def ae(batch):
NUM_FILTERS = 8
input_shape = batch.input.shape
target_shape = batch.target.shape
seq_length = input_shape[1]
input_layer = InputLayer(shape=input_shape)
# Dense layers
dense_layer_0 = DenseLayer(input_layer,
num_units=(seq_length - 3) * NUM_FILTERS)
dense_layer_1 = DenseLayer(dense_layer_0, num_units=128)
dense_layer_2 = DenseLayer(dense_layer_1,
num_units=(seq_length - 3) * NUM_FILTERS)
# Output
final_dense_layer = DenseLayer(dense_layer_2,
num_units=target_shape[1] * target_shape[2],
nonlinearity=None)
output_layer = ReshapeLayer(final_dense_layer, shape=target_shape)
net = Net(output_layer,
tags=['AE'],
description=
"Like AE in e575 but much larger layers. Still no conv layers.",
predecessor_experiment="e575")
return net
def ae(batch):
NUM_FILTERS = 8
input_shape = batch.input.shape
target_shape = input_shape
seq_length = input_shape[1]
output_layer = build_net(
input_shape=input_shape,
layers=[
{
'type': DenseLayer,
'num_units': (seq_length - 3) * NUM_FILTERS,
'nonlinearity': rectify
},
{
'type': DenseLayer,
'num_units': 128,
'nonlinearity': rectify
},
{
'type': DenseLayer,
'num_units': (seq_length - 3) * NUM_FILTERS,
'nonlinearity': rectify
},
# Output
{
'type': DenseLayer,
'num_units': target_shape[1] * target_shape[2],
'nonlinearity': None
},
{
'type': ReshapeLayer,
'shape': target_shape
}
])
net = Net(output_layer,
tags=['AE'],
description="Identical AE to e576 but with rectify.",
predecessor_experiment="e576")
return net
def ae(batch):
input_shape = batch.input.shape
target_shape = batch.target.shape
input_layer = InputLayer(shape=input_shape)
# Dense layers
dense_layer_0 = DenseLayer(input_layer, num_units=128)
dense_layer_1 = DenseLayer(dense_layer_0, num_units=64)
dense_layer_2 = DenseLayer(dense_layer_1, num_units=128)
# Output
final_dense_layer = DenseLayer(dense_layer_2,
num_units=target_shape[1] * target_shape[2],
nonlinearity=None)
output_layer = ReshapeLayer(final_dense_layer, shape=target_shape)
net = Net(output_layer,
tags=['AE'],
description="Like AE in e567 but without conv layer",
predecessor_experiment="e567")
return net
def ae(batch):
NUM_FILTERS = 8
input_shape = batch.input.shape
seq_length = input_shape[1]
output_layer = build_net(
input_shape=input_shape,
layers=[
{
'type': DimshuffleLayer,
'pattern': (0, 2, 1) # (batch, features, time)
},
{
'type': Conv1DLayer, # convolve over the time axis
'num_filters': NUM_FILTERS,
'filter_size': 4,
'stride': 1,
'nonlinearity': None,
'pad': 'valid'
},
{
'type': DimshuffleLayer,
'pattern': (0, 2, 1) # back to (batch, time, features)
},
{
'type': DenseLayer,
'num_units': (seq_length - 3) * NUM_FILTERS,
'nonlinearity': rectify
},
{
'type': DenseLayer,
'num_units': 128,
'nonlinearity': rectify
},
{
'type': DenseLayer,
'num_units': (seq_length - 3) * NUM_FILTERS,
'nonlinearity': rectify
},
{
'type': ReshapeLayer,
'shape': (-1, (seq_length - 3), NUM_FILTERS)
},
{
'type': DimshuffleLayer,
'pattern': (0, 2, 1) # (batch, features, time)
},
{ # DeConv
'type': Conv1DLayer,
'num_filters': 1,
'filter_size': 4,
'stride': 1,
'nonlinearity': None,
'pad': 'full'
},
{
'type': DimshuffleLayer,
'pattern': (0, 2, 1) # back to (batch, time, features)
}
])
net = Net(output_layer,
tags=['AE', 'Conv1D'],
description="Identical AE to e567. Trained on 5 appliances.",
predecessor_experiment="e567")
return net