def main(config):
hidden_size = config.hidden_size
hidden_layer_depth = config.hidden_layer_depth
latent_length = config.latent_length
batch_size = config.batch_size
learning_rate = config.learning_rate
n_epochs = config.n_epochs
dropout_rate = config.dropout_rate
optimizer = config.optimizer # options: ADAM, SGD
cuda = config.cuda # options: True, False
print_every = config.print_every
clip = config.clip # options: True, False
max_grad_norm = config.max_grad_norm
loss = config.loss # options: SmoothL1Loss, MSELoss
block = config.block # options: LSTM, GRU
dload = config.dload
# TODO best practice
# sequence_length = 32
#
# number_of_features = 32 * 3
sequence_length = 32 * 32
number_of_features = 3
vrae = VRAE(sequence_length=sequence_length,
number_of_features=number_of_features,
hidden_size=hidden_size,
hidden_layer_depth=hidden_layer_depth,
latent_length=latent_length,
batch_size=batch_size,
learning_rate=learning_rate,
n_epochs=n_epochs,
dropout_rate=dropout_rate,
optimizer=optimizer,
cuda=cuda,
print_every=print_every,
clip=clip,
max_grad_norm=max_grad_norm,
loss=loss,
block=block,
dload=dload)
class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = nn.Conv2d(3, 6, 5)
self.pool = nn.MaxPool2d(2, 2)
self.conv2 = nn.Conv2d(6, 16, 5)
self.fc1 = nn.Linear(16 * 5 * 5, 120)
self.fc2 = nn.Linear(120, 84)
self.fc3 = nn.Linear(84, 10)
return vrae
def define_vrae_model(trial, sequence_length, number_of_features):
hidden_layer_depth = trial.suggest_int("n_hidden_layers", 1, 2)
hidden_size = trial.suggest_int("hidden_size", 30, 60, step=10)
latent_length = trial.suggest_int("latent_length",7, 21, step=7)
batch_size = trial.suggest_int("batch_size", 32, 64, step=32)
learning_rate = trial.suggest_float("learning_rate", 0.0001, 0.0005, step=0.0002)
n_epochs = trial.suggest_int("n_epochs", 800, 1400, step=200)
dropout_rate = trial.suggest_float("learning_rate", 0.1, 0.2, step=0.1)
optimizer = trial.suggest_categorical("optimizer", ["Adam", "SGD"])
loss = trial.suggest_categorical("loss", ["SmoothL1Loss", "MSELoss"])
# block = trial.suggest_categorical("block_choose", ["LSTM", "GRU"])
# hidden_layer_depth = 1
# hidden_size = 30
# latent_length = 10
# batch_size = 64
# learning_rate = 0.00005
# n_epochs = 5
# dropout_rate = 0.0005
# n_epochs = 2
# optimizer = "Adam"
# loss = "SmoothL1Loss"
block = "LSTM"
dload = './ECG200_model_dir/'
print_every = 30
cuda = True
clip = True
max_grad_norm = 5
vrae = VRAE(sequence_length=sequence_length,
number_of_features=number_of_features,
hidden_size=hidden_size,
hidden_layer_depth=hidden_layer_depth,
latent_length=latent_length,
batch_size=batch_size,
learning_rate=learning_rate,
n_epochs=n_epochs,
dropout_rate=dropout_rate,
optimizer=optimizer,
cuda=cuda,
print_every=print_every,
clip=clip,
max_grad_norm=max_grad_norm,
loss=loss,
block=block,
dload=dload)
return vrae
def main(config):
hidden_size = config.hidden_size
hidden_layer_depth = config.hidden_layer_depth
latent_length = config.latent_length
batch_size = config.batch_size
learning_rate = config.learning_rate
n_epochs = config.n_epochs
dropout_rate = config.dropout_rate
optimizer = config.optimizer # options: ADAM, SGD
cuda = config.cuda # options: True, False
print_every = config.print_every
clip = config.clip # options: True, False
max_grad_norm = config.max_grad_norm
loss = config.loss # options: SmoothL1Loss, MSELoss
block = config.block # options: LSTM, GRU
dload = config.dload
sequence_length = X_train.shape[1]
number_of_features = X_train.shape[2]
vrae = VRAE(sequence_length=sequence_length,
number_of_features=number_of_features,
hidden_size=hidden_size,
hidden_layer_depth=hidden_layer_depth,
latent_length=latent_length,
batch_size=batch_size,
learning_rate=learning_rate,
n_epochs=n_epochs,
dropout_rate=dropout_rate,
optimizer=optimizer,
cuda=cuda,
print_every=print_every,
clip=clip,
max_grad_norm=max_grad_norm,
loss=loss,
block=block,
dload=dload)
return vrae
y_train -= base
y_val -= base
train_dataset = TensorDataset(torch.from_numpy(X_train))
test_dataset = TensorDataset(torch.from_numpy(X_val))
sequence_length = X_train.shape[1]
number_of_features = X_train.shape[2]
vrae = VRAE(sequence_length=sequence_length,
number_of_features = number_of_features,
hidden_size = hidden_size,
hidden_layer_depth = hidden_layer_depth,
latent_length = latent_length,
batch_size = batch_size,
learning_rate = learning_rate,
n_epochs = n_epochs,
dropout_rate = dropout_rate,
optimizer = optimizer,
cuda = cuda,
print_every=print_every,
clip=clip,
max_grad_norm=max_grad_norm,
loss = loss,
block = block,
dload = dload)
vrae.fit(train_dataset, save = False)
vrae.save('vraetc110.pth')
#Traversal Plot
z_run = vrae.transform(test_dataset)
for i in range(20):