def Discriminator(real, fake):
in_len = 784
m_disc = 512
batch_size = 64
pair = T.concatenate([real, fake], axis=0)
h1 = HiddenLayer(in_len, m_disc)
h2 = HiddenLayer(m_disc, m_disc)
h3 = HiddenLayer(m_disc, 1, activation='sigmoid')
pc1 = h1.output(pair)
pc2 = h2.output(pc1)
pc3 = h3.output(pc2)
p_real = pc3[:real.shape[0], :].flatten()
p_gen = pc3[-real.shape[0]:, :].flatten()
d_cost_real = binary_crossentropy(p_real, T.ones(p_real.shape)).mean()
d_cost_real = (d_cost_real * (d_cost_real < 0.9)).mean()
d_cost_gen = binary_crossentropy(p_gen, T.zeros(p_gen.shape)).mean()
d_cost_gen = (d_cost_gen * (d_cost_gen > 0.1)).mean()
g_cost_d = binary_crossentropy(p_gen, T.ones(p_gen.shape)).mean()
d_cost = (d_cost_real + d_cost_gen) / 2.0
g_cost = g_cost_d
layers = [h1, h2, h3]
params = layers2params(layers)
return d_cost, g_cost, params
def __init__(self,
rng,
input,
input_dim,
n_in,
n_hidden,
n_out,
test=False,
classifier=None,
dropout_p=0.5):
if test == False:
input = input.reshape(input_dim)
# Initialize hidden layer
self.hiddenLayer = HiddenLayer(
rng=rng,
input=input,
n_in=n_in,
n_out=n_hidden,
)
# Initialize output layer
self.last_layer = HiddenLayer(rng=rng,
input=self.hiddenLayer.output,
n_in=n_hidden,
n_out=n_out)
# save parameters
self.params = self.hiddenLayer.params + self.last_layer.params
else:
input = input.reshape(input_dim)
self.hiddenLayer = classifier.hiddenLayer.TestVersion(
rng, input, n_in, n_out)
self.last_layer = classifier.last_layer.TestVersion(
rng, input=self.hiddenLayer.output, n_in=n_hidden, n_out=n_out)
def __init__(self,
input,
n_in,
n_hidden,
n_out,
rng=None,
is_binary=False,
params_first_phase=None):
"""
Initialize the parameters for the multilayer perceptron
:type rng: numpy.random.RandomState
:param rng: a random number generator used to initialize weights
:type input: theano.tensor.TensorType
:param input: symbolic variable that describes the input of the
architecture (one minibatch)
:type n_in: int
:param n_in: number of input units, the dimension of the space in
which the datapoints lie
:type n_hidden: int
:param n_hidden: number of hidden units
:type n_out: int
:param n_out: number of output units, the dimension of the space in which
the labels lie.
"""
self.input = input
if rng == None:
rng = numpy.random.RandomState(1234)
if DEBUGGING:
theano.config.compute_test_value = 'raise'
self.input.tag.test_value = numpy.random.rand(1800, n_in)
# Since we are dealing with a one hidden layer MLP, this will
# translate into a TanhLayer connected to the LogisticRegression
# layer; this can be replaced by a SigmoidalLayer, or a layer
# implementing any other nonlinearity
self.hiddenLayer = HiddenLayer(rng=rng, input=self.input,
n_in=n_in, n_out=n_hidden,
activation=T.tanh)
self.state = "train"
self.is_binary = is_binary
self.out_dir = "out/"
# The logistic regression layer gets as input the hidden units
# of the hidden layer
self.logRegressionLayer = LogisticRegressionLayer(
input=self.hiddenLayer.output,
n_in=n_hidden,
n_out=n_out,
is_binary=is_binary,
rng=rng)
# L1 norm ; one regularization option is to enforce L1 norm to
# be small
self.L1 = abs(self.hiddenLayer.W).sum() \
+ abs(self.logRegressionLayer.W).sum()
# square of L2 norm ; one regularization option is to enforce
# square of L2 norm to be small
self.L2_sqr = (self.hiddenLayer.W ** 2).sum() \
+ (self.logRegressionLayer.W ** 2).sum()
# negative log likelihood of the MLP is given by the
# crossentropy of the output of the model, computed in the
# logistic regression layer
if is_binary:
self.crossentropy = self.logRegressionLayer.crossentropy
else:
self.crossentropy = self.logRegressionLayer.crossentropy_categorical
# negative log likelihood of the MLP is given by the negative
# log likelihood of the output of the model, computed in the
# logistic regression layer
self.negative_log_likelihood = self.logRegressionLayer.negative_log_likelihood
# same holds for the function computing the number of errors
self.errors = self.logRegressionLayer.errors
# Class memberships
self.class_memberships = self.logRegressionLayer.get_class_memberships(self.hiddenLayer.get_outputs(self.input))
self.train_pkl_file = "post_train_n_hidden_" + str(n_hidden) + "n_out_" + str(n_out)
self.test_pkl_file = "post_test_n_hidden_" + str(n_hidden) + "n_out_" + str(n_out)
if params_first_phase is not None:
self.params = params_first_phase
# the parameters of the model are the parameters of the two layer it is
# made out of
self.params = self.hiddenLayer.params + self.logRegressionLayer.params
self.data_dict = { 'Ws':[],
'costs':[],
'test_probs':[],
'train_probs':[],
'test_scores':[]}
class PosttrainMLP:
"""Post training:- Second phase MLP.
A multilayer perceptron is a feedforward artificial neural network model
that has one layer or more of hidden units and nonlinear activations.
Intermediate layers usually have as activation function thanh or the
sigmoid function (defined here by a ``SigmoidalLayer`` class) while the
top layer is a softamx layer (defined here by a ``LogisticRegression``
class).
"""
def __init__(self,
input,
n_in,
n_hidden,
n_out,
rng=None,
is_binary=False,
params_first_phase=None):
"""
Initialize the parameters for the multilayer perceptron
:type rng: numpy.random.RandomState
:param rng: a random number generator used to initialize weights
:type input: theano.tensor.TensorType
:param input: symbolic variable that describes the input of the
architecture (one minibatch)
:type n_in: int
:param n_in: number of input units, the dimension of the space in
which the datapoints lie
:type n_hidden: int
:param n_hidden: number of hidden units
:type n_out: int
:param n_out: number of output units, the dimension of the space in which
the labels lie.
"""
self.input = input
if rng == None:
rng = numpy.random.RandomState(1234)
if DEBUGGING:
theano.config.compute_test_value = 'raise'
self.input.tag.test_value = numpy.random.rand(1800, n_in)
# Since we are dealing with a one hidden layer MLP, this will
# translate into a TanhLayer connected to the LogisticRegression
# layer; this can be replaced by a SigmoidalLayer, or a layer
# implementing any other nonlinearity
self.hiddenLayer = HiddenLayer(rng=rng, input=self.input,
n_in=n_in, n_out=n_hidden,
activation=T.tanh)
self.state = "train"
self.is_binary = is_binary
self.out_dir = "out/"
# The logistic regression layer gets as input the hidden units
# of the hidden layer
self.logRegressionLayer = LogisticRegressionLayer(
input=self.hiddenLayer.output,
n_in=n_hidden,
n_out=n_out,
is_binary=is_binary,
rng=rng)
# L1 norm ; one regularization option is to enforce L1 norm to
# be small
self.L1 = abs(self.hiddenLayer.W).sum() \
+ abs(self.logRegressionLayer.W).sum()
# square of L2 norm ; one regularization option is to enforce
# square of L2 norm to be small
self.L2_sqr = (self.hiddenLayer.W ** 2).sum() \
+ (self.logRegressionLayer.W ** 2).sum()
# negative log likelihood of the MLP is given by the
# crossentropy of the output of the model, computed in the
# logistic regression layer
if is_binary:
self.crossentropy = self.logRegressionLayer.crossentropy
else:
self.crossentropy = self.logRegressionLayer.crossentropy_categorical
# negative log likelihood of the MLP is given by the negative
# log likelihood of the output of the model, computed in the
# logistic regression layer
self.negative_log_likelihood = self.logRegressionLayer.negative_log_likelihood
# same holds for the function computing the number of errors
self.errors = self.logRegressionLayer.errors
# Class memberships
self.class_memberships = self.logRegressionLayer.get_class_memberships(self.hiddenLayer.get_outputs(self.input))
self.train_pkl_file = "post_train_n_hidden_" + str(n_hidden) + "n_out_" + str(n_out)
self.test_pkl_file = "post_test_n_hidden_" + str(n_hidden) + "n_out_" + str(n_out)
if params_first_phase is not None:
self.params = params_first_phase
# the parameters of the model are the parameters of the two layer it is
# made out of
self.params = self.hiddenLayer.params + self.logRegressionLayer.params
self.data_dict = { 'Ws':[],
'costs':[],
'test_probs':[],
'train_probs':[],
'test_scores':[]}
def _shared_dataset(self, data_x, name="x"):
shared_x = theano.shared(numpy.asarray(list(data_x), dtype=theano.config.floatX))
shared_x.name = name
return shared_x
def save_data(self):
output = None
if self.state == "train":
output = open(self.out_dir + self.train_pkl_file + ".pkl", 'wb')
else:
output = open(self.out_dir + self.test_pkl_file + ".pkl", 'wb')
pkl.dump(self.data_dict, output)
self.data_dict['Ws'] = []
self.data_dict['costs'] = []
self.data_dict['test_scores'] = []
def posttrain(self,
learning_rate=0.1,
L1_reg=0.00,
L2_reg=0.0001,
n_epochs=80,
data=None,
labels=None,
cost_type=Costs.Crossentropy,
save_exp_data=False,
batch_size=20):
if data is None:
raise Exception("Post-training can't start without pretraining class membership probabilities.")
if labels is None:
raise Exception("Post-training can not start without posttraining class labels.")
self.state = "train"
self.learning_rate = learning_rate
train_set_x = self._shared_dataset(data, name="training_set")
train_set_y = labels
# compute number of minibatches for training
n_examples = train_set_x.get_value(borrow=True).shape[0]
n_train_batches = n_examples / batch_size
######################
# BUILD ACTUAL MODEL #
######################
print '...postraining the model'
# allocate symbolic variables for the data
index = T.lscalar('index') # index to a [mini]batch
y = T.ivector('y') # the labels are presented as 1D vector of int32
mode = "FAST_COMPILE" #DEBUG_MODE"
if DEBUGGING:
self.input.tag.test_value = numpy.random.rand(self)
index.tag.test_value = 0
y.tag.test_value = numpy.ones(n_examples)
mode = "DEBUG_MODE"
# the cost we minimize during training is the negative log likelihood of
# the model plus the regularization terms (L1 and L2); cost is expressed
# here symbolically.
cost = None
if cost_type == Costs.NegativeLikelihood:
cost = self.negative_log_likelihood(y) \
+ L1_reg * self.L1 \
+ L2_reg * self.L2_sqr
elif cost_type == Costs.Crossentropy:
cost = self.crossentropy(y) \
+ L1_reg * self.L1 \
+ L2_reg * self.L2_sqr
gparams = []
for param in self.params:
gparam = T.grad(cost, param)
gparams.append(gparam)
# specify how to update the parameters of the model as selfa dictionary
updates = {}
# given two list the zip A = [a1, a2, a3, a4] and B = [b1, b2, b3, b4] of
# same length, zip generates a list C of same size, where each element
# is a pair formed from the two lists :
# C = [(a1, b1), (a2, b2), (a3, b3) , (a4, b4)]
for param, gparam in zip(self.params, gparams):
updates[param] = param - learning_rate * gparam
# compiling a Theano function `train_model` that returns the cost, butx
# in the same time updates the parameter of the model based on the rules
# defined in `updates`
# p_y_given_x = self.class_memberships
train_model = theano.function(inputs=[index],
outputs=cost,
updates = updates,
givens = {
self.input: train_set_x[index * batch_size:(index + 1) * batch_size],
y: train_set_y[index * batch_size: (index + 1) * batch_size]
},
mode=mode)
if DEBUGGING:
theano.printing.debugprint(train_model)
epoch = 0
costs = []
Ws = []
while (epoch < n_epochs):
for minibatch_index in xrange(n_train_batches):
print "Postraining in Minibatch %i " % (minibatch_index)
minibatch_avg_cost = train_model(minibatch_index)
costs.append(float(minibatch_avg_cost))
Ws.append(self.params[2])
epoch +=1
if save_exp_data:
self.data_dict['Ws'].append(Ws)
self.data_dict['costs'].append([costs])
self.save_data()
return costs
def posttest(self,
data=None,
labels=None,
save_exp_data = False,
batch_size=20):
if data is None:
raise Exception("Post-training can't start without pretraining class membership probabilities.")
if labels is None:
raise Exception("Post-training can not start without posttraining class-membership probabilities.")
test_set_x = shared_dataset(data)
test_set_y = labels
self.state = "test"
# compute number of minibatches for training, validation and testing
n_examples = test_set_x.get_value(borrow=True).shape[0]
n_test_batches = n_examples / batch_size
print '...post-testing the model'
# allocate symbolic variables for the data
index = T.lscalar() # index to a [mini]batch
y = T.ivector('y') # the labels are presented as 1D vector of
# [int] labels
mode = "FAST_RUN"
if DEBUGGING:
theano.config.compute_test_value = 'raise'
index.tag.test_value = 0
y.tag.test_value = numpy.ones(n_examples)
mode = "DEBUG_MODE"
# the cost we minimize during training is the negative log likelihood of
# the model plus the regularization terms (L1 and L2); cost is expressed
# here symbolically
# compiling a Theano function `test_model` that returns the cost, but
# in the same time updates the parameter of the model based on the rules
# defined in `updates`
test_model = theano.function(inputs=[index],
outputs=self.errors(y),
givens={
self.input: test_set_x[index * batch_size:(index + 1) * batch_size],
y: test_set_y[index * batch_size: (index + 1) * batch_size]},
mode=mode)
###############
# TEST MODEL #
###############
test_losses = []
for minibatch_index in xrange(n_test_batches):
test_losses.append(float(test_model(minibatch_index)))
test_score = numpy.mean(test_losses)
print("Minibatch %i, mean test error %f" % (minibatch_index, test_losses[minibatch_index] * 100))
print "In the end test score is %f" %(test_score * 100)
if save_exp_data:
self.data_dict['test_scores'].append(test_losses)
self.save_data()
return test_score, test_losses
def __init__(self, rng, input, n_in, n_hidden, n_out):
"""Initialize the parameters for the multilayer perceptron
:type rng: numpy.random.RandomState
:param rng: a random number generator used to initialize weights
:type input: theano.tensor.TensorType
:param input: symbolic variable that describes the input of the
architecture (one minibatch)
:type n_in: int
:param n_in: number of input units, the dimension of the space in
which the datapoints lie
:type n_hidden: int
:param n_hidden: number of hidden units
:type n_out: int
:param n_out: number of output units, the dimension of the space in
which the labels lie
"""
# Since we are dealing with a one hidden layer MLP, this will translate
# into a HiddenLayer with a tanh activation function connected to the
# LogisticRegression layer; the activation function can be replaced by
# sigmoid or any other nonlinear function
self.hiddenLayer = HiddenLayer(rng=rng,
input=input,
n_in=n_in,
n_out=n_hidden,
activation=T.tanh)
self.W = theano.shared(name='W',
value=numpy.random.uniform(
-0.2, 0.2, (n_hidden * n_out)).astype(
theano.config.floatX))
self.b = theano.shared(name='b',
value=numpy.zeros(n_out, ).astype(
theano.config.floatX))
self.outputLayer = T.nnet.softmax(
T.dot(self.hiddenLayer.output, self.W) + self.b)
# L1 norm ; one regularization option is to enforce L1 norm to
# be small
self.L1 = (abs(self.hiddenLayer.W).sum() + abs(self.W).sum())
# square of L2 norm ; one regularization option is to enforce
# square of L2 norm to be small
self.L2_sqr = ((self.hiddenLayer.W**2).sum() + (self.W**2).sum())
# negative log likelihood of the MLP is given by the negative
# log likelihood of the output of the model, computed in the
# logistic regression layer
self.predProb = theano.function([input],
self.outputLayer,
name='f_pred_prob')
self.pred = theano.function([input],
self.outputLayer.argmax(axis=1),
name='f_pred')
# the parameters of the model are the parameters of the two layer it is
# made out of
self.params = self.hiddenLayer.params + [self.W + self.b]
