def recognizing(imageData, loadedFile='data.pkl', kernels=[8, 10]):
"""
Load the data and build the CNN which have been trained.
Then print the result.
"""
dataBase, manLabel = loading(imageData)
prm0, prm1, prm2, prm3 = load_save_data.loadData(loadedFile)
m = T.matrix('m')
inputData = m.reshape((dataBase.shape[0], 1, 57, 47))
l0 = CNN.ConvPoolLayer(inputData=inputData,
filterShape=(kernels[0], 1, 5, 5),
imageShape=(dataBase.shape[0], 1, 57, 47),
poolsize=(2, 2),
weights=prm0[0],
biases=prm0[1])
l1 = CNN.ConvPoolLayer(inputData=l0.output,
filterShape=(kernels[1], kernels[0], 5, 5),
imageShape=(dataBase.shape[0], kernels[0], 26, 21),
poolsize=(2, 2),
weights=prm1[0],
biases=prm1[1])
l2 = CNN.HiddenLayer(inputData=l1.output.reshape(
(dataBase.shape[0], kernels[1] * 11 * 8)),
numIn=kernels[1] * 11 * 8,
numOut=2000,
weights=prm2[0],
biases=prm2[1],
activation=T.tanh)
l3 = CNN.LogisticRegression(inputData=l2.output,
numIn=2000,
numOut=40,
weights=prm3[0],
biases=prm3[1])
func = theano.function([m], l3.yRslt)
result = func(dataBase)
"""for ind in range(dataBase.shape[0]):
if result[ind] != manLabel[ind]:
print ('%i is mis-predicted as %i' % (manLabel[ind], result[ind]))
img1 = dataBase[ind].reshape((57, 47))
plt.subplot(121)
plt.imshow(img1)
pylab.gray()
img2 = dataBase[result[ind] * 10].reshape((57, 47))
plt.subplot(122)
plt.imshow(img2)
pylab.gray()
pylab.show()
"""
#first term is age and second term is gender
if int(gender_matrix[result[0]]) == 0:
gender = "male"
else:
gender = "female"
print old_matrix[result[0]], ' ', gender
def train(self):
# self.trainBuntton["state"] = 'disabled'
#sizes, epochs, eta, update_queue, validation_threshold, error_threshold
trainingPercent = int(self.trainingPercent.get())
epochs = int(self.epochs.get())
et = float(self.error_threshold.get())
vt = int(self.validation_threshold.get())
poolinRule = self.poolingRule.get()
mini_batch_size = int(self.mini_batch_size.get())
try:
import pickle
weights = pickle.load(open(self.address.get(), "rb"))
except:
weights = None
if not self.plotter.is_alive():
self.plotter.start()
if self.cnn is not None:
if self.cnn.exitcode != 0:
print 'Another learner is working.'
return
self.cnn.terminate()
# # self.cnn = CNN.Network(sizes, epochs, eta, mu, self.updatesQueue, vt, et, widrow, bs, activationFunc, weights)
# self.cnn = CNN.Network([CNN.ConvPoolLayer(image_shape=(mini_batch_size, 1, 28, 28), filter_shape=(20, 1, 5, 5), poolsize=(2, 2)),
# CNN.FullyConnectedLayer(n_in=20*12*12, n_out=100),
# CNN.SoftmaxLayer(n_in=100, n_out=10)], mini_batch_size, self.updatesQueue)
self.cnn = CNN.Network([
CNN.ConvPoolLayer(image_shape=(mini_batch_size, 1, 28, 28),
filter_shape=(20, 1, 5, 5),
poolsize=(2, 2)),
CNN.ConvPoolLayer(image_shape=(mini_batch_size, 20, 12, 12),
filter_shape=(30, 20, 5, 5),
poolsize=(2, 2)),
CNN.ConvPoolLayer(image_shape=(mini_batch_size, 30, 4, 4),
filter_shape=(40, 30, 2, 2),
poolsize=(3, 3)),
CNN.FullyConnectedLayer(n_in=40 * 1 * 1, n_out=100),
CNN.SoftmaxLayer(n_in=100, n_out=10)
], mini_batch_size, self.updatesQueue)
self.cnn.start()
def DataTraining(learningRate=0.05,
epochs=200,
imageData='olivettifaces.gif',
kernels=[8, 10],
batchsize=40):
"""
This is the training function, kind of like relaxation step.
First, give a number(here I choose 23333) to generate random number and load the data.
Then construct the convolution neural network(It's a hard part, by debugging and debugging the parameters)
After constructing, we will build three models, and use SGD in training model.
Then training.
"""
rng = numpy.random.RandomState(23333)
imageData = load_save_data.loadImageData(imageData)
trainX, trainY = imageData[0]
validX, validY = imageData[1]
testX, testY = imageData[2]
numTrainBatches = trainX.get_value(borrow=True).shape[0] / batchsize
numValidBatches = validX.get_value(borrow=True).shape[0] / batchsize
numTestBatches = testX.get_value(borrow=True).shape[0] / batchsize
index = T.lscalar()
x = T.matrix('x')
y = T.ivector('y')
inputData = x.reshape((batchsize, 1, 57, 47))
l0 = CNN.ConvPoolLayer(inputData=inputData,
filterShape=(kernels[0], 1, 5, 5),
imageShape=(batchsize, 1, 57, 47),
poolsize=(2, 2),
rng=rng)
# 26 = (57 - 5) / 2, 21 = (47 - 5) / 2
l1 = CNN.ConvPoolLayer(inputData=l0.output,
filterShape=(kernels[1], kernels[0], 5, 5),
imageShape=(batchsize, kernels[0], 26, 21),
poolsize=(2, 2),
rng=rng)
# 11 = (26 - 5 + 1) / 2, 8 = (21 - 5) / 2
l2 = CNN.HiddenLayer(inputData=l1.output.reshape(
(batchsize, kernels[1] * 11 * 8)),
numIn=kernels[1] * 11 * 8,
numOut=2000,
activation=T.tanh,
rng=rng)
l3 = CNN.LogisticRegression(inputData=l2.output, numIn=2000, numOut=40)
cost = l3.logLikelihood(y)
"""
================================================================================================================================
"""
#Here I will give a brief introduction to how to build the model.
#Using the given index, we can get the corresponding x and y(what given does).
#Because l3.error() is applied, l2 - l1 - l0 were also applied. i.e. CNN is applied
#In this way, model was built.
testModel = theano.function(
[index],
l3.error(y),
givens={
x: testX[index * batchsize:(index + 1) * batchsize],
y: testY[index * batchsize:(index + 1) * batchsize]
})
validationModel = theano.function(
[index],
l3.error(y),
givens={
x: validX[index * batchsize:(index + 1) * batchsize],
y: validY[index * batchsize:(index + 1) * batchsize]
})
#SGD(stochastic gradient decent)
parameters = l0.parameters + l1.parameters + l2.parameters + l3.parameters
grads = T.grad(cost, parameters)
updates = [(p, p - learningRate * g) for p, g in zip(parameters, grads)]
trainingModel = theano.function(
[index],
cost,
updates=updates,
givens={
x: trainX[index * batchsize:(index + 1) * batchsize],
y: trainY[index * batchsize:(index + 1) * batchsize]
})
"""
=================================================================================================================================
"""
print 'training'
ptc = 800
ptcInc = 2
threshold = 0.99
#To ensure that always validate on validation data
validationFrequency = min(numTrainBatches, ptc / 2)
bstValidationL = numpy.inf
bstIter = 0
start = time.clock()
epoch = 0
done = False
#Now it's time to train, really tricky.
#Kind of like SVM, a epoch will go through all data set. For each while loop, we need to train
#each batch by using the for loop. If the current loss is less than the best validation loss
#we will update them, and if the current loss is less than the threshold x best validation loss
#we will update the patience.
while (epoch < epochs) and (not done):
epoch += 1
for ind in xrange(numTrainBatches):
iter = (epoch - 1) * numTrainBatches + ind
c = trainingModel(ind)
if (iter + 1) % validationFrequency == 0:
validationL = [
validationModel(i) for i in xrange(numValidBatches)
]
currentL = numpy.mean(validationL)
#print('epoch %i, validation error %f %%' % (epoch, currentL * 100.0))
if currentL < bstValidationL:
if currentL < bstValidationL * threshold:
ptc = max(ptc, iter * ptcInc)
bstValidationL = currentL
bstIter = iter
load_save_data.saveData(l0.parameters, l1.parameters,
l2.parameters, l3.parameters)
testL = [testModel(i) for i in xrange(numTestBatches)]
tstscore = numpy.mean(testL)
print(('epoch %i, model error: %f %%') %
(epoch, tstscore * 100.0))
if ptc <= iter:
done = True
break
end = time.clock()
print('Optimization: ')
print('validation score: %f %% iter: %i' %
(bstValidationL * 100.0, bstIter + 1))
print('time: %.2f ' % (end - start))
def cnn_mnist(dataset):
learning_rate = 0.1
n_epochs = 200
batch_size = 500
#nkerns : number of kernels on each layer
nkerns = [20, 10]
train_set_x, train_set_y = dataset[0]
valid_set_x, valid_set_y = dataset[1]
test_set_x, test_set_y = dataset[2]
n_train_batches = train_set_x.get_value(borrow=True).shape[0] // batch_size
n_valid_batches = valid_set_x.get_value(borrow=True).shape[0] // batch_size
n_test_batches = test_set_x.get_value(borrow=True).shape[0] // batch_size
index = T.lscalar()
x = T.matrix('x')
y = T.ivector('y')
print('Building the model...')
layer0_input = x.reshape((batch_size, 1, 28, 28))
layer0 = CNN.ConvPoolLayer(rng,
input=layer0_input,
image_shape=(batch_size, 1, 28, 28),
filter_shape=(nkerns[0], 1, 5, 5),
poolsize=(2, 2))
layer1 = CNN.ConvPoolLayer(rng,
input=layer0.output,
image_shape=(batch_size, nkerns[0], 12, 12),
filter_shape=(nkerns[1], nkerns[0], 5, 5),
poolsize=(2, 2))
layer2_input = layer1.output.flatten(2)
layer2 = MultiLayerPerceptron.HiddenLayer(rng,
input=layer2_input,
n_in=nkerns[1] * 4 * 4,
n_out=80,
activation=T.tanh)
layer3 = LogisticRegression.LogisticRegression(input=layer2.output,
n_in=80,
n_out=10)
cost = layer3.NLL(y)
test_model = theano.function(
[index],
layer3.errors(y),
givens={
x: test_set_x[index * batch_size:(index + 1) * batch_size],
y: test_set_y[index * batch_size:(index + 1) * batch_size]
})
print('1. Test built.')
validate_model = theano.function(
[index],
layer3.errors(y),
givens={
x: valid_set_x[index * batch_size:(index + 1) * batch_size],
y: valid_set_y[index * batch_size:(index + 1) * batch_size]
})
print('2. Valid built.')
params = layer3.params + layer2.params + layer1.params + layer0.params
grads = T.grad(cost, params)
updates = [(param_i, param_i - learning_rate * grad_i)
for param_i, grad_i in zip(params, grads)]
print('3. Derivative calculated.')
train_model = theano.function(
[index],
cost,
updates=updates,
givens={
x: train_set_x[index * batch_size:(index + 1) * batch_size],
y: train_set_y[index * batch_size:(index + 1) * batch_size]
})
print('4. Train built.')
print('Train model...')
patience = 10000
patience_increase = 2
improvement_threshold = 0.995
validation_frequency = min(n_train_batches, patience // 2)
best_validation_loss = numpy.inf
best_iter = 0
test_score = 0
start_time = timeit.default_timer()
epoch = 0
done_looping = False
while (epoch < n_epochs) and (not done_looping):
epoch = epoch + 1
for minibatch_index in range(n_train_batches):
iter = (epoch - 1) * n_train_batches + minibatch_index
# if iter % 100 == 0:
# print('training @ iter = ',iter)
cost_ij = train_model(minibatch_index)
if (iter + 1) % validation_frequency == 0:
validation_losses = [
validate_model(i) for i in range(n_valid_batches)
]
this_validation_loss = numpy.mean(validation_losses)
print('epoch %i, minibatch %i/%i, validation error %.2f %%' %
(epoch, minibatch_index + 1, n_train_batches,
this_validation_loss * 100))
if this_validation_loss < best_validation_loss:
if this_validation_loss < best_validation_loss * improvement_threshold:
patience = max(patience, iter * patience_increase)
best_validation_loss = this_validation_loss
best_iter = iter
test_losses = [
test_model(i) for i in range(n_test_batches)
]
test_score = numpy.mean(test_losses)
print(
'\tepoch %i, minibatch %i/%i, test error of best model %.2f %%'
% (epoch, minibatch_index + 1, n_train_batches,
test_score * 100))
if patience <= iter:
done_looping = True
break
end_time = timeit.default_timer()
print('Optimization complete')
print(
'Best validation score of %f %% obtained at iteration %i, with test performance %.2f %%'
% (best_validation_loss * 100, best_iter + 1, test_score * 100))