def get_model(n_features=29,
n_classes=2,
learning_rate=LEARNING_RATE,
verbose=VERBOSE,
batch_size=BATCH_SIZE,
n_steps=N_STEPS,
width=WIDTH,
save_step=100,
cuda=True,
preprocessing=None,
skew=None):
def data_augment(X, y, W, training=True):
if training:
z = np.random.normal(loc=0, scale=WIDTH, size=(X.shape[0]))
X = skew(X, z)
if preprocessing is not None:
X, y, W = preprocessing(X, y, W)
return X, y, W
net = Net(n_features, n_classes)
model = NeuralNet(net,
n_classes=n_classes,
learning_rate=learning_rate,
preprocessing=data_augment,
verbose=verbose,
batch_size=batch_size,
n_steps=n_steps,
width=width,
save_step=save_step,
cuda=cuda)
model.name = 'DataAugmentNeuralNet'
return model
def get_model(n_features=29,
n_classes=2,
learning_rate=LEARNING_RATE,
verbose=VERBOSE,
batch_size=BATCH_SIZE,
n_steps=N_STEPS,
width=WIDTH,
save_step=100,
cuda=True,
preprocessing=None,
skew=None):
def data_augment(X, y, W, training=True):
if training:
z_list = [
np.random.normal(loc=0, scale=WIDTH, size=(X.shape[0]))
for _ in range(5)
]
X = np.concatenate([
X,
] + [skew(X, z) for z in z_list], axis=0)
y = np.concatenate([
y,
] + [y for _ in range(5)], axis=0)
if W is not None:
W = np.concatenate([
W,
] + [W for _ in range(5)], axis=0)
if preprocessing is not None:
X, y, W = preprocessing(X, y, W)
return X, y, W
net = Net(n_features, n_classes)
model = NeuralNet(net,
n_classes=n_classes,
learning_rate=learning_rate,
preprocessing=data_augment,
verbose=verbose,
batch_size=batch_size,
n_steps=n_steps,
width=width,
save_step=save_step,
cuda=cuda)
model.name = 'TrueDataAugmentNeuralNet'
return model
