def __init__(self, numpy_rng, theano_rng=None, n_ins=5000,
hidden_layers_sizes=[500, 500], n_outs=2,
corruption_levels=[0.1, 0.1]):
""" This class is made to support a variable number of layers.
:type numpy_rng: numpy.random.RandomState
:param numpy_rng: numpy random number generator used to draw initial
weights
:type theano_rng: theano.tensor.shared_randomstreams.RandomStreams
:param theano_rng: Theano random generator; if None is given one is
generated based on a seed drawn from `rng`
:type n_ins: int
:param n_ins: dimension of the input to the sdA
:type n_layers_sizes: list of ints
:param n_layers_sizes: intermediate layers size, must contain
at least one value
:type n_outs: int
:param n_outs: dimension of the output of the network
:type corruption_levels: list of float
:param corruption_levels: amount of corruption to use for each
layer
"""
self.sigmoid_layers = []
self.dA_layers = []
self.params = []
self.n_layers = len(hidden_layers_sizes)
assert self.n_layers > 0
if not theano_rng:
theano_rng = RandomStreams(numpy_rng.randint(2 ** 30))
# allocate symbolic variables for the data
self.x = T.matrix('x') # the data is presented as rasterized images
self.y = T.matrix('y') # the actual values for each examples
# The SdA is an MLP, for which all weights of intermediate layers
# are shared with a different denoising autoencoders
# We will first construct the SdA as a deep multilayer perceptron,
# and when constructing each sigmoidal layer we also construct a
# denoising autoencoder that shares weights with that layer
# During pretraining we will train these autoencoders (which will
# lead to chainging the weights of the MLP as well)
# During finetunining we will finish training the SdA by doing
# stochastich gradient descent on the MLP
for i in xrange(self.n_layers):
# construct the sigmoidal layer
# 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 = n_ins
else:
input_size = hidden_layers_sizes[i - 1]
# the input to this layer is either the activation of the hidden
# layer below or the input of the SdA if you are on the first
# layer
if i == 0:
layer_input = self.x
else:
layer_input = self.sigmoid_layers[-1].output
sigmoid_layer = HiddenLayer(rng=numpy_rng,
input=layer_input,
n_in=input_size,
n_out=hidden_layers_sizes[i],
activation=T.nnet.sigmoid)
# add the layer to our list of layers
self.sigmoid_layers.append(sigmoid_layer)
# its arguably a philosophical question...
# but we are going to only declare that the parameters of the
# sigmoid_layers are parameters of the StackedDAA
# the visible biases in the dA are parameters of those
# dA, but not the SdA
self.params.extend(sigmoid_layer.params)
# Construct a denoising autoencoder that shared weights with this
# layer
dA_layer = dA(numpy_rng=numpy_rng,
theano_rng=theano_rng,
input=layer_input,
n_visible=input_size,
n_hidden=hidden_layers_sizes[i],
W=sigmoid_layer.W,
bhid=sigmoid_layer.b)
self.dA_layers.append(dA_layer)
# We now need to add a regression layer on top of the MLP
self.regLayer = Regression(
input=self.sigmoid_layers[-1].output,
n_in=hidden_layers_sizes[-1], n_out=n_outs)
self.params.extend(self.regLayer.params)
# construct a function that implements one step of finetunining
# compute the cost for second phase of training,
# defined as the squared error
self.finetune_cost = self.regLayer.errors(self.y)
# compute the gradients with respect to the model parameters
# symbolic variable that points to the number of errors made on the
# minibatch given by self.x and self.y
self.errors = self.regLayer.errors(self.y)
def build_model(self, load_previous_weights=False):
"""Creates the net's layers from the model settings."""
if load_previous_weights == True:
self.load_weights()
# Load the data
datasets = self.load_samples()
# Train, Validation, Test 100000, 20000, 26... fot Mitocondria set
# Train, Validation, Test 50000, 10000, 10000 times 28x28 = 784 for MNIST dataset
self.train_set_x, self.train_set_y = datasets[0]
self.valid_set_x, self.valid_set_y = datasets[1]
self.test_set_x, self.test_set_y = datasets[2]
# Assumes the width equals the height
img_width_size = numpy.sqrt(self.test_set_x.shape[2].eval()).astype(int)
print img_width_size
assert self.test_set_x.shape[2].eval() == img_width_size * img_width_size, 'input image not square'
print "Image shape %s x %s" % (img_width_size, img_width_size)
nbr_channels = self.test_set_x.shape[1].eval()
self.input_shape = (nbr_channels, img_width_size, img_width_size)
# Compute number of minibatches for training, validation and testing
# Divide the total number of elements in the set by the batch size
self.n_train_batches = self.train_set_x.get_value(borrow=True).shape[0]
self.n_valid_batches = self.valid_set_x.get_value(borrow=True).shape[0]
self.n_test_batches = self.test_set_x.get_value(borrow=True).shape[0]
self.n_train_batches /= self.batch_size
self.n_valid_batches /= self.batch_size
self.n_test_batches /= self.batch_size
print 'Size train_batches %d, n_valid_batches %d, n_test_batches %d' % (self.n_train_batches, self.n_valid_batches, self.n_test_batches)
######################
# BUILD ACTUAL MODEL #
######################
print 'Building the model ...'
# The input is an 4D array of size, number of images in the batch size, number of channels
# (or number of feature maps), image width and height.
#TODO(vpetresc) make nbr of channels variable (1 or 3)
layer_input = self.x.reshape((self.batch_size, nbr_feature_maps, self.input_shape[1], self.input_shape[2]))
pooled_width = self.input_shape[1]
pooled_height = self.input_shape[2]
# Add convolutional layers followed by pooling
clayers = []
idx = 0
for clayer_params in self.convolutional_layers:
print 'Adding conv layer nbr filter %d, Ksize %d' % (clayer_params.num_filters, clayer_params.filter_w)
if load_previous_weights == False:
layer = LeNetConvPoolLayer(self.rng, input=layer_input,
image_shape=(self.batch_size, nbr_feature_maps, pooled_width, pooled_height),
filter_shape=(clayer_params.num_filters, nbr_feature_maps,
clayer_params.filter_w, clayer_params.filter_w),
poolsize=(self.poolsize, self.poolsize))
else:
layer = LeNetConvPoolLayer(self.rng, input=layer_input,
image_shape=(self.batch_size, nbr_feature_maps, pooled_width, pooled_height),
filter_shape=(clayer_params.num_filters, nbr_feature_maps,
clayer_params.filter_w, clayer_params.filter_w),
poolsize=(self.poolsize, self.poolsize),
W=self.cached_weights[itdx+1], b=self.cached_weights[idx])
clayers.append(layer)
pooled_width = (pooled_width - clayer_params.filter_w + 1) / self.poolsize
pooled_height = (pooled_height - clayer_params.filter_w + 1) / self.poolsize
layer_input = layer.output
nbr_feature_maps = clayer_params.num_filters
idx += 2
# Flatten the output of the previous layers and add
# fully connected sigmoidal layers
layer_input = layer_input.flatten(2)
nbr_input = nbr_feature_maps * pooled_width * pooled_height
hlayers = []
for hlayer_params in self.hidden_layers:
print 'Adding hidden layer fully connected %d' % (hlayer_params.num_outputs)
if load_previous_weights == False:
layer = HiddenLayer(self.rng, input=layer_input, n_in=nbr_input,
n_out=hlayer_params.num_outputs, activation=T.tanh)
else:
layer = HiddenLayer(self.rng, input=layer_input, n_in=nbr_input,
n_out=hlayer_params.num_outputs, activation=T.tanh,
W=self.cached_weights[idx+1], b=self.cached_weights[idx])
idx +=2
nbr_input = hlayer_params.num_outputs
layer_input = layer.output
hlayers.append(layer)
# classify the values of the fully-connected sigmoidal layer
if load_previous_weights == False:
self.output_layer = Regression(input=layer_input,
n_in=nbr_input,
n_out=self.last_layer.num_outputs)
else:
self.output_layer = Regression(input=layer_input,
n_in=nbr_input,
n_out=self.last_layer.num_outputs,
W=self.cached_weights[idx+1], b=self.cached_weights[idx])
# the cost we minimize during training is the NLL of the model
self.cost = self.output_layer.squared_loss(self.y)
# Create a list of all model parameters to be fit by gradient descent.
# The parameters are added in reversed order because ofthe order
# in the backpropagation algorithm.
self.params = self.output_layer.params
for hidden_layer in reversed(hlayers):
self.params += hidden_layer.params
for conv_layer in reversed(clayers):
self.params += conv_layer.params
# create a list of gradients for all model parameters
self.grads = T.grad(self.cost, self.params)
class SdA(object):
"""Stacked denoising auto-encoder class (SdA)
A stacked denoising autoencoder model is obtained by stacking several
dAs. The hidden layer of the dA at layer `i` becomes the input of
the dA at layer `i+1`. The first layer dA gets as input the input of
the SdA, and the hidden layer of the last dA represents the output.
Note that after pretraining, the SdA is dealt with as a normal MLP,
the dAs are only used to initialize the weights.
"""
def __init__(self, numpy_rng, theano_rng=None, n_ins=5000,
hidden_layers_sizes=[500, 500], n_outs=2,
corruption_levels=[0.1, 0.1]):
""" This class is made to support a variable number of layers.
:type numpy_rng: numpy.random.RandomState
:param numpy_rng: numpy random number generator used to draw initial
weights
:type theano_rng: theano.tensor.shared_randomstreams.RandomStreams
:param theano_rng: Theano random generator; if None is given one is
generated based on a seed drawn from `rng`
:type n_ins: int
:param n_ins: dimension of the input to the sdA
:type n_layers_sizes: list of ints
:param n_layers_sizes: intermediate layers size, must contain
at least one value
:type n_outs: int
:param n_outs: dimension of the output of the network
:type corruption_levels: list of float
:param corruption_levels: amount of corruption to use for each
layer
"""
self.sigmoid_layers = []
self.dA_layers = []
self.params = []
self.n_layers = len(hidden_layers_sizes)
assert self.n_layers > 0
if not theano_rng:
theano_rng = RandomStreams(numpy_rng.randint(2 ** 30))
# allocate symbolic variables for the data
self.x = T.matrix('x') # the data is presented as rasterized images
self.y = T.matrix('y') # the actual values for each examples
# The SdA is an MLP, for which all weights of intermediate layers
# are shared with a different denoising autoencoders
# We will first construct the SdA as a deep multilayer perceptron,
# and when constructing each sigmoidal layer we also construct a
# denoising autoencoder that shares weights with that layer
# During pretraining we will train these autoencoders (which will
# lead to chainging the weights of the MLP as well)
# During finetunining we will finish training the SdA by doing
# stochastich gradient descent on the MLP
for i in xrange(self.n_layers):
# construct the sigmoidal layer
# 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 = n_ins
else:
input_size = hidden_layers_sizes[i - 1]
# the input to this layer is either the activation of the hidden
# layer below or the input of the SdA if you are on the first
# layer
if i == 0:
layer_input = self.x
else:
layer_input = self.sigmoid_layers[-1].output
sigmoid_layer = HiddenLayer(rng=numpy_rng,
input=layer_input,
n_in=input_size,
n_out=hidden_layers_sizes[i],
activation=T.nnet.sigmoid)
# add the layer to our list of layers
self.sigmoid_layers.append(sigmoid_layer)
# its arguably a philosophical question...
# but we are going to only declare that the parameters of the
# sigmoid_layers are parameters of the StackedDAA
# the visible biases in the dA are parameters of those
# dA, but not the SdA
self.params.extend(sigmoid_layer.params)
# Construct a denoising autoencoder that shared weights with this
# layer
dA_layer = dA(numpy_rng=numpy_rng,
theano_rng=theano_rng,
input=layer_input,
n_visible=input_size,
n_hidden=hidden_layers_sizes[i],
W=sigmoid_layer.W,
bhid=sigmoid_layer.b)
self.dA_layers.append(dA_layer)
# We now need to add a regression layer on top of the MLP
self.regLayer = Regression(
input=self.sigmoid_layers[-1].output,
n_in=hidden_layers_sizes[-1], n_out=n_outs)
self.params.extend(self.regLayer.params)
# construct a function that implements one step of finetunining
# compute the cost for second phase of training,
# defined as the squared error
self.finetune_cost = self.regLayer.errors(self.y)
# compute the gradients with respect to the model parameters
# symbolic variable that points to the number of errors made on the
# minibatch given by self.x and self.y
self.errors = self.regLayer.errors(self.y)
def pretraining_functions(self, train_set_x, batch_size):
''' Generates a list of functions, each of them implementing one
step in trainnig the dA corresponding to the layer with same index.
The function will require as input the minibatch index, and to train
a dA you just need to iterate, calling the corresponding function on
all minibatch indexes.
:type train_set_x: theano.tensor.TensorType
:param train_set_x: Shared variable that contains all datapoints used
for training the dA
:type batch_size: int
:param batch_size: size of a [mini]batch
:type learning_rate: float
:param learning_rate: learning rate used during training for any of
the dA layers
'''
# index to a [mini]batch
index = T.lscalar('index') # index to a minibatch
corruption_level = T.scalar('corruption') # % of corruption to use
learning_rate = T.scalar('lr') # learning rate to use
# number of batches
n_batches = train_set_x.get_value(borrow=True).shape[0] / batch_size
# begining of a batch, given `index`
batch_begin = index * batch_size
# ending of a batch given `index`
batch_end = batch_begin + batch_size
pretrain_fns = []
first_layer = True
for dA in self.dA_layers:
# get the cost and the updates list
if first_layer:
cost, updates = dA.get_cost_updates(corruption_level,
learning_rate,
noise='gaussian')
first_layer = False
else:
cost, updates = dA.get_cost_updates(corruption_level,
learning_rate,
noise='gaussian')
# compile the theano function
fn = theano.function(inputs=[index,
theano.Param(corruption_level, default=0.2),
theano.Param(learning_rate, default=0.1)],
outputs=cost,
updates=updates,
givens={self.x: train_set_x[batch_begin:
batch_end]})
# append `fn` to the list of functions
pretrain_fns.append(fn)
return pretrain_fns
def build_finetune_functions(self, datasets, batch_size):
'''Generates a function `train` that implements one step of
finetuning, a function `validate` that computes the error on
a batch from the validation set, and a function `test` that
computes the error on a batch from the testing set
:type datasets: list of pairs of theano.tensor.TensorType
:param datasets: It is a list that contain all the datasets;
the has to contain three pairs, `train`,
`valid`, `test` in this order, where each pair
is formed of two Theano variables, one for the
datapoints, the other for the labels
:type batch_size: int
:param batch_size: size of a minibatch
:type learning_rate: float
:param learning_rate: learning rate used during finetune stage
'''
(train_set_x, train_set_y) = datasets[0]
(valid_set_x, valid_set_y) = datasets[1]
(test_set_x, test_set_y) = datasets[2]
# compute number of minibatches for training, validation and testing
n_valid_batches = valid_set_x.get_value(borrow=True).shape[0]
n_valid_batches /= batch_size
n_test_batches = test_set_x.get_value(borrow=True).shape[0]
n_test_batches /= batch_size
index = T.lscalar('index') # index to a [mini]batch
learning_rate = T.scalar('lr')
# compute the gradients with respect to the model parameters
gparams = T.grad(self.finetune_cost, self.params)
# compute list of fine-tuning updates
updates = []
for param, gparam in zip(self.params, gparams):
updates.append((param, param - gparam * learning_rate))
train_fn = theano.function(inputs=[index,
theano.Param(learning_rate, default=0.1)],
outputs=self.finetune_cost,
updates=updates,
givens={
self.x: train_set_x[index * batch_size:
(index + 1) * batch_size],
self.y: train_set_y[index * batch_size:
(index + 1) * batch_size]},
name='train')
test_score_i = theano.function([index], self.errors,
givens={
self.x: test_set_x[index * batch_size:
(index + 1) * batch_size],
self.y: test_set_y[index * batch_size:
(index + 1) * batch_size]},
name='test')
valid_score_i = theano.function([index], self.errors,
givens={
self.x: valid_set_x[index * batch_size:
(index + 1) * batch_size],
self.y: valid_set_y[index * batch_size:
(index + 1) * batch_size]},
name='valid')
# Create a function that scans the entire validation set
def valid_score():
return [valid_score_i(i) for i in xrange(n_valid_batches)]
# Create a function that scans the entire test set
def test_score():
return [test_score_i(i) for i in xrange(n_test_batches)]
return train_fn, valid_score, test_score
class CNNTrainRegression(CNNBase):
"""The class takes a proto bufer as input, setups a CNN according to the
settings, trains the network and saves the weights in a file
"""
def __init__(self, protofile, cached_weights):
"""
:param protofile: describes the arhitecture of the network
:type protofile: string
:param cached_weights: filename of the weights
:type cached_weights: string
"""
self.cnntype = 'TRAIN' #: extension that is added to the logfile name
super(CNNTrain, self).__init__(protofile, cached_weights)
#: Array of train data of size num samples x dimension of sample
self.train_set_x = None
#: Array of target data of size num samples x 1
self.train_set_y = None
#: Array of valid data of size num samples x dimension of sample
self.valid_set_x = None
self.valid_set_y = None
self.test_set_x = None
self.test_set_y = None
#: The number of batches in which train_set_x is divided
self.n_train_batches = 0
self.n_valid_batches = 0
self.n_test_batches = 0
#: The cost that is minimized by the algorithm, usually log likelihood
self.cost = 0
#: Array of symbolic gradients of the cost wrt. weights
self.grads = None
#: Array of weights (plus biases) for which the gradient is computed
self.params = None
#: Usually logistic regression
self.output_layer = None
#: The size of the input array that is being passed to the algorithm
#: It has size num samples x num channels x img width x img height
self.input_shape = None
self.index = T.lscalar() #: index to a [mini]batch
self.x = T.matrix('x') #: the data is presented as rasterized images
self.y = T.ivector('y') #: the labels are presented as 1D vector of ints
def build_model(self, load_previous_weights=False):
"""Creates the net's layers from the model settings."""
if load_previous_weights == True:
self.load_weights()
# Load the data
datasets = self.load_samples()
# Train, Validation, Test 100000, 20000, 26... fot Mitocondria set
# Train, Validation, Test 50000, 10000, 10000 times 28x28 = 784 for MNIST dataset
self.train_set_x, self.train_set_y = datasets[0]
self.valid_set_x, self.valid_set_y = datasets[1]
self.test_set_x, self.test_set_y = datasets[2]
# Assumes the width equals the height
img_width_size = numpy.sqrt(self.test_set_x.shape[2].eval()).astype(int)
print img_width_size
assert self.test_set_x.shape[2].eval() == img_width_size * img_width_size, 'input image not square'
print "Image shape %s x %s" % (img_width_size, img_width_size)
nbr_channels = self.test_set_x.shape[1].eval()
self.input_shape = (nbr_channels, img_width_size, img_width_size)
# Compute number of minibatches for training, validation and testing
# Divide the total number of elements in the set by the batch size
self.n_train_batches = self.train_set_x.get_value(borrow=True).shape[0]
self.n_valid_batches = self.valid_set_x.get_value(borrow=True).shape[0]
self.n_test_batches = self.test_set_x.get_value(borrow=True).shape[0]
self.n_train_batches /= self.batch_size
self.n_valid_batches /= self.batch_size
self.n_test_batches /= self.batch_size
print 'Size train_batches %d, n_valid_batches %d, n_test_batches %d' % (self.n_train_batches, self.n_valid_batches, self.n_test_batches)
######################
# BUILD ACTUAL MODEL #
######################
print 'Building the model ...'
# The input is an 4D array of size, number of images in the batch size, number of channels
# (or number of feature maps), image width and height.
#TODO(vpetresc) make nbr of channels variable (1 or 3)
layer_input = self.x.reshape((self.batch_size, nbr_feature_maps, self.input_shape[1], self.input_shape[2]))
pooled_width = self.input_shape[1]
pooled_height = self.input_shape[2]
# Add convolutional layers followed by pooling
clayers = []
idx = 0
for clayer_params in self.convolutional_layers:
print 'Adding conv layer nbr filter %d, Ksize %d' % (clayer_params.num_filters, clayer_params.filter_w)
if load_previous_weights == False:
layer = LeNetConvPoolLayer(self.rng, input=layer_input,
image_shape=(self.batch_size, nbr_feature_maps, pooled_width, pooled_height),
filter_shape=(clayer_params.num_filters, nbr_feature_maps,
clayer_params.filter_w, clayer_params.filter_w),
poolsize=(self.poolsize, self.poolsize))
else:
layer = LeNetConvPoolLayer(self.rng, input=layer_input,
image_shape=(self.batch_size, nbr_feature_maps, pooled_width, pooled_height),
filter_shape=(clayer_params.num_filters, nbr_feature_maps,
clayer_params.filter_w, clayer_params.filter_w),
poolsize=(self.poolsize, self.poolsize),
W=self.cached_weights[itdx+1], b=self.cached_weights[idx])
clayers.append(layer)
pooled_width = (pooled_width - clayer_params.filter_w + 1) / self.poolsize
pooled_height = (pooled_height - clayer_params.filter_w + 1) / self.poolsize
layer_input = layer.output
nbr_feature_maps = clayer_params.num_filters
idx += 2
# Flatten the output of the previous layers and add
# fully connected sigmoidal layers
layer_input = layer_input.flatten(2)
nbr_input = nbr_feature_maps * pooled_width * pooled_height
hlayers = []
for hlayer_params in self.hidden_layers:
print 'Adding hidden layer fully connected %d' % (hlayer_params.num_outputs)
if load_previous_weights == False:
layer = HiddenLayer(self.rng, input=layer_input, n_in=nbr_input,
n_out=hlayer_params.num_outputs, activation=T.tanh)
else:
layer = HiddenLayer(self.rng, input=layer_input, n_in=nbr_input,
n_out=hlayer_params.num_outputs, activation=T.tanh,
W=self.cached_weights[idx+1], b=self.cached_weights[idx])
idx +=2
nbr_input = hlayer_params.num_outputs
layer_input = layer.output
hlayers.append(layer)
# classify the values of the fully-connected sigmoidal layer
if load_previous_weights == False:
self.output_layer = Regression(input=layer_input,
n_in=nbr_input,
n_out=self.last_layer.num_outputs)
else:
self.output_layer = Regression(input=layer_input,
n_in=nbr_input,
n_out=self.last_layer.num_outputs,
W=self.cached_weights[idx+1], b=self.cached_weights[idx])
# the cost we minimize during training is the NLL of the model
self.cost = self.output_layer.squared_loss(self.y)
# Create a list of all model parameters to be fit by gradient descent.
# The parameters are added in reversed order because ofthe order
# in the backpropagation algorithm.
self.params = self.output_layer.params
for hidden_layer in reversed(hlayers):
self.params += hidden_layer.params
for conv_layer in reversed(clayers):
self.params += conv_layer.params
# create a list of gradients for all model parameters
self.grads = T.grad(self.cost, self.params)
def train_model(self):
"""Abstract method to be implemented by subclasses"""
raise NotImplementedError()