def fit_fn(ds_train, ds_val, verbose):
net = LeNet5()
criterion = nn.CrossEntropyLoss()
optimizer = SGD(net.parameters(),
lr=0.01,
momentum=0.9,
weight_decay=1e-4,
nesterov=True)
# lr_scheduler = MultiStepLR(optimizer, [10, 20], gamma=0.1)
lr_scheduler = CosineAnnealingLR(optimizer,
T_max=30,
eta_min=0.001,
warmup=5,
warmup_eta_min=0.01)
metrics = {
'loss': TrainLoss(),
'acc': Accuracy(),
}
test_metrics = {
'loss': Loss(criterion),
'acc': Accuracy(),
}
trainer = Trainer(net,
criterion,
optimizer,
lr_scheduler,
metrics=metrics,
test_metrics=test_metrics,
work_dir="./checkpoints/MNIST-LeNet5")
trainer._verbose = False
# summary(net, (1, 32, 32))
train_loader = DataLoader(ds_train,
batch_size=128,
shuffle=True,
num_workers=2,
pin_memory=True)
val_loader = DataLoader(ds_val, batch_size=128)
accs = trainer.fit(train_loader, 5, val_loader=val_loader)['acc']
return accs[-1], max(accs)
('sep_conv_5x5', 0), ('sep_conv_3x3', 1),
('avg_pool_3x3', 0), ('dil_conv_3x3', 1)],
normal_concat=[2, 3, 4, 5],
reduce=[('sep_conv_5x5', 1), ('max_pool_3x3', 0),
('sep_conv_5x5', 1), ('sep_conv_5x5', 2),
('sep_conv_3x3', 0), ('sep_conv_3x3', 3),
('sep_conv_3x3', 1), ('sep_conv_3x3', 2)],
reduce_concat=[2, 3, 4, 5])
drop_path = 0.3
epochs = 600
# net = NASNet(36, 20, True, drop_path, 10, PC_DARTS_cifar)
net = NASNet(4, 5, True, drop_path, 10, PC_DARTS_cifar)
criterion = CrossEntropyLoss(auxiliary_weight=0.4)
optimizer = SGD(net.parameters(), lr=0.025, momentum=0.9, weight_decay=3e-4)
lr_scheduler = CosineAnnealingLR(optimizer, epochs, min_lr=0)
train_metrics = {
'loss': TrainLoss(),
'acc': Accuracy(),
}
eval_metrics = {
'loss': Loss(CrossEntropyLoss()),
'acc': Accuracy(),
}
trainer = CNNLearner(net,
criterion,
optimizer,
lr_scheduler,
batch_size = 128
steps_per_epoch = math.ceil(len(ds_train) / batch_size)
min_lr = 0.01 * mul
max_lr = 0.1 * mul
net = LeNet5()
criterion = nn.CrossEntropyLoss()
optimizer = SGD(net.parameters(),
lr=min_lr,
momentum=0.9,
weight_decay=1e-4,
nesterov=True)
# lr_scheduler = MultiStepLR(optimizer, [10, 20], gamma=0.1)
lr_scheduler = CosineAnnealingLR(optimizer,
T_max=30,
eta_min=0.001,
warmup=5,
warmup_eta_min=0.001)
metrics = {
'loss': TrainLoss(),
'acc': Accuracy(),
}
test_metrics = {
'loss': Loss(criterion),
'acc': Accuracy(),
}
trainer = Trainer(net,
criterion,
ds_test = CIFAR10(data_home,
train=False,
download=True,
transform=test_transform)
net = LeNet5()
# net = efficientnet_b0(num_classes=10, dropout=0.3, drop_connect=0.2)
criterion = nn.CrossEntropyLoss()
optimizer = SGD(net.parameters(),
lr=0.01,
momentum=0.9,
weight_decay=1e-4,
nesterov=True)
lr_scheduler = CosineAnnealingLR(optimizer,
100,
eta_min=1e-3,
warmup=5,
warmup_eta_min=1e-3)
metrics = {
'loss': TrainLoss(),
'acc': Accuracy(),
}
trainer = Trainer(net,
criterion,
optimizer,
lr_scheduler,
metrics=metrics,
save_path="./checkpoints",
name="CIFAR10-EfficientNet")
mul = 1
batch_size = 128
steps_per_epoch = math.ceil(len(ds_train) / batch_size)
min_lr = 0.01 * mul
net = LeNet5()
criterion = nn.CrossEntropyLoss()
optimizer = SGD(net.parameters(),
lr=min_lr,
momentum=0.9,
weight_decay=1e-4,
nesterov=True)
lr_scheduler = CosineAnnealingLR(optimizer,
epochs=30,
min_lr=0.001,
warmup_epoch=5,
warmup_min_lr=0.001)
train_metrics = {
'loss': TrainLoss(),
'acc': Accuracy(),
}
eval_metrics = {
'loss': Loss(criterion),
'acc': Accuracy(),
}
trainer = CNNLearner(net,
criterion,