def train_bottom_layer(train_set,
validation_set,
batch_size=20,
k=1,
layers_sizes=[40],
pretraining_epochs=[800],
pretrain_lr=[0.005],
lambda_1=0.0,
lambda_2=0.1,
rng=None,
graph_output=False):
print('Visible nodes: %i' % train_set.get_value().shape[1])
print('Output nodes: %i' % layers_sizes[-1])
dbn = DBN(numpy_rng=rng,
n_ins=train_set.get_value().shape[1],
hidden_layers_sizes=layers_sizes[:-1],
n_outs=layers_sizes[-1])
dbn.training(train_set,
batch_size,
k=k,
pretraining_epochs=pretraining_epochs,
pretrain_lr=pretrain_lr,
lambda_1=lambda_1,
lambda_2=lambda_2,
validation_set_x=validation_set,
graph_output=graph_output)
output_train_set = dbn.get_output(train_set)
if validation_set is not None:
output_val_set = dbn.get_output(validation_set)
else:
output_val_set = None
return dbn, output_train_set, output_val_set
