y = next(iter(train_dataloaders))
logger.append([x, y])
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
is_use_cuda = torch.cuda.is_available()
cudnn.benchmark = True
# 设置基础网络
if 'resnet50' == args.model.split('_')[0]:
my_model = models.resnet50(pretrained=False)
elif 'resnet50-cbam' == args.model.split('_')[0]:
my_model = resnet_cbam.resnet50_cbam(pretrained=False)
elif 'resnet101' == args.model.split('_')[0]:
my_model = models.resnet101(pretrained=False)
else:
raise ModuleNotFoundError
# endregion
# region --------------------------------------网络训练配置设置----------------------------------------
# 损失函数设定
loss_fn = [nn.CrossEntropyLoss()]
# 优化器设置
optimizer = optim.SGD(my_model.parameters(),
lr=0.1,
momentum=0.9,
weight_decay=1e-4)
# 学习率优化函数设置
lr_schedule = lr_scheduler.MultiStepLR(optimizer,
milestones=[30, 60],
gamma=0.1)
# 累计误差TOP5
metric = [ClassErrorMeter([1, 5], True)]
# 迭代次数
epoch = int(args.epoch)
# 传入训练器
my_trainer = Trainer(my_model, args.model, loss_fn, optimizer, lr_schedule,
500, is_use_cuda, train_dataloaders, val_dataloaders,
metric, 0, epoch, args.debug, logger, writer)
# 训练
my_trainer.fit()
logger.append('训练完毕')
def main(args):
if 0 == len(args.resume):
logger = Logger('./logs/' + args.model + '.log')
else:
logger = Logger('./logs/' + args.model + '.log', True)
logger.append(vars(args))
if args.display:
writer = SummaryWriter()
else:
writer = None
gpus = args.gpu.split(',')
data_transforms = {
'train':
transforms.Compose([
transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
'val':
transforms.Compose([
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
}
train_datasets = datasets.ImageFolder(os.path.join(args.data_root, 't256'),
data_transforms['train'])
val_datasets = datasets.ImageFolder(os.path.join(args.data_root, 'v256'),
data_transforms['val'])
train_dataloaders = torch.utils.data.DataLoader(
train_datasets,
batch_size=args.batch_size * len(gpus),
shuffle=True,
num_workers=8)
val_dataloaders = torch.utils.data.DataLoader(val_datasets,
batch_size=1024,
shuffle=False,
num_workers=8)
if args.debug:
x, y = next(iter(train_dataloaders))
logger.append([x, y])
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
is_use_cuda = torch.cuda.is_available()
cudnn.benchmark = True
if 'resnet50' == args.model.split('_')[0]:
my_model = models.resnet50(pretrained=False)
elif 'resnet101' == args.model.split('_')[0]:
my_model = models.resnet101(pretrained=False)
elif 'mnasnet' == args.model.split('_')[0]:
my_model = mnasnet.MnasNet()
elif 'shufflenetv2' == args.model.split('_')[0]:
my_model = shufflenetv2.ShuffleNetV2()
else:
raise ModuleNotFoundError
#my_model.apply(fc_init)
if is_use_cuda and 1 == len(gpus):
my_model = my_model.cuda()
elif is_use_cuda and 1 < len(gpus):
my_model = nn.DataParallel(my_model.cuda())
loss_fn = [nn.CrossEntropyLoss()]
optimizer = optim.SGD(my_model.parameters(),
lr=0.1,
momentum=0.9,
weight_decay=1e-4)
lr_schedule = lr_scheduler.MultiStepLR(optimizer,
milestones=[30, 60],
gamma=0.1) #
metric = [ClassErrorMeter([1, 5], True)]
start_epoch = 0
num_epochs = 90
my_trainer = Trainer(my_model, args.model, loss_fn, optimizer, lr_schedule, 500, is_use_cuda, train_dataloaders, \
val_dataloaders, metric, start_epoch, num_epochs, args.debug, logger, writer)
my_trainer.fit()
logger.append('Optimize Done!')
def main(cfg):
"""Runs main training procedure."""
print('Starting training...')
print('Current working directory is:', os.getcwd())
# fix random seeds for reproducibility
seed_everything(seed=cfg['seed'])
# neptune logging
neptune.init(project_qualified_name=cfg['neptune_project_name'],
api_token=cfg['neptune_api_token'])
neptune.create_experiment(name=cfg['neptune_experiment'], params=cfg)
num_classes = 1 if len(cfg['classes']) == 1 else (len(cfg['classes']) + 1)
activation = 'sigmoid' if num_classes == 1 else 'softmax2d'
background = False if cfg['ignore_channels'] else True
aux_params = dict(
pooling=cfg['pooling'], # one of 'avg', 'max'
dropout=cfg['dropout'], # dropout ratio, default is None
activation='sigmoid', # activation function, default is None
classes=num_classes) # define number of output labels
# configure model
models = {
'unet':
Unet(encoder_name=cfg['encoder_name'],
encoder_weights=cfg['encoder_weights'],
decoder_use_batchnorm=cfg['use_batchnorm'],
classes=num_classes,
activation=activation,
aux_params=aux_params),
'unetplusplus':
UnetPlusPlus(encoder_name=cfg['encoder_name'],
encoder_weights=cfg['encoder_weights'],
decoder_use_batchnorm=cfg['use_batchnorm'],
classes=num_classes,
activation=activation,
aux_params=aux_params),
'deeplabv3plus':
DeepLabV3Plus(encoder_name=cfg['encoder_name'],
encoder_weights=cfg['encoder_weights'],
classes=num_classes,
activation=activation,
aux_params=aux_params)
}
assert cfg['architecture'] in models.keys()
model = models[cfg['architecture']]
# configure loss
losses = {
'dice_loss':
DiceLoss(include_background=background,
softmax=False,
batch=cfg['combine']),
'generalized_dice':
GeneralizedDiceLoss(include_background=background,
softmax=False,
batch=cfg['combine'])
}
assert cfg['loss'] in losses.keys()
loss = losses[cfg['loss']]
# configure optimizer
optimizers = {
'adam': Adam([dict(params=model.parameters(), lr=cfg['lr'])]),
'adamw': AdamW([dict(params=model.parameters(), lr=cfg['lr'])]),
'rmsprop': RMSprop([dict(params=model.parameters(), lr=cfg['lr'])])
}
assert cfg['optimizer'] in optimizers.keys()
optimizer = optimizers[cfg['optimizer']]
# configure metrics
metrics = {
'dice_score':
DiceMetric(include_background=background, reduction='mean'),
'dice_smp':
Fscore(threshold=cfg['rounding'],
ignore_channels=cfg['ignore_channels']),
'iou_smp':
IoU(threshold=cfg['rounding'], ignore_channels=cfg['ignore_channels']),
'generalized_dice':
GeneralizedDiceLoss(include_background=background,
softmax=False,
batch=cfg['combine']),
'dice_loss':
DiceLoss(include_background=background,
softmax=False,
batch=cfg['combine']),
'cross_entropy':
BCELoss(reduction='mean'),
'accuracy':
Accuracy(ignore_channels=cfg['ignore_channels'])
}
assert all(m['name'] in metrics.keys() for m in cfg['metrics'])
metrics = [(metrics[m['name']], m['name'], m['type'])
for m in cfg['metrics']] # tuple of (metric, name, type)
# configure scheduler
schedulers = {
'steplr':
StepLR(optimizer, step_size=cfg['step_size'], gamma=0.5),
'cosine':
CosineAnnealingLR(optimizer,
cfg['epochs'],
eta_min=cfg['eta_min'],
last_epoch=-1)
}
assert cfg['scheduler'] in schedulers.keys()
scheduler = schedulers[cfg['scheduler']]
# configure augmentations
train_transform = load_train_transform(transform_type=cfg['transform'],
patch_size=cfg['patch_size'])
valid_transform = load_valid_transform(patch_size=cfg['patch_size'])
train_dataset = SegmentationDataset(df_path=cfg['train_data'],
transform=train_transform,
normalize=cfg['normalize'],
tissuemix=cfg['tissuemix'],
probability=cfg['probability'],
blending=cfg['blending'],
warping=cfg['warping'],
color=cfg['color'])
valid_dataset = SegmentationDataset(df_path=cfg['valid_data'],
transform=valid_transform,
normalize=cfg['normalize'])
# save intermediate augmentations
if cfg['eval_dir']:
default_dataset = SegmentationDataset(df_path=cfg['train_data'],
transform=None,
normalize=None)
transform_dataset = SegmentationDataset(df_path=cfg['train_data'],
transform=None,
normalize=None,
tissuemix=cfg['tissuemix'],
probability=cfg['probability'],
blending=cfg['blending'],
warping=cfg['warping'],
color=cfg['color'])
for idx in range(0, min(500, len(default_dataset)), 10):
image_input, image_mask = default_dataset[idx]
image_input = image_input.transpose((1, 2, 0))
image_input = image_input.astype(np.uint8)
image_mask = image_mask.transpose(
1, 2, 0) # Why do we need transpose here?
image_mask = image_mask.astype(np.uint8)
image_mask = image_mask.squeeze()
image_mask = image_mask * 255
image_transform, _ = transform_dataset[idx]
image_transform = image_transform.transpose(
(1, 2, 0)).astype(np.uint8)
idx_str = str(idx).zfill(3)
skimage.io.imsave(os.path.join(cfg['eval_dir'],
f'{idx_str}a_image_input.png'),
image_input,
check_contrast=False)
plt.imsave(os.path.join(cfg['eval_dir'],
f'{idx_str}b_image_mask.png'),
image_mask,
vmin=0,
vmax=1)
skimage.io.imsave(os.path.join(cfg['eval_dir'],
f'{idx_str}c_image_transform.png'),
image_transform,
check_contrast=False)
del transform_dataset
train_loader = DataLoader(train_dataset,
batch_size=cfg['batch_size'],
num_workers=cfg['workers'],
shuffle=True)
valid_loader = DataLoader(valid_dataset,
batch_size=cfg['batch_size'],
num_workers=cfg['workers'],
shuffle=False)
trainer = Trainer(model=model,
device=cfg['device'],
save_checkpoints=cfg['save_checkpoints'],
checkpoint_dir=cfg['checkpoint_dir'],
checkpoint_name=cfg['checkpoint_name'])
trainer.compile(optimizer=optimizer,
loss=loss,
metrics=metrics,
num_classes=num_classes)
trainer.fit(train_loader,
valid_loader,
epochs=cfg['epochs'],
scheduler=scheduler,
verbose=cfg['verbose'],
loss_weight=cfg['loss_weight'])
# validation inference
best_model = model
best_model.load_state_dict(
torch.load(os.path.join(cfg['checkpoint_dir'],
cfg['checkpoint_name'])))
best_model.to(cfg['device'])
best_model.eval()
# setup directory to save plots
if os.path.isdir(cfg['plot_dir']):
# remove existing dir and content
shutil.rmtree(cfg['plot_dir'])
# create absolute destination
os.makedirs(cfg['plot_dir'])
# valid dataset without transformations and normalization for image visualization
valid_dataset_vis = SegmentationDataset(df_path=cfg['valid_data'],
transform=valid_transform,
normalize=None)
if cfg['save_checkpoints']:
for n in range(len(valid_dataset)):
image_vis = valid_dataset_vis[n][0].astype('uint8')
image_vis = image_vis.transpose((1, 2, 0))
image, gt_mask = valid_dataset[n]
gt_mask = gt_mask.transpose((1, 2, 0))
gt_mask = gt_mask.squeeze()
x_tensor = torch.from_numpy(image).to(cfg['device']).unsqueeze(0)
pr_mask, _ = best_model.predict(x_tensor)
pr_mask = pr_mask.cpu().numpy().round()
pr_mask = pr_mask.squeeze()
save_predictions(out_path=cfg['plot_dir'],
index=n,
image=image_vis,
ground_truth_mask=gt_mask,
predicted_mask=pr_mask,
average='macro')
def main():
global args
if args.seed is not None:
random.seed(args.seed)
torch.manual_seed(args.seed)
if args.distributed:
init_distributed_mode(args)
else:
args.local_rank = 0
if args.local_rank == 0 and not os.path.exists(args.model_dir):
os.makedirs(args.model_dir)
model, criterion, optimizer, lr_scheduler = initialize_setting()
train_dataset, val_dataset, test_dataset, num_classes, in_shape = \
data_loader.__dict__[args.dataset](
args.data, augmentation=True, cutout=args.cutout)
num_train = len(train_dataset)
indices = list(range(num_train))
split = int(np.floor(args.train_portion * num_train))
train_sampler = DistributedSampler(train_dataset) \
if args.distributed else torch.utils.data.sampler.SubsetRandomSampler(indices[:split])
train_loader = DataLoader(train_dataset,
batch_size=args.batch_size,
shuffle=(train_sampler is None),
num_workers=args.workers,
pin_memory=True,
sampler=train_sampler)
if args.train_portion < 1:
valid_sampler = torch.utils.data.sampler.SubsetRandomSampler(
indices[split:num_train])
val_loader = DataLoader(val_dataset,
batch_size=args.batch_size,
shuffle=False,
sampler=valid_sampler,
num_workers=args.workers,
pin_memory=True)
else:
val_loader = None
test_loader = DataLoader(test_dataset,
batch_size=args.batch_size,
shuffle=False,
num_workers=args.workers,
pin_memory=True)
cudnn.benchmark = True
writer = SummaryWriter(args.model_dir) if args.local_rank == 0 else None
best_acc1 = 0.0
trainer = Trainer(model, criterion, optimizer, train_sampler, lr_scheduler,
train_loader, val_loader, test_loader, writer, best_acc1,
args)
acc1, best_acc1 = trainer.fit(init=True)
if args.local_rank == 0:
save_result(model, in_shape, acc1, best_acc1, args)
writer.close()