def train_mlp(NET, sgd_params, datasets):
"""Run mlp training test."""
# Train the net
NT.train_mlp(NET=NET, \
sgd_params=sgd_params, \
datasets=datasets)
return
def train_mlp(NET, sgd_params, datasets):
"""Run mlp training test."""
# Train the net
NT.train_mlp(NET=NET, \
sgd_params=sgd_params, \
datasets=datasets)
return
def train_ss_mlp(NET, sgd_params, datasets):
"""Run semi-supervised EA-regularized test."""
# Run training on the given NET
NT.train_ss_mlp(NET=NET, \
sgd_params=sgd_params, \
datasets=datasets)
return
def train_ss_mlp(NET, sgd_params, datasets):
"""Run semi-supervised EA-regularized test."""
# Run training on the given NET
NT.train_ss_mlp(NET=NET, \
sgd_params=sgd_params, \
datasets=datasets)
return
def train_dae(NET, dae_layer, sgd_params, datasets):
"""Run DAE training test."""
# Run denoising autoencoder training on the given layer of NET
NT.train_dae(NET=NET, \
dae_layer=dae_layer, \
sgd_params=sgd_params, \
datasets=datasets)
return
def train_dae(NET, dae_layer, sgd_params, datasets):
"""Run DAE training test."""
# Run denoising autoencoder training on the given layer of NET
NT.train_dae(NET=NET, \
dae_layer=dae_layer, \
sgd_params=sgd_params, \
datasets=datasets)
return
def train_dae(NET, dae_layer, mlp_params, sgd_params):
"""Run DAE training test."""
# Load some data to train/validate/test with
dataset = 'data/mnist.pkl.gz'
datasets = load_udm(dataset)
# Run denoising autoencoder training on the given layer of NET
NT.train_dae(NET=NET, \
dae_layer=dae_layer, \
mlp_params=mlp_params, \
sgd_params=sgd_params, \
datasets=datasets)
return 1
def train_dae(NET, dae_layer, mlp_params, sgd_params):
"""Run DAE training test."""
# Load some data to train/validate/test with
dataset = 'data/mnist.pkl.gz'
datasets = load_udm(dataset)
# Run denoising autoencoder training on the given layer of NET
NT.train_dae(NET=NET, \
dae_layer=dae_layer, \
mlp_params=mlp_params, \
sgd_params=sgd_params, \
datasets=datasets)
return
def train_ss_mlp(NET, mlp_params, sgd_params, rng, su_count=1000):
"""Run semisupervised DEV-regularized test."""
# Load some data to train/validate/test with
dataset = 'data/mnist.pkl.gz'
datasets = load_udm_ss(dataset, su_count, rng)
# Tell the net that it's semisupervised, which will force it to use only
# unlabeled examples for computing the DEV regularizer.
NET.is_semisupervised = 1
# Run training on the given NET
NT.train_ss_mlp(NET=NET, \
mlp_params=mlp_params, \
sgd_params=sgd_params, \
datasets=datasets)
return 1
def train_ss_mlp(NET, mlp_params, sgd_params, rng, su_count=1000):
"""Run semisupervised DEV-regularized test."""
# Load some data to train/validate/test with
dataset = 'data/mnist.pkl.gz'
datasets = load_udm_ss(dataset, su_count, rng)
# Tell the net that it's semisupervised, which will force it to use only
# unlabeled examples for computing the DEV regularizer.
NET.is_semisupervised = 1
# Run training on the given NET
NT.train_ss_mlp(NET=NET, \
mlp_params=mlp_params, \
sgd_params=sgd_params, \
datasets=datasets)
return
def train_mlp(NET, mlp_params, sgd_params):
"""Run mlp training test."""
# Load some data to train/validate/test with
#dataset = 'data/mnist.pkl.gz'
#datasets = load_udm(dataset)
dataset = 'data/mnist_batches.npz'
datasets = load_mnist(dataset)
# Tell the net that it's not semisupervised, which will force it to use
# _all_ examples for computing the DEV regularizer.
NET.is_semisupervised = 0
# Train the net
NT.train_mlp(NET=NET, \
mlp_params=mlp_params, \
sgd_params=sgd_params, \
datasets=datasets)
return 1
def train_mlp(NET, mlp_params, sgd_params):
"""Run mlp training test."""
# Load some data to train/validate/test with
#dataset = 'data/mnist.pkl.gz'
#datasets = load_udm(dataset)
dataset = 'data/mnist_batches.npz'
datasets = load_mnist(dataset)
# Tell the net that it's not semisupervised, which will force it to use
# _all_ examples for computing the DEV regularizer.
NET.is_semisupervised = 0
# Train the net
NT.train_mlp(NET=NET, \
mlp_params=mlp_params, \
sgd_params=sgd_params, \
datasets=datasets)
return
