def validation_err(vl_start_index=0, vl_limit=None):
global if_load_trained_model
global validate_model
global validate_results
if if_load_trained_model == 0:
load_trained_model()
valid_set = tdtf.read_data_patch_to_ndarray(valid_dataset_route, vl_start_index, vl_limit)
print valid_set[1]
valid_set_x, valid_set_y = valid_set
validation_loss = validate_model(valid_set_x, valid_set_y)
validation_pred_y = validate_results(valid_set_x)
print validation_pred_y
label = [0, 0]
right = [0, 0]
for i in range(len(validation_pred_y)):
y_pred = validation_pred_y[i]
y = valid_set_y[i]
right[y] += 1
if y != y_pred:
if y == 0:
label[1] += 1
else:
label[0] += 1
t_num = len(validation_pred_y)
right_num = t_num - label[0] - label[1]
print "total %d, 0:1=%d:%d, right %d , wrong label to bg %d , label to 0 %d " % (t_num, right[0], right[1], right_num, label[1], label[0])
print('validation error %f %%' % \
(validation_loss * 100.))
def validation_err(vl_start_index=0, vl_limit=None):
global if_load_trained_model
global validate_model
global validate_results
if if_load_trained_model == 0:
load_trained_model()
valid_set = tdtf.read_data_patch_to_ndarray(valid_dataset_route,
vl_start_index, vl_limit)
print valid_set[1]
valid_set_x, valid_set_y = valid_set
validation_loss = validate_model(valid_set_x, valid_set_y)
validation_pred_y = validate_results(valid_set_x)
print validation_pred_y
label = [0, 0]
right = [0, 0]
for i in range(len(validation_pred_y)):
y_pred = validation_pred_y[i]
y = valid_set_y[i]
right[y] += 1
if y != y_pred:
if y == 0:
label[1] += 1
else:
label[0] += 1
t_num = len(validation_pred_y)
right_num = t_num - label[0] - label[1]
print "total %d, 0:1=%d:%d, right %d , wrong label to bg %d , label to 0 %d " % (
t_num, right[0], right[1], right_num, label[1], label[0])
print('validation error %f %%' % \
(validation_loss * 100.))
def train_by_lenet5(tr_start_index, tr_limit, vl_start_index, vl_limit, output_filename="tmp.file", learning_rate=0.13, n_epochs=5000):
global train_dataset_route
global valid_dataset_route
output_file = open(output_filename, 'w')
print train_dataset_route, type(train_dataset_route)
"""
:type learning_rate: float
:param learning_rate: learning rate used (factor for the stochastic
gradient)
:type n_epochs: int
:param n_epochs: maximal number of epochs to run the optimizer
:type nkerns: list of ints
:param nkerns: number of kernels on each layer
"""
train_set = tdtf.read_data_patch_to_ndarray(train_dataset_route, tr_start_index, tr_limit)
datasets = load_data.shared_dataset(train_set)
train_set_x, train_set_y = datasets
valid_set = tdtf.read_data_patch_to_ndarray(valid_dataset_route, vl_start_index, vl_limit)
print valid_set[1]
datasets = load_data.shared_dataset(valid_set)
valid_set_x, valid_set_y = datasets
# compute number of minibatches for training, validation and testing
n_train_batches = train_set_x.get_value(borrow=True).shape[0]
n_valid_batches = valid_set_x.get_value(borrow=True).shape[0]
n_train_batches /= batch_size
n_valid_batches /= batch_size
# allocate symbolic variables for the data
######################
# BUILD ACTUAL MODEL #
######################
print '... building the model'
# the cost we minimize during training is the NLL of the model
cost = layer3.negative_log_likelihood(y)
# create a function to compute the mistakes that are made by the model
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]})
# create a list of all model parameters to be fit by gradient descent
params = layer3.params + layer2.params + layer1.params + layer0.params
# create a list of gradients for all model parameters
grads = T.grad(cost, params)
# train_model is a function that updates the model parameters by
# SGD Since this model has many parameters, it would be tedious to
# manually create an update rule for each model parameter. We thus
# create the updates list by automatically looping over all
# (params[i],grads[i]) pairs.
updates = []
for param_i, grad_i in zip(params, grads):
updates.append((param_i, param_i - learning_rate * grad_i))
train_model = theano.function([index], [cost, layer3.errors(y), layer3.params[0][0][0]], 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]})
###############
# TRAIN MODEL #
###############
print '... training'
# early-stopping parameters
patience = 50000 # look as this many examples regardless
patience_increase = 2 # wait this much longer when a new best is
# found
improvement_threshold = 0.995 # a relative improvement of this much is
# considered significant
validation_frequency = min(n_train_batches, patience / 2)
# go through this many
# minibatche before checking the network
# on the validation set; in this case we
# check every epoch
best_params = None
best_validation_loss = numpy.inf
best_iter = 0
test_score = 0.
start_time = time.clock()
epoch = 0
done_looping = False
min_train_cost = 10000
decreasing_num = 0
last_train_err = 1
last_train_cost = 1
while (epoch < n_epochs) and (not done_looping):
epoch = epoch + 1
for minibatch_index in xrange(n_train_batches):
iter = (epoch - 1) * n_train_batches + minibatch_index
if iter % 100 == 0:
print 'training @ iter = ', iter , ' patience = ' , patience
cost_ij, train_err, par = train_model(minibatch_index)
decreasing_rate = (last_train_err - train_err) / (last_train_err) * 100.
last_train_err = train_err
if last_train_err == 0:
last_train_err += 0.0000001
c_d_rate = (last_train_cost - cost_ij) / (last_train_cost) * 100.
last_train_cost = cost_ij
print >> output_file, ('epoch %i, minibatch %i/%i, train_cost %f , train_error %.2f %%, decreasing rate %f %%, cost_decreasing rate %f %%, W00 ' % \
(epoch, minibatch_index + 1, n_train_batches,
cost_ij,
train_err* 100.
,decreasing_rate
,c_d_rate))
#print layer1.params[0:1][0][0:3]
#print layer2.params[0:1][0][0:3]
if cost_ij < min_train_cost:
decreasing_num = 0
min_train_cost = cost_ij
layer0_state = layer0.__getstate__()
layer1_state = layer1.__getstate__()
layer2_state = layer2.__getstate__()
layer3_state = layer3.__getstate__()
trained_model_list = [layer0_state, layer1_state, layer2_state, layer3_state]
trained_model_array = numpy.asarray(trained_model_list)
classifier_file = open(train_model_route, 'w')
cPickle.dump([1,2,3], classifier_file, protocol=2)
numpy.save(classifier_file, trained_model_array)
classifier_file.close()
else:
print "decreasing"
decreasing_num += 1
if decreasing_num > 100:
done_looping = True
break
if (iter + 1) % validation_frequency == 0:
# compute zero-one loss on validation set
validation_losses = [validate_model(i) for i
in xrange(n_valid_batches)]
this_validation_loss = numpy.mean(validation_losses)
print >> output_file, ('epoch %i, minibatch %i/%i, validation error %f %%' % \
(epoch, minibatch_index + 1, n_train_batches, \
this_validation_loss * 100.))
# if we got the best validation score until now
if this_validation_loss < best_validation_loss:
#improve patience if loss improvement is good enough
if this_validation_loss < best_validation_loss * \
improvement_threshold:
patience = max(patience, iter * patience_increase)
# save best validation score and iteration number
best_validation_loss = this_validation_loss
best_iter = iter
if patience <= iter:
done_looping = True
print patience , iter
break
end_time = time.clock()
print >> sys.stderr, ('The code for file ' +
os.path.split(__file__)[1] +
' ran for %.2fm' % ((end_time - start_time) / 60.))
print >> output_file, ('Optimization complete.')
print >> output_file, ('Best validation score of %f %% obtained at iteration %i,'\
'with test performance %f %%' %
(best_validation_loss * 100., best_iter + 1, test_score * 100.))
output_file.close()
def train_by_lenet5(tr_start_index,
tr_limit,
vl_start_index,
vl_limit,
output_filename="tmp.file",
learning_rate=0.13,
n_epochs=5000):
global train_dataset_route
global valid_dataset_route
output_file = open(output_filename, 'w')
print train_dataset_route, type(train_dataset_route)
"""
:type learning_rate: float
:param learning_rate: learning rate used (factor for the stochastic
gradient)
:type n_epochs: int
:param n_epochs: maximal number of epochs to run the optimizer
:type nkerns: list of ints
:param nkerns: number of kernels on each layer
"""
train_set = tdtf.read_data_patch_to_ndarray(train_dataset_route,
tr_start_index, tr_limit)
datasets = load_data.shared_dataset(train_set)
train_set_x, train_set_y = datasets
valid_set = tdtf.read_data_patch_to_ndarray(valid_dataset_route,
vl_start_index, vl_limit)
print valid_set[1]
datasets = load_data.shared_dataset(valid_set)
valid_set_x, valid_set_y = datasets
# compute number of minibatches for training, validation and testing
n_train_batches = train_set_x.get_value(borrow=True).shape[0]
n_valid_batches = valid_set_x.get_value(borrow=True).shape[0]
n_train_batches /= batch_size
n_valid_batches /= batch_size
# allocate symbolic variables for the data
######################
# BUILD ACTUAL MODEL #
######################
print '... building the model'
# the cost we minimize during training is the NLL of the model
cost = layer3.negative_log_likelihood(y)
# create a function to compute the mistakes that are made by the model
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]
})
# create a list of all model parameters to be fit by gradient descent
params = layer3.params + layer2.params + layer1.params + layer0.params
# create a list of gradients for all model parameters
grads = T.grad(cost, params)
# train_model is a function that updates the model parameters by
# SGD Since this model has many parameters, it would be tedious to
# manually create an update rule for each model parameter. We thus
# create the updates list by automatically looping over all
# (params[i],grads[i]) pairs.
updates = []
for param_i, grad_i in zip(params, grads):
updates.append((param_i, param_i - learning_rate * grad_i))
train_model = theano.function(
[index], [cost, layer3.errors(y), layer3.params[0][0][0]],
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]
})
###############
# TRAIN MODEL #
###############
print '... training'
# early-stopping parameters
patience = 50000 # look as this many examples regardless
patience_increase = 2 # wait this much longer when a new best is
# found
improvement_threshold = 0.995 # a relative improvement of this much is
# considered significant
validation_frequency = min(n_train_batches, patience / 2)
# go through this many
# minibatche before checking the network
# on the validation set; in this case we
# check every epoch
best_params = None
best_validation_loss = numpy.inf
best_iter = 0
test_score = 0.
start_time = time.clock()
epoch = 0
done_looping = False
min_train_cost = 10000
decreasing_num = 0
last_train_err = 1
last_train_cost = 1
while (epoch < n_epochs) and (not done_looping):
epoch = epoch + 1
for minibatch_index in xrange(n_train_batches):
iter = (epoch - 1) * n_train_batches + minibatch_index
if iter % 100 == 0:
print 'training @ iter = ', iter, ' patience = ', patience
cost_ij, train_err, par = train_model(minibatch_index)
decreasing_rate = (last_train_err -
train_err) / (last_train_err) * 100.
last_train_err = train_err
if last_train_err == 0:
last_train_err += 0.0000001
c_d_rate = (last_train_cost - cost_ij) / (last_train_cost) * 100.
last_train_cost = cost_ij
print >> output_file, ('epoch %i, minibatch %i/%i, train_cost %f , train_error %.2f %%, decreasing rate %f %%, cost_decreasing rate %f %%, W00 ' % \
(epoch, minibatch_index + 1, n_train_batches,
cost_ij,
train_err* 100.
,decreasing_rate
,c_d_rate))
#print layer1.params[0:1][0][0:3]
#print layer2.params[0:1][0][0:3]
if cost_ij < min_train_cost:
decreasing_num = 0
min_train_cost = cost_ij
layer0_state = layer0.__getstate__()
layer1_state = layer1.__getstate__()
layer2_state = layer2.__getstate__()
layer3_state = layer3.__getstate__()
trained_model_list = [
layer0_state, layer1_state, layer2_state, layer3_state
]
trained_model_array = numpy.asarray(trained_model_list)
classifier_file = open(train_model_route, 'w')
cPickle.dump([1, 2, 3], classifier_file, protocol=2)
numpy.save(classifier_file, trained_model_array)
classifier_file.close()
else:
print "decreasing"
decreasing_num += 1
if decreasing_num > 100:
done_looping = True
break
if (iter + 1) % validation_frequency == 0:
# compute zero-one loss on validation set
validation_losses = [
validate_model(i) for i in xrange(n_valid_batches)
]
this_validation_loss = numpy.mean(validation_losses)
print >> output_file, ('epoch %i, minibatch %i/%i, validation error %f %%' % \
(epoch, minibatch_index + 1, n_train_batches, \
this_validation_loss * 100.))
# if we got the best validation score until now
if this_validation_loss < best_validation_loss:
#improve patience if loss improvement is good enough
if this_validation_loss < best_validation_loss * \
improvement_threshold:
patience = max(patience, iter * patience_increase)
# save best validation score and iteration number
best_validation_loss = this_validation_loss
best_iter = iter
if patience <= iter:
done_looping = True
print patience, iter
break
end_time = time.clock()
print >> sys.stderr, ('The code for file ' + os.path.split(__file__)[1] +
' ran for %.2fm' % ((end_time - start_time) / 60.))
print >> output_file, ('Optimization complete.')
print >> output_file, ('Best validation score of %f %% obtained at iteration %i,'\
'with test performance %f %%' %
(best_validation_loss * 100., best_iter + 1, test_score * 100.))
output_file.close()
def sgd_optimization_mnist(tr_start_index=1, tr_limit=5000, vl_start_index=1, vl_limit=5000,
learning_rate=0.015, n_epochs=5000
, output_filename="ls.out"):
output_file = open(output_filename,'w')
# allocate symbolic variables for the data
index = T.lscalar() # index to a [mini]batch
x = T.matrix('x') # the data is presented as rasterized images
y = T.ivector('y') # the labels are presented as 1D vector of
# [int] labels
in_shape = layer0_input_shape[0] * layer0_input_shape[1]
batch_size = tr_limit
train_set = tdtf.read_data_patch_to_ndarray(train_dataset_route, tr_start_index, tr_limit)
datasets = load_data.shared_dataset(train_set)
train_set_x, train_set_y = datasets
valid_set = tdtf.read_data_patch_to_ndarray(valid_dataset_route, vl_start_index, vl_limit)
print valid_set[1]
datasets = load_data.shared_dataset(valid_set)
valid_set_x, valid_set_y = datasets
# compute number of minibatches for training, validation and testing
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
if not if_load_trained_model :
trained_model_pkl = open(train_model_route, 'r')
trained_model_state_list = cPickle.load(trained_model_pkl)
trained_model_state_array = numpy.load(trained_model_pkl)
classifier_state = trained_model_state_array[0]
classifier = LogisticRegression(input=x, n_in=in_shape, n_out=layer0_output_shape
, W=classifier_state[0], b=classifier_state[1])
else:
######################
# BUILD ACTUAL MODEL #
######################
#print '... building the model'
# construct the logistic regression class
rng = numpy.random.RandomState(23555)
W_bound=1
tmp_W = theano.shared(numpy.asarray(
rng.uniform(low=0, high=W_bound, size=(in_shape, layer0_output_shape)), dtype=theano.config.floatX),
borrow=True)
classifier = LogisticRegression(input=x, n_in=in_shape, n_out=layer0_output_shape)
#,W=tmp_W)
# the cost we minimize during training is the negative log likelihood of
# the model in symbolic format
cost = classifier.negative_log_likelihood(y)
# compiling a Theano function that computes the mistakes that are made by
# the model on a minibatch
validate_model = theano.function(inputs=[index],
outputs=classifier.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]})
# compute the gradient of cost with respect to theta = (W,b)
g_W = T.grad(cost=cost, wrt=classifier.W)
g_b = T.grad(cost=cost, wrt=classifier.b)
# specify how to update the parameters of the model as a list of
# (variable, update expression) pairs.
updates = [(classifier.W, classifier.W - learning_rate * g_W),
(classifier.b, classifier.b - learning_rate * g_b)]
# compiling a Theano function `train_model` that returns the cost, but in
# the same time updates the parameter of the model based on the rules
# defined in `updates`
train_model = theano.function(inputs=[index], \
outputs=[cost, classifier.errors(y)], \
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]})
###############
# TRAIN MODEL #
###############
#print '... training the model'
# early-stopping parameters
patience = 50000 # look as this many examples regardless
patience_increase = 2 # wait this much longer when a new best is
# found
improvement_threshold = 0.995 # a relative improvement of this much is
# considered significant
validation_frequency = min(n_train_batches, patience / 2)
# go through this many
# minibatche before checking the network
# on the validation set; in this case we
# check every epoch
best_params = None
best_validation_loss = numpy.inf
test_score = 0.
start_time = time.clock()
best_train_loss = numpy.inf
done_looping = False
epoch = 0
last_train_err = 1
last_train_cost = 1
while (epoch < n_epochs) and (not done_looping):
epoch = epoch + 1
for minibatch_index in xrange(n_train_batches):
minibatch_avg_cost, train_err = train_model(minibatch_index)
decreasing_rate = (last_train_err - train_err) / (last_train_err) * 100.
last_train_err = train_err
c_d_rate = (last_train_cost - minibatch_avg_cost) / (last_train_cost) * 100.
last_train_cost = minibatch_avg_cost
print >> output_file, ('epoch %i, minibatch %i/%i, train_cost %f , train_error %.2f %%, decreasing rate %f %%, cost_decreasing rate %f %%' % \
(epoch, minibatch_index + 1, n_train_batches,
minibatch_avg_cost,
train_err* 100.
,decreasing_rate
,c_d_rate))
if best_train_loss > train_err:
best_train_loss = train_err
# iteration number
iter = (epoch - 1) * n_train_batches + minibatch_index
if (iter + 1) % validation_frequency == 0:
# compute zero-one loss on validation set
validation_losses = [validate_model(i)
for i in xrange(n_valid_batches)]
this_validation_loss = numpy.mean(validation_losses)
print >> output_file, ('epoch %i, minibatch %i/%i, validation error %f %%' % \
(epoch, minibatch_index + 1, n_train_batches,
this_validation_loss * 100.))
# if we got the best validation score until now
if this_validation_loss < best_validation_loss:
#improve patience if loss improvement is good enough
if this_validation_loss < best_validation_loss * \
improvement_threshold:
patience = max(patience, iter * patience_increase)
best_validation_loss = this_validation_loss
# load trained_model to
'''
layer_state = classifier.__getstate__()
trained_model_list = [layer_state]
trained_model_array = numpy.asarray(trained_model_list)
classifier_file = open(train_model_route, 'w')
cPickle.dump([1,2,3], classifier_file, protocol=2)
numpy.save(classifier_file, trained_model_array)
classifier_file.close()
'''
'''
test_losses = [test_model(i)
for i in xrange(n_test_batches)]
test_score = numpy.mean(test_losses)
test_res = [test_results(i)
for i in xrange(n_test_batches)]
print((' epoch %i, minibatch %i/%i, test error of best'
' model %f %%') %
(epoch, minibatch_index + 1, n_train_batches,
test_score * 100.))
'''
if patience <= iter:
done_looping = True
break
end_time = time.clock()
print >> output_file, (('Optimization complete with best validation score of %f %%,'
'with test performance %f %%'
'with best train_performance %f %%') %
(best_validation_loss * 100., test_score * 100., best_train_loss * 100.))
print >> output_file, 'The code run for %d epochs, with %f epochs/sec' % (
epoch, 1. * epoch / (end_time - start_time))
print >> sys.stderr, ('The code for file ' +
os.path.split(__file__)[1] +
' ran for %.1fs' % ((end_time - start_time)))
output_file.close()
