def test_model_updates(model, dataset):
before = list(model.parameters())[0].clone().detach().cpu().numpy()
cnn.train_model(model, dataset, epochs=1)
after = list(model.parameters())[0].clone().detach().cpu().numpy()
for b, a in zip(before, after):
# Make sure something changed.
assert (b != a).any()
def train(net, train_loader, dtype, device, parameters):
"""
Simplified training function for ax optimization.
Args:
net: Neural network object.
train_loader: Training data loader.
parameters (dict): Dictionary of parameter values.
dtype: Data type.
device: Device specifying cpu or gpu training.
"""
cnn.train_model(model=net,
dataset=train_loader,
dtype=dtype,
device=device,
optim_args={"lr": parameters["lr"]},
**parameters)
return
def train(net, train_loader, dtype, device, parameters):
"""
Simplified training function for ax optimization.
Args:
net: Neural network object.
train_loader: Training data loader.
parameters (dict): Dictionary of parameter values.
dtype: Data type.
device: Device specifying cpu or gpu training.
"""
cnn.train_model(
model=net,
dataset=train_loader,
optim_fn=optim.Adam,
early_stop=False,
epochs=parameters["epochs"],
optim_args={
"lr": parameters["lr"],
"momentum": parameters["momentum"]
},
)
return
for i in range(REPEATS):
print(f"Running trial {i+1}/{REPEATS}")
net = cnn.NgnCnn(args.neurons, seed=i)
for group in group_config:
net_copy = copy.deepcopy(net)
if "Dropout" in group:
net_copy.dropout = 0.2
if "Neural Noise" in group:
net_copy.neural_noise = (-0.2, 0.5)
net_copy.to(device)
parameters = group_config[group]
log, _ = cnn.train_model(
model=net_copy,
dataset=data_loader,
dtype=dtype,
device=device,
**parameters,
)
accuracy = cnn.predict(net_copy,
data_loader,
device=device,
valid=False)
train_accuracy = cnn.predict(net_copy, data_loader, train=True)
print(group, accuracy)
results.iloc[counter] = [
group, train_accuracy["Accuracy"][0], accuracy["Accuracy"][0], i
]
counter += 1