def train_ME(datafile,
rng,
clip=None,
batch_size=20,
k=10,
lambda_1=0.0,
lambda_2=0.1,
layers_sizes=[40],
pretraining_epochs=[80000],
pretrain_lr=[0.005],
holdout=0.1,
repeats=10,
graph_output=False,
datadir='data'):
print('*** Training on ME ***')
train_set, validation_set = load_n_preprocess_data(datafile,
clip=clip,
holdout=holdout,
repeats=repeats,
datadir=datadir)
return train_bottom_layer(train_set,
validation_set,
batch_size=batch_size,
k=k,
layers_sizes=layers_sizes,
pretraining_epochs=pretraining_epochs,
pretrain_lr=pretrain_lr,
lambda_1=lambda_1,
lambda_2=lambda_2,
rng=rng,
graph_output=graph_output)
def train_DM(datafile,
rng,
clip=None,
batch_size=20,
k=1,
lambda_1=0,
lambda_2=1,
layers_sizes=[400, 40],
pretraining_epochs=[8000, 800],
pretrain_lr=[0.005, 0.1],
holdout=0.1,
repeats=10,
graph_output=False,
datadir='data'):
print('*** Training on DM ***')
train_set, validation_set = load_n_preprocess_data(datafile,
clip=clip,
holdout=holdout,
repeats=repeats,
# transform_fn=numpy.power,
# exponent=1.0/6.0,
datadir=datadir)
return train_bottom_layer(train_set, validation_set,
batch_size=batch_size,
k=k,
layers_sizes=layers_sizes,
pretraining_epochs=pretraining_epochs,
pretrain_lr=pretrain_lr,
lambda_1=lambda_1,
lambda_2=lambda_2,
rng=rng,
graph_output=graph_output)
def MLP_output_from_datafile(self,
datafile,
holdout=0.0,
repeats=1,
clip=None,
transform_fn=None,
exponent=1.0,
datadir='data'):
train_set, validation_set = load_n_preprocess_data(datafile,
holdout=holdout,
clip=clip,
transform_fn=transform_fn,
exponent=exponent,
repeats=repeats,
shuffle=False,
datadir=datadir)
return (self.get_output(train_set), self.get_output(validation_set))
