def __init__(self, numpy_rng, theano_rng=None, cfg = None, dnn = None):
""" Stacked Denoising Autoencoders for DNN Pre-training """
self.cfg = cfg
self.hidden_layers_sizes = cfg.hidden_layers_sizes
self.n_ins = cfg.n_ins
self.hidden_layers_number = len(self.hidden_layers_sizes)
self.dA_layers = []
self.sigmoid_layers = []
self.params = []
if not theano_rng:
theano_rng = RandomStreams(numpy_rng.randint(2 ** 30))
# allocate symbolic variables for the data
self.x = dnn.x
for i in xrange(self.hidden_layers_number):
# the size of the input is either the number of hidden units of
# the layer below, or the input size if we are on the first layer
if i == 0:
input_size = self.n_ins
layer_input = self.x
else:
input_size = self.hidden_layers_sizes[i - 1]
layer_input = dnn.layers[i-1].output
# Construct a denoising autoencoder that shared weights with this layer
if i == 0:
reconstruct_activation = cfg.firstlayer_reconstruct_activation
else:
reconstruct_activation = cfg.hidden_activation
self.sigmoid_layers.append(dnn.layers[i])
self.params.extend(dnn.layers[i].params)
dA_layer = dA(numpy_rng=numpy_rng,
theano_rng=theano_rng,
input=layer_input,
n_visible=input_size,
n_hidden=self.hidden_layers_sizes[i],
W=dnn.layers[i].W,
bhid=dnn.layers[i].b,
sparsity = cfg.sparsity,
sparsity_weight = cfg.sparsity_weight,
hidden_activation = cfg.hidden_activation,
reconstruct_activation = reconstruct_activation,
contraction_level= self.cfg.contraction_level,
n_batchsize = self.cfg.batch_size)
self.dA_layers.append(dA_layer)
def __init__(self, numpy_rng, theano_rng=None, cfg = None):
self.cfg = cfg
self.hidden_layers_sizes = cfg.hidden_layers_sizes
self.n_ins = cfg.n_ins+ cfg.n_outs
self.hidden_layers_number = len(self.hidden_layers_sizes)
self.dA_layers = []
self.sigmoid_layers = []
self.params = []
if not theano_rng:
theano_rng = RandomStreams(numpy_rng.randint(2 ** 30))
# allocate symbolic variables for the data
self.xy = T.matrix('xy', dtype=theano.config.floatX)
for i in xrange(self.hidden_layers_number):
# the size of the input is either the number of hidden units of
# the layer below, or the input size if we are on the first layer
if i == 0:
input_size = self.n_ins
else:
if self.cfg.settings.has_key('firstlayer_xy') and self.cfg.settings['firstlayer_xy'] == 1:
input_size = self.hidden_layers_sizes[i - 1]
else:
input_size = self.hidden_layers_sizes[i - 1] + cfg.n_outs
# Construct a denoising autoencoder that shared weights with this layer
if i == 0:
reconstruct_activation = cfg.firstlayer_reconstruct_activation
else:
reconstruct_activation = cfg.hidden_activation
dA_layer = dA(numpy_rng=numpy_rng,
theano_rng=theano_rng,
input=None,
n_visible=input_size,
n_hidden=self.hidden_layers_sizes[i],
sparsity = cfg.sparsity,
sparsity_weight = cfg.sparsity_weight,
hidden_activation = cfg.hidden_activation,
reconstruct_activation = reconstruct_activation,
contraction_level= self.cfg.contraction_level,
n_batchsize = self.cfg.batch_size)
self.dA_layers.append(dA_layer)