test_set.X = 2* test_set.X.reshape(-1, 784) - 1.
# flatten targets
test_set.y = test_set.y.reshape(-1)
print('Building the MLP...')
# Prepare Theano variables for inputs and targets
input = T.matrix('inputs')
target = T.vector('targets')
mlp = lasagne.layers.InputLayer(shape=(None, 784),input_var=input)
# Input layer is not binary -> use baseline kernel in first hidden layer
mlp = binary_ops.DenseLayer(
mlp,
nonlinearity=lasagne.nonlinearities.identity,
num_units=num_units,
kernel = "baseline")
mlp = lasagne.layers.BatchNormLayer(mlp)
mlp = lasagne.layers.NonlinearityLayer(mlp,nonlinearity=binary_ops.SignTheano)
for k in range(1,n_hidden_layers):
mlp = binary_ops.DenseLayer(
mlp,
nonlinearity=lasagne.nonlinearities.identity,
num_units=num_units,
kernel = kernel)
mlp = lasagne.layers.BatchNormLayer(mlp)
cnn = binary_ops.Conv2DLayer(cnn,
num_filters=512,
filter_size=(3, 3),
pad=1,
nonlinearity=lasagne.nonlinearities.identity,
convOp=convOp)
cnn = lasagne.layers.MaxPool2DLayer(cnn, pool_size=(2, 2))
cnn = lasagne.layers.BatchNormLayer(cnn)
cnn = lasagne.layers.NonlinearityLayer(cnn,
nonlinearity=binary_ops.SignTheano)
cnn = binary_ops.DenseLayer(cnn,
nonlinearity=lasagne.nonlinearities.identity,
num_units=1024,
kernel=kernel)
cnn = lasagne.layers.BatchNormLayer(cnn)
cnn = lasagne.layers.NonlinearityLayer(cnn,
nonlinearity=binary_ops.SignTheano)
cnn = binary_ops.DenseLayer(cnn,
nonlinearity=lasagne.nonlinearities.identity,
num_units=1024,
kernel=kernel)
cnn = lasagne.layers.BatchNormLayer(cnn)
cnn = lasagne.layers.NonlinearityLayer(cnn,
nonlinearity=binary_ops.SignTheano)