def main():
models = ['resnet18', 'resnet34', 'resnet101', 'resnet152']
layers = ['layer4', 'layer3', 'layer2', 'layer1']
for model_name in models:
sample_model = experiment('whole', model_name, num_classes=4, pretrained=True)
all_experiments = []
for l in layers:
all_experiments += get_experiments_per_layer(sample_model, l)
all_experiments += ['whole']
for exp in all_experiments:
parser = argparse.ArgumentParser(description='Microscopy Classification CLEF')
parser.add_argument('--batch-size', type=int, default=32, metavar='N', help='input batch size for training (default: 32)')
parser.add_argument('--test-batch-size', type=int, default=32, metavar='N', help='input batch size for testing (default: 1000)')
parser.add_argument('--epochs', type=int, default=1, metavar='N', help='number of epochs to train (default: 10)')
parser.add_argument('--lr', type=float, default=1e-3, metavar='LR', help='learning rate (default: 0.001)')
parser.add_argument('--seed', type=int, default=443, metavar='S', help='random seed (default: 443)')
parser.add_argument('--patience', type=int, default=5, metavar='N', help='how many epochs to wait before stopping a non-improving execution')
parser.add_argument('--num-workers', type=int, default=16, metavar='N', help='processors for data loading tasks')
parser.add_argument('--num-output-classes', type=int, default=4, metavar='N', help='number of classes in the dataset')
parser.add_argument('--csv-path', type=str, default='/workspace/labels/microscopy.csv', help='location of input images')
parser.add_argument('--out-dir', type=str, default='/workspace/outputs', help='location for output data')
parser.add_argument('--project_name', type=str, default='microscopy', help='Project name on Wandb')
parser.add_argument('--architecture', type=str, default='resnet18', help='Highlevel description of the deep learning model used')
parser.add_argument('--experiment', type=str, default='fc', help='Experiment variation')
parser.add_argument('--pretrained', type=int, default='1', help='pretrain on imagenet')
parser.add_argument('--infra', type=str, default='', help='Description of the infrastructure used for training')
parser.add_argument('--notes', type=str, default='', help='Any particular note about the run')
parser.add_argument('--augmentation', type=int, default='1', help='Use data augmentation')
parser.add_argument('--weight-sampler', type=int, default='1', help='Weight sampling for unbalanced set')
parser.add_argument('--weight-loss', type=int, default='1', help='Apply a weighted loss function to fight unbalanced dataset')
args = parser.parse_args()
# update args
args.architecture = model_name
args.experiment = exp
args.lr = 5e-6
args.epochs = 100
args.patience = 100
# all with weight loss and weight sampler and augmentation (see defaults)
provider = MicroscopyTrainDataLoader(args.csv_path, seed=args.seed)
model = experiment(args.experiment, args.architecture, num_classes=args.num_output_classes, pretrained= (args.pretrained == 1))
args.augmentation = (args.augmentation == 1)
args.weight_sampler = (args.weight_sampler == 1)
args.weight_loss = (args.weight_loss == 1)
run = ExperimentalRun(model, provider, args, notes=args.notes, tags=['clef', 'microscopy'])
run.train()
def main():
parser = argparse.ArgumentParser(description='Chart Image Classification')
parser.add_argument('--batch-size',
type=int,
default=32,
metavar='N',
help='input batch size for training (default: 32)')
parser.add_argument('--test-batch-size',
type=int,
default=32,
metavar='N',
help='input batch size for testing (default: 1000)')
parser.add_argument('--epochs',
type=int,
default=10,
metavar='N',
help='number of epochs to train (default: 10)')
parser.add_argument('--lr',
type=float,
default=1e-3,
metavar='LR',
help='learning rate (default: 0.001)')
parser.add_argument('--seed',
type=int,
default=443,
metavar='S',
help='random seed (default: 443)')
#parser.add_argument('--log-interval', type=int, default=10, metavar='N', help='how many batches to wait before logging training status')
parser.add_argument(
'--patience',
type=int,
default=5,
metavar='N',
help='how many epochs to wait before stopping a non-improving execution'
)
parser.add_argument('--num-workers',
type=int,
default=16,
metavar='N',
help='processors for data loading tasks')
parser.add_argument('--num-output-classes',
type=int,
default=10,
metavar='N',
help='number of classes in the dataset')
parser.add_argument('--dataset-path',
type=str,
default='./data',
help='location of input images')
parser.add_argument('--labels-path',
type=str,
default='../labels_task1.csv',
help='location of labels for task')
parser.add_argument('--out-dir',
type=str,
default='./outputs',
help='location for output data')
parser.add_argument('--project_name',
type=str,
default='chart',
help='Project name on Wandb')
parser.add_argument(
'--architecture',
type=str,
default='',
help='Highlevel description of the deep learning model used')
parser.add_argument(
'--infra',
type=str,
default='',
help='Description of the infrastructure used for training')
parser.add_argument('--notes',
type=str,
default='',
help='Any particular note about the run')
args = parser.parse_args()
classes = [
'Donut', 'Grouped horizontal bar', 'Grouped vertical bar',
'Horizontal box', 'Line', 'Pie', 'Scatter', 'Stacked horizontal bar',
'Stacked vertical bar', 'Vertical box'
]
provider = DataProprocessing(args.dataset_path,
None,
args.labels_path,
seed=args.seed)
model = models.resnet50(pretrained=False)
# Reset the last layer
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, args.num_output_classes)
run = ExperimentalRun(model,
provider,
args,
notes=args.notes,
tags=['chart', 'image classification'])
run.train()
def main():
parser = argparse.ArgumentParser(
description='Microscopy Classification CLEF')
parser.add_argument('--batch-size',
type=int,
default=32,
metavar='N',
help='input batch size for training (default: 32)')
parser.add_argument('--test-batch-size',
type=int,
default=32,
metavar='N',
help='input batch size for testing (default: 1000)')
parser.add_argument('--epochs',
type=int,
default=1,
metavar='N',
help='number of epochs to train (default: 10)')
parser.add_argument('--lr',
type=float,
default=1e-3,
metavar='LR',
help='learning rate (default: 0.001)')
parser.add_argument('--seed',
type=int,
default=443,
metavar='S',
help='random seed (default: 443)')
parser.add_argument(
'--patience',
type=int,
default=5,
metavar='N',
help='how many epochs to wait before stopping a non-improving execution'
)
parser.add_argument('--num-workers',
type=int,
default=16,
metavar='N',
help='processors for data loading tasks')
parser.add_argument('--num-output-classes',
type=int,
default=4,
metavar='N',
help='number of classes in the dataset')
parser.add_argument('--csv-path',
type=str,
default='/workspace/labels/microscopy.csv',
help='location of input images')
parser.add_argument('--out-dir',
type=str,
default='/workspace/outputs',
help='location for output data')
parser.add_argument('--project_name',
type=str,
default='microscopy',
help='Project name on Wandb')
parser.add_argument(
'--architecture',
type=str,
default='resnet18',
help='Highlevel description of the deep learning model used')
parser.add_argument('--experiment',
type=str,
default='fc',
help='Experiment variation')
parser.add_argument('--pretrained',
type=int,
default='1',
help='pretrain on imagenet')
parser.add_argument(
'--infra',
type=str,
default='',
help='Description of the infrastructure used for training')
parser.add_argument('--notes',
type=str,
default='',
help='Any particular note about the run')
parser.add_argument('--augmentation',
type=int,
default='0',
help='Use data augmentation')
parser.add_argument('--weight-sampler',
type=int,
default='0',
help='Weight sampling for unbalanced set')
parser.add_argument(
'--weight-loss',
type=int,
default='0',
help='Apply a weighted loss function to fight unbalanced dataset')
args = parser.parse_args()
provider = MicroscopyTrainDataLoader(args.csv_path, seed=args.seed)
args.augmentation = (args.augmentation == 1)
args.weight_sampler = (args.weight_sampler == 1)
args.weight_loss = (args.weight_loss == 1)
JSON_INPUT_PATH = "/workspace/src/experiments/microscopy/shallow-resnet50.json"
with open(JSON_INPUT_PATH) as json_file:
models = json.load(json_file)
resnet50_4_2 = load_shallow_model('resnet50.layer4-2', models)
resnet50_4_1 = load_shallow_model('resnet50.layer4-1', models)
resnet50_4_0 = load_shallow_model('resnet50.layer4-0', models)
resnet50_3_5 = load_shallow_model('resnet50.layer3-5', models)
resnet50_3_4 = load_shallow_model('resnet50.layer3-4', models)
resnet50_3_3 = load_shallow_model('resnet50.layer3-3', models)
ensemble = MyEnsemble(resnet50_4_2, resnet50_4_1, resnet50_4_0,
resnet50_3_5, resnet50_3_4, resnet50_3_3)
for p in ensemble.parameters():
p.requires_grad = False
for p in ensemble.classifier.parameters():
p.requires_grad = True
run = ExperimentalRun(ensemble,
provider,
args,
notes=args.notes,
tags=['clef', 'microscopy', 'ensemble'])
run.train()
def main():
parser = argparse.ArgumentParser(
description='High Level Biomedical Image Classification')
parser.add_argument('--batch-size',
type=int,
default=32,
metavar='N',
help='input batch size for training (default: 32)')
parser.add_argument('--test-batch-size',
type=int,
default=32,
metavar='N',
help='input batch size for testing (default: 1000)')
parser.add_argument('--epochs',
type=int,
default=10,
metavar='N',
help='number of epochs to train (default: 10)')
parser.add_argument('--lr',
type=float,
default=1e-3,
metavar='LR',
help='learning rate (default: 0.001)')
parser.add_argument('--seed',
type=int,
default=443,
metavar='S',
help='random seed (default: 443)')
#parser.add_argument('--log-interval', type=int, default=10, metavar='N', help='how many batches to wait before logging training status')
parser.add_argument(
'--patience',
type=int,
default=5,
metavar='N',
help='how many epochs to wait before stopping a non-improving execution'
)
parser.add_argument('--num-workers',
type=int,
default=16,
metavar='N',
help='processors for data loading tasks')
parser.add_argument('--num-output-classes',
type=int,
default=10,
metavar='N',
help='number of classes in the dataset')
parser.add_argument('--dataset-path',
type=str,
default='./data',
help='location of input images')
parser.add_argument('--labels-path',
type=str,
default='../labels_task1.csv',
help='location of labels for task')
parser.add_argument('--out-dir',
type=str,
default='./outputs',
help='location for output data')
parser.add_argument('--project_name',
type=str,
default='high-level biomedical modality',
help='Project name on Wandb')
parser.add_argument(
'--architecture',
type=str,
default='',
help='Highlevel description of the deep learning model used')
parser.add_argument(
'--infra',
type=str,
default='',
help='Description of the infrastructure used for training')
parser.add_argument('--notes',
type=str,
default='',
help='Any particular note about the run')
args = parser.parse_args()
classes = [
'MICROSCOPY', 'GRAPHICS', 'ORGANISMS', 'MOLECULAR', 'EXPERIMENTAL',
'OTHER'
]
classes.sort()
provider = TrainDataLoader(args.dataset_path,
args.labels_path,
classes,
seed=args.seed)
if args.architecture == 'resnet50':
model = models.resnet50(pretrained=False)
# Reset the last layer
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, args.num_output_classes)
elif args.architecture == 'resnet18':
model = models.resnet18(pretrained=False)
# Reset the last layer
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, args.num_output_classes)
elif args.architecture == 'resnet34':
model = models.resnet34(pretrained=False)
# Reset the last layer
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, args.num_output_classes)
elif args.architecture == 'efficientnet':
# https://github.com/lukemelas/EfficientNet-PyTorch/issues/152
model = EfficientNet.from_pretrained(
'efficientnet-b0', num_classes=args.num_output_classes)
for param in model.parameters():
param.requires_grad = False
model._fc.weight.requires_grad = True
model._fc.bias.requires_grad = True
run = ExperimentalRun(model,
provider,
args,
notes=args.notes,
tags=['clef', 'modality'])
run.train()