class Configs(MNISTConfigs, TrainValidConfigs):
"""
## Configurable Experiment Definition
"""
optimizer: torch.optim.Adam
model: nn.Module
set_seed = SeedConfigs()
device: torch.device = DeviceConfigs()
epochs: int = 10
is_save_models = True
model: nn.Module
inner_iterations = 10
accuracy_func = Accuracy()
loss_func = nn.CrossEntropyLoss()
def init(self):
tracker.set_queue("loss.*", 20, True)
tracker.set_scalar("accuracy.*", True)
hook_model_outputs(self.mode, self.model, 'model')
self.state_modules = [self.accuracy_func]
def step(self, batch: any, batch_idx: BatchIndex):
# Get the batch
data, target = batch[0].to(self.device), batch[1].to(self.device)
# Add global step if we are in training mode
if self.mode.is_train:
tracker.add_global_step(len(data))
# Run the model and specify whether to log the activations
with self.mode.update(is_log_activations=batch_idx.is_last):
output = self.model(data)
# Calculate the loss
loss = self.loss_func(output, target)
# Calculate the accuracy
self.accuracy_func(output, target)
# Log the loss
tracker.add("loss.", loss)
# Optimize if we are in training mode
if self.mode.is_train:
# Calculate the gradients
loss.backward()
# Take optimizer step
self.optimizer.step()
# Log the parameter and gradient L2 norms once per epoch
if batch_idx.is_last:
tracker.add('model', self.model)
tracker.add('optimizer', (self.optimizer, {
'model': self.model
}))
# Clear the gradients
self.optimizer.zero_grad()
# Save logs
tracker.save()
class Configs(MNISTConfigs, TrainValidConfigs):
seed = SeedConfigs()
device: torch.device = DeviceConfigs()
epochs: int = 10
train_batch_size = 1
valid_batch_size = 1
is_save_models = True
model: nn.Module
loss_func = nn.CrossEntropyLoss()
accuracy_func = SimpleAccuracy()
class Configs(MNISTConfigs, TrainValidConfigs):
seed = SeedConfigs()
device: torch.device = DeviceConfigs()
epochs: int = 10
is_save_models = True
model: nn.Module
learning_rate: float = 2.5e-4
momentum: float = 0.5
loss_func = 'cross_entropy_loss'
accuracy_func = 'simple_accuracy'
class Configs(MNISTConfigs, TrainValidConfigs):
optimizer: torch.optim.Adam
model: nn.Module
set_seed = SeedConfigs()
device: torch.device = DeviceConfigs()
epochs: int = 10
is_save_models = True
model: nn.Module
inner_iterations = 10
accuracy_func = Accuracy()
loss_func = nn.CrossEntropyLoss()
def init(self):
tracker.set_queue("loss.*", 20, True)
tracker.set_scalar("accuracy.*", True)
hook_model_outputs(self.mode, self.model, 'model')
self.state_modules = [self.accuracy_func]
def step(self, batch: any, batch_idx: BatchIndex):
data, target = batch[0].to(self.device), batch[1].to(self.device)
if self.mode.is_train:
tracker.add_global_step(len(data))
with self.mode.update(is_log_activations=batch_idx.is_last):
output = self.model(data)
loss = self.loss_func(output, target)
self.accuracy_func(output, target)
tracker.add("loss.", loss)
if self.mode.is_train:
loss.backward()
self.optimizer.step()
if batch_idx.is_last:
tracker.add('model', self.model)
self.optimizer.zero_grad()
tracker.save()