def test_basic(self):
model = PseudoModel()
dataset = TrainingPseudoDataSet()
dataset.load_records()
train_dl, val_dl = dataset.partition_data([0.75, 0.25],
TrainingPseudoDataSet)
# model = SimpleNet()
# dataset = TrainingAbnormalDataSet()
# dataset.load_records(keep_annotations=True)
# train_dl, val_dl = dataset.partition_data([0.75, 0.25], TrainingAbnormalDataSet)
batch_aug = BatchAugmenter()
# batch_aug.compose([
# MixUpImageWithAnnotations(probability=1.0)
# ])
task = AbnormalClassificationTask(model,
train_dl,
SGD(model.parameters(),
lr=0.03,
momentum=0.9),
batch_augmenter=batch_aug)
# task = AbnormalClassificationTask(model, train_dl, Adam(model.parameters(), lr=0.03, betas=(0.9, 0.999), weight_decay=0.01), batch_augmenter=batch_aug)
task.max_iter = 100_000_000
# task = TrainingTask()
val_hook = PeriodicStepFuncHook(400000,
lambda: task.validation(val_dl, model))
checkpoint_hook = CheckpointHook(100000, "test", 1000000, 5)
scheduler = LRScheduler.LinearWarmup(0, 3000)
scheduler2 = LRScheduler.LambdaLR(0, 3000, lambda step: 1.0)
task.register_hook(LogTrainingLoss(frequency=100))
task.register_hook(StepTimer())
# task.register_hook(val_hook)
task.register_hook(checkpoint_hook)
# task.register_hook(TrainingVisualizationHook(batch=False))
task.register_lrschedulers(scheduler2)
task.register_lrschedulers(scheduler)
task.begin_or_resume()
assert 1 == 1
if __name__ == "__main__":
from src.data.abnormal_dataset import TrainingAbnormalDataSet
from src.training_tasks.tasks.AbnormalClassificationTask import AbnormalClassificationTask
from src.utils.hooks import StepTimer, PeriodicStepFuncHook, LogTrainingLoss
from torch import optim
from src.training_tasks import BackpropAggregators
model = Res50()
dataloader = TrainingAbnormalDataSet()
dataloader.load_records(keep_annotations=False)
train_dl, val_dl = dataloader.partition_data([0.75, 0.25], TrainingAbnormalDataSet)
task = AbnormalClassificationTask(model, train_dl, optim.Adam(model.parameters(), lr=0.0001), backward_agg=BackpropAggregators.MeanLosses)
task.max_iter = 25000
val_hook = PeriodicStepFuncHook(5000, lambda: task.validation(val_dl, model))
checkpoint_hook = CheckpointHook(1000, "resnet50_test3")
task.register_hook(LogTrainingLoss())
task.register_hook(StepTimer())
task.register_hook(val_hook)
task.register_hook(checkpoint_hook)
task.begin_or_resume()
train_dl, val_dl = dataloader.partition_data([0.75, 0.25], TrainingMulticlassDataset)
steps_per_epoch = len(train_dl) // config.artificial_batch_size
batch_aug = BatchAugmenter()
# batch_aug.compose([MixUpImageWithAnnotations(probability=0.5)])
task = MulticlassDetectionTask(model, train_dl, optim.Adam(model.parameters(), lr=0.00001), backward_agg=BackpropAggregators.MeanLosses, batch_augmenter=batch_aug)
# Loss exploded with this optimizer
# task = MulticlassDetectionTask(model, train_dl, optim.SGD(model.parameters(), lr=0.003, momentum=0.9), backward_agg=BackpropAggregators.MeanLosses, batch_augmenter=batch_aug)
task.max_iter = steps_per_epoch * 2500
validation_iteration = 500
train_acc_hook = PeriodicStepFuncHook(validation_iteration , lambda: task.validation(train_dl, model))
val_hook = PeriodicStepFuncHook(validation_iteration, lambda: task.validation(val_dl, model))
checkpoint_hook = CheckpointHook(validation_iteration, "TimmModel_PostmAPBug_Test1", permanent_checkpoints=validation_iteration, keep_last_n_checkpoints=0)
lr_steps = [1.0, 0.1, 0.01, 0.001, 0.0001]
steps = [ steps_per_epoch * 50, steps_per_epoch * 100, steps_per_epoch * 150]
def lr_stepper(current_step):
idx = 0
for step in steps:
if current_step <= step:
return lr_steps[idx]
idx += 1
return lr_steps[-1]
batch_aug = BatchAugmenter()
batch_aug.compose([MixUpImage(probability=0.75)])
task = AbnormalClassificationTask(
model,
train_dl,
optim.Adam(model.parameters(), lr=0.0001),
backward_agg=BackpropAggregators.MeanLosses,
batch_augmenter=batch_aug)
# task = AbnormalClassificationTask(model, train_dl, optim.SGD(model.parameters(), lr=0.003, momentum=0.9),
# backward_agg=BackpropAggregators.MeanLosses, batch_augmenter=batch_aug)
task.max_iter = steps_per_epoch * 25
val_hook = PeriodicStepFuncHook(
steps_per_epoch, lambda: task.tim__validation(val_dl, model))
train_acc_hook = PeriodicStepFuncHook(
steps_per_epoch, lambda: task.tim__validation(train_dl, model))
checkpoint_hook = CheckpointHook(steps_per_epoch,
"TimmModel_BiTRes_X1_TestFive",
permanent_checkpoints=steps_per_epoch,
keep_last_n_checkpoints=5)
lr_steps = [1.0, 0.1, 0.01, 0.001]
# steps = [ steps_per_epoch * 10, steps_per_epoch * 15, steps_per_epoch * 20]
steps = [-1, -1, steps_per_epoch * 3]
def lr_stepper(current_step):
idx = 0
for step in steps: