def get_layer_trainer_sgd_autoencoder(layer,
trainset,
batch_size=10,
learning_rate=0.1,
max_epochs=100,
name=''):
# configs on sgd
train_algo = SGD(
learning_rate=learning_rate,
# learning_rule = AdaDelta(),
learning_rule=Momentum(init_momentum=0.5),
cost=MeanSquaredReconstructionError(),
batch_size=batch_size,
monitoring_dataset=trainset,
termination_criterion=EpochCounter(max_epochs=max_epochs),
update_callbacks=None)
log_callback = LoggingCallback(name)
return Train(model=layer,
algorithm=train_algo,
extensions=[
log_callback,
OneOverEpoch(start=1, half_life=5),
MomentumAdjustor(final_momentum=0.7,
start=10,
saturate=100)
],
dataset=trainset)
def get_ae_pretrainer(layer, data, batch_size, epochs=30):
init_lr = 0.05
train_algo = SGD(
batch_size=batch_size,
learning_rate=init_lr,
learning_rule=Momentum(init_momentum=0.5),
monitoring_batches=batch_size,
monitoring_dataset=data,
# for ContractiveAutoencoder:
# cost=cost.SumOfCosts(costs=[[1., MeanSquaredReconstructionError()],
# [0.5, cost.MethodCost(method='contraction_penalty')]]),
# for HigherOrderContractiveAutoencoder:
# cost=cost.SumOfCosts(costs=[[1., MeanSquaredReconstructionError()],
# [0.5, cost.MethodCost(method='contraction_penalty')],
# [0.5, cost.MethodCost(method='higher_order_penalty')]]),
# for DenoisingAutoencoder:
cost=MeanSquaredReconstructionError(),
termination_criterion=EpochCounter(epochs))
return Train(model=layer,
algorithm=train_algo,
dataset=data,
extensions=[
MomentumAdjustor(final_momentum=0.9, start=0,
saturate=25),
LinearDecayOverEpoch(start=1,
saturate=25,
decay_factor=.02)
])
def __init__(self,
runner,
model_params,
resume=False,
resume_data=None,
s3_data=None,
**kwargs):
dataset = create_dense_design_matrix(x=runner.dp.train_set_x)
if resume:
model, model_params = self.resume_model(model_params, resume_data)
else:
model = self.new_model(model_params, dataset=dataset)
termination_criterion = MaxEpochNumber(model_params['maxnum_iter'])
algorithm = SGD(learning_rate=model_params['learning_rate']['init'],
monitoring_dataset=dataset,
cost=MeanSquaredReconstructionError(),
termination_criterion=termination_criterion,
batch_size=model_params['batch_size'])
ext = AutoEncoderStatReporter(runner,
resume=resume,
resume_data=resume_data,
save_freq=model_params['save_freq'])
self.train_obj = Train(dataset=dataset,
model=model,
algorithm=algorithm,
extensions=[ext])
def train_model():
global ninput, noutput
simdata = SimulationData(
sim_path="../../javaDataCenter/generarDadesV1/CA_SDN_topo1/")
simdata.load_data()
simdata.preprocessor()
dataset = simdata.get_matrix()
structure = get_structure()
layers = []
for pair in structure:
layers.append(get_autoencoder(pair))
model = DeepComposedAutoencoder(layers)
training_alg = SGD(learning_rate=1e-3,
cost=MeanSquaredReconstructionError(),
batch_size=1296,
monitoring_dataset=dataset,
termination_criterion=EpochCounter(max_epochs=50))
extensions = [MonitorBasedLRAdjuster()]
experiment = Train(dataset=dataset,
model=model,
algorithm=training_alg,
save_path='training2.pkl',
save_freq=10,
allow_overwrite=True,
extensions=extensions)
experiment.main_loop()
def set_training_criteria(self, learning_rate=0.05, cost=MeanSquaredReconstructionError(), batch_size=10, max_epochs=10): self.training_alg = SGD(learning_rate = learning_rate, cost = cost, batch_size = batch_size, monitoring_dataset = self.datasets, termination_criterion = EpochCounter(max_epochs))
def create_layer_one(self):
which_set = "train"
one_hot = True
start = 0
# Creating 5 random patch layers based on 8,000 samples (Saturation point where the objective no longer improves.
stop = 800
# GridPatchCIFAR10 Randomly selects 5 16x16 patches from each image, and we do this 5 times. This helps increase training time and captures more information. Similar to how the neurons in the eye are attached to a specific region in the image.
dataset = GridPatchCIFAR10(which_set=which_set, one_hot=one_hot, start=start, stop=stop)
# Denoising autoencoder model hyper-parameters
nvis = 768
nhid = 512
irange = 0.05
corruption_lvl = 0.2
corruptor = BinomialCorruptor(corruption_level=corruption_lvl)
activation_encoder = "tanh"
# Linear activation
activation_decoder = None
# Creating the denoising autoencoder
model = DenoisingAutoencoder(nvis=nvis, nhid=nhid, irange=irange, corruptor=corruptor, act_enc=activation_encoder, act_dec=activation_decoder)
# Parameters for SGD learning algorithm instantiated below
learning_rate = 0.001
batch_size = 100
monitoring_batches = 5
monitoring_dataset = dataset
cost = MeanSquaredReconstructionError()
max_epochs = 10
termination_criterion = EpochCounter(max_epochs=max_epochs)
# SGD Learning algorithm
algorithm = SGD(learning_rate=learning_rate, batch_size=batch_size, monitoring_batches=monitoring_batches, monitoring_dataset=dataset, cost=cost, termination_criterion=termination_criterion)
processes = []
for i in range(0,5):
print "Training DAE Sub-Layer: ", i
save_path = self.save_path+str(i)+".pkl"
save_freq = 1
train = Train(dataset=dataset,model=model,algorithm=algorithm, save_path=save_path, save_freq=save_freq)
p = Process(target=train.main_loop, args=())
p.start()
processes.append(p)
for process in processes:
process.join()
def get_layer_trainer(layer):
# configs on sgd
config = {'learning_rate': 0.1,
'cost' : MeanSquaredReconstructionError(),
'batch_size': 10,
'monitoring_batches': 10,
'monitoring_dataset': ToyDataset(),
'termination_criterion': EpochCounter(max_epochs=100),
'update_callbacks': None
}
train_algo = UnsupervisedExhaustiveSGD(**config)
model = layer
callbacks = None
return LayerTrainer(model, train_algo, callbacks)
def get_layer_trainer_sgd_autoencoder(layer, trainset):
# configs on sgd
train_algo = SGD(
learning_rate=0.1,
cost=MeanSquaredReconstructionError(),
batch_size=10,
monitoring_batches=10,
monitoring_dataset=trainset,
termination_criterion=EpochCounter(max_epochs=MAX_EPOCHS_UNSUPERVISED),
update_callbacks=None)
model = layer
extensions = None
return Train(model=model,
algorithm=train_algo,
extensions=extensions,
dataset=trainset)
def get_ae_trainer(model, dataset, save_path, epochs=5):
"""
An Autoencoder (AE) trainer
"""
config = {
'learning_rate': 1e-2,
'train_iteration_mode': 'shuffled_sequential',
'batch_size': 250,
#'batches_per_iter' : 2000,
'learning_rule': RMSProp(),
'monitoring_dataset': dataset,
'cost': MeanSquaredReconstructionError(),
'termination_criterion': EpochCounter(max_epochs=epochs),
}
return Train(model=model,
algorithm=SGD(**config),
dataset=dataset,
save_path=save_path,
save_freq=1) #, extensions=extensions)
def main():
# Only the trainset is processed by this function.
print 'getting preprocessed data to train model'
pp_trainset, testset = get_processed_dataset()
# remember to change here when changing datasets
print 'loading unprocessed data for input displays'
trainset = cifar10.CIFAR10(which_set="train")
dmat = trainset.get_design_matrix()
nvis = dmat.shape[1]
model = DenoisingAutoencoder(
corruptor=BinomialCorruptor(corruption_level=0.5),
nhid=nhid,
nvis=nvis,
act_enc='sigmoid',
act_dec='sigmoid',
irange=.01)
algorithm = SGD(
learning_rate=0.1,
cost=MeanSquaredReconstructionError(),
batch_size=1000,
monitoring_batches=10,
monitoring_dataset=pp_trainset,
termination_criterion=EpochCounter(max_epochs=MAX_EPOCHS_UNSUPERVISED),
update_callbacks=None)
extensions = None
trainer = Train(model=model,
algorithm=algorithm,
save_path='testrun.pkl',
save_freq=1,
extensions=extensions,
dataset=pp_trainset)
trainer.main_loop()
def get_layer_trainer_sgd_autoencoder(layer,
trainset,
batch_size=10,
monitoring_batches=5,
learning_rate=0.1,
max_epochs=100,
name=''):
# configs on sgd
train_algo = SGD(
learning_rate=learning_rate, # 0.1,
# learning_rule = AdaDelta(),
cost=MeanSquaredReconstructionError(),
batch_size=10,
monitoring_batches=5,
monitoring_dataset=trainset,
termination_criterion=EpochCounter(max_epochs=max_epochs),
update_callbacks=None)
log_callback = LoggingCallback(name)
return Train(model=layer,
algorithm=train_algo,
extensions=[log_callback],
dataset=trainset)
def main():
# the data isn't going to be fully processed, so we may have to
# do some stuff to the testset still to make it work.
trainset, testset = get_processed_dataset()
# Creating the patch-pairs:
design_matrix = trainset.get_design_matrix()
processed_patch_size = design_matrix.shape[1]
num_images = train_size
examples_per_image = patches_per_image * (patches_per_image - 1)
num_examples = examples_per_image * num_images
stamps = trainset.stamps
max_stamp = input_width - patch_width
d_size = (2 * max_stamp + 1)**input_dim
patch_pairs = np.zeros((num_examples, 2 * processed_patch_size))
distances = np.zeros((num_examples, input_dim))
distances_onehot = np.zeros((num_examples, d_size))
examples = np.zeros((num_examples, 2 * processed_patch_size + d_size))
nvis = 2 * processed_patch_size + d_size
def flatten_encoding(encoding, max_stamp):
dims = len(encoding)
flat_encoding = 0
for i in xrange(dims - 1):
flat_encoding += encoding[i]
flat_encoding *= max_stamp
flat_encoding += encoding[-1]
# Can be done without (or with less) for loops?
print 'begin for loop'
for i in xrange(num_images):
if (i % 1000 == 0):
print i, '-th outer loop...'
for j in xrange(patches_per_image):
patch1_num = i * patches_per_image + j
patch1_pos = stamps[patch1_num, :]
for k in xrange(patches_per_image):
example_num = i * examples_per_image + j * (patches_per_image -
1) + k
if (k > j):
example_num -= 1
if (k != j):
patch2_num = i * patches_per_image + k
patch2_pos = stamps[patch2_num, :]
distance = patch1_pos - patch2_pos
distances[example_num] = distance
distance_encoding = distance + max_stamp
distance_encoding = flatten_encoding(
distance_encoding, max_stamp)
distances_onehot[example_num, distance_encoding] = 1
p1 = design_matrix[patch1_num]
p2 = design_matrix[patch2_num]
patch_pairs[example_num] = np.hstack((p1, p2))
examples[example_num] = np.hstack(
(patch_pairs[example_num],
distances_onehot[example_num]))
print 'end for loop'
trainset.set_design_matrix(examples)
model = DenoisingAutoencoder(
corruptor=BinomialCorruptor(corruption_level=0.5),
nhid=nhid,
nvis=nvis,
act_enc='sigmoid',
act_dec='sigmoid',
irange=.01)
algorithm = SGD(
learning_rate=0.1,
cost=MeanSquaredReconstructionError(),
batch_size=100,
monitoring_batches=10,
monitoring_dataset=trainset,
termination_criterion=EpochCounter(max_epochs=MAX_EPOCHS_UNSUPERVISED),
update_callbacks=None)
extensions = None
trainer = Train(model=model,
algorithm=algorithm,
save_path='run.pkl',
save_freq=1,
extensions=extensions,
dataset=trainset)
trainer.main_loop()
def __init__(self, cost=MeanSquaredReconstructionError()):
self.cost = cost
self.train_iteration_mode = 'even_sequential'
self.monitor_iteration_mode = 'even_sequential'
def main_train(work_dir="../results/avicenna/",
corruption_level=0.3,
nvis=75,
nhid=600,
tied_weights=True,
act_enc="sigmoid",
act_dec=None,
max_epochs=2,
learning_rate=0.001,
batch_size=20,
monitoring_batches=5,
save_freq=1,
n_components_trans_pca=7):
conf = {
'corruption_level': corruption_level,
'nvis': nvis,
'nhid': nhid,
'tied_weights': tied_weights,
'act_enc': act_enc,
'act_dec': act_dec,
'max_epochs': max_epochs,
'learning_rate': learning_rate,
'batch_size': batch_size,
'monitoring_batches': monitoring_batches,
'save_freq': save_freq,
'n_components_trans_pca': n_components_trans_pca
}
start = time.clock()
############### TRAIN THE DAE
train_file = work_dir + "train_pca" + str(conf['nvis']) + ".npy"
save_path = work_dir + "train_pca" + str(conf['nvis']) + "_dae" + str(
conf['nhid']) + "_model.pkl"
trainset = NpyDataset(file=train_file)
trainset.yaml_src = 'script'
corruptor = BinomialCorruptor(corruption_level=conf['corruption_level'])
dae = DenoisingAutoencoder(nvis=conf['nvis'],
nhid=conf['nhid'],
tied_weights=conf['tied_weights'],
corruptor=corruptor,
act_enc=conf['act_enc'],
act_dec=conf['act_dec'])
cost = MeanSquaredReconstructionError()
termination_criterion = EpochCounter(max_epochs=conf['max_epochs'])
algorithm = UnsupervisedExhaustiveSGD(
learning_rate=conf['learning_rate'],
batch_size=conf['batch_size'],
monitoring_batches=conf['monitoring_batches'],
monitoring_dataset=trainset,
cost=cost,
termination_criterion=termination_criterion)
train_obj = Train(dataset=trainset,
model=dae,
algorithm=algorithm,
save_freq=conf['save_freq'],
save_path=save_path)
train_obj.main_loop()
############### APPLY THE MODEL ON THE TRAIN DATASET
print("Applying the model on the train dataset...")
model = load(save_path)
save_train_path = work_dir + "train_pca" + str(
conf['nvis']) + "_dae" + str(conf['nhid']) + ".npy"
dump_obj = FeatureDump(encoder=model,
dataset=trainset,
path=save_train_path)
dump_obj.main_loop()
############### APPLY THE MODEL ON THE VALID DATASET
print("Applying the model on the valid dataset...")
valid_file = work_dir + "valid_pca" + str(conf['nvis']) + ".npy"
validset = NpyDataset(file=valid_file)
validset.yaml_src = 'script'
save_valid_path = work_dir + "valid_pca" + str(
conf['nvis']) + "_dae" + str(conf['nhid']) + ".npy"
dump_obj = FeatureDump(encoder=model,
dataset=validset,
path=save_valid_path)
dump_obj.main_loop()
############### APPLY THE MODEL ON THE TEST DATASET
print("Applying the model on the test dataset...")
test_file = work_dir + "test_pca" + str(conf['nvis']) + ".npy"
testset = NpyDataset(file=test_file)
testset.yaml_src = 'script'
save_test_path = work_dir + "test_pca" + str(conf['nvis']) + "_dae" + str(
conf['nhid']) + ".npy"
dump_obj = FeatureDump(encoder=model, dataset=testset, path=save_test_path)
dump_obj.main_loop()
############### COMPUTE THE ALC SCORE ON VALIDATION SET
valid_data = ift6266h12.load_npy(save_valid_path)
label_data = ift6266h12.load_npy(
'/data/lisa/data/UTLC/numpy_data/avicenna_valid_y.npy')
alc_1 = score(valid_data, label_data)
############### APPLY THE TRANSDUCTIVE PCA
test_data = ift6266h12.load_npy(save_test_path)
trans_pca = PCA(n_components=conf['n_components_trans_pca'])
final_valid = trans_pca.fit_transform(valid_data)
final_test = trans_pca.fit_transform(test_data)
save_valid_path = work_dir + "valid_pca" + str(
conf['nvis']) + "_dae" + str(conf['nhid']) + "_tpca" + str(
conf['n_components_trans_pca']) + ".npy"
save_test_path = work_dir + "test_pca" + str(conf['nvis']) + "_dae" + str(
conf['nhid']) + "_tpca" + str(conf['n_components_trans_pca']) + ".npy"
np.save(save_valid_path, final_valid)
np.save(save_test_path, final_test)
############### COMPUTE THE NEW ALC SCORE ON VALIDATION SET
alc_2 = score(final_valid, label_data)
############### OUTPUT AND RETURN THE RESULTS
timeSpent = ((time.clock() - start) / 60.)
print 'FINAL RESULTS (PCA-' + str(conf['nvis']) + ' DAE-' + str(conf['nhid']) + ' TransPCA-' + str(conf['n_components_trans_pca']) + ') ALC after DAE: ', alc_1, ' FINAL ALC: ', alc_2, \
' Computed in %5.2f min' % (timeSpent)
return timeSpent, alc_1, alc_2
def main():
# Only the trainset is processed by this function.
print 'getting preprocessed data for training model'
pp_trainset, testset = get_processed_dataset()
# remember to change here when changing datasets
print 'loading unprocessed data for input displays'
trainset = cifar10.CIFAR10(which_set="train")
dmat = pp_trainset.get_design_matrix()
nvis = dmat.shape[1]
model = DenoisingAutoencoder(
corruptor=BinomialCorruptor(corruption_level=0.3),
nhid=nhid,
nvis=nvis,
act_enc='sigmoid',
act_dec='sigmoid',
irange=.01)
algorithm = SGD(
learning_rate=learning_rate,
cost=MeanSquaredReconstructionError(),
batch_size=100,
monitoring_batches=10,
monitoring_dataset=pp_trainset,
termination_criterion=EpochCounter(max_epochs=MAX_EPOCHS_UNSUPERVISED),
update_callbacks=None)
extensions = None
trainer = Train(model=model,
algorithm=algorithm,
save_path='run.pkl',
save_freq=1,
extensions=extensions,
dataset=pp_trainset)
trainer.main_loop()
####################
# Plot and Save:
# choose random patch-pairs to plot
stamps = pp_trainset.stamps
num_examples = stamps.shape[0]
to_plot = np.random.randint(0, num_examples, num2plot)
# use to_plot indices to extract data
stamps_data = stamps[to_plot]
image_numbers = stamps[to_plot, 0].astype(int)
X = trainset.X
images_data = trainset.get_topological_view(X[image_numbers])
p1x = stamps_data[:, 1]
p1y = stamps_data[:, 2]
p2x = stamps_data[:, 3]
p2y = stamps_data[:, 4]
# For input ppd's, once we've identified the patches, we just outline them and draw an arrow for d
# This might mess with original trainset (I dunno), in which case, we should make a copy
add_outlines(images_data, p1x, p1y, patch_width)
add_outlines(images_data, p2x, p2y, patch_width)
##################################################
# translating outputs back into things we can plot
dataset = pp_trainset
Xout = dataset.X.astype('float32')
max_stamp = input_width - patch_width
d_size = (2 * max_stamp + 1)**input_dim
# displacement
d_enc = Xout[:, -d_size:]
d_out_flat = np.argmax(d_enc, axis=1)
d_shape = [2 * max_stamp + 1, 2 * max_stamp + 1] # assumed 2D
d_out = flat_to_2D(d_out_flat, d_shape)
d_out[to_plot, ]
# patches
vc = dataset.view_converter
p_enc = Xout[:, :len(Xout.T) - d_size]
p_size = p_enc.shape[1] / 2
p1_enc = p_enc[:, :p_size]
p2_enc = p_enc[:, p_size:]
p1_enc = vc.design_mat_to_topo_view(p1_enc)
p2_enc = vc.design_mat_to_topo_view(p2_enc)
pp = dataset.preprocessor
gcn = pp.items[1]
means = gcn.means
normalizers = gcn.normalizers
toshape = (num_examples, )
for i in range(input_dim):
toshape += (1, )
if num_channels != 1:
toshape += (1, )
# When the number of patches and patch-pairs differs, this breaks.
# I need to match up normalizers/means with their corresponding patches
# undoing the PCA might be breaking too, but without errors...
normalizers1 = expand_p1(normalizers)
normalizers2 = expand_p2(normalizers)
means1 = expand_p1(means)
means2 = expand_p2(means)
p1_enc *= normalizers1.reshape(toshape)
p1_enc += means1.reshape(toshape)
p2_enc *= normalizers2.reshape(toshape)
p2_enc += means2.reshape(toshape)
# Now, we pull off the same examples from the data to compare to dAE inputs in plots
outputs = copy.deepcopy(images_data)
insertpatches(outputs, p1_enc[to_plot], p1x, p1y, patch_width)
insertpatches(outputs, p2_enc[to_plot], p2x, p2y, patch_width)
plt.figure()
for i in range(num2plot):
# Inputs
plt.subplot(num2plot, 2, 2 * i + 1)
plt.imshow(images_data[i], cmap=cm.Greys_r)
print stamps_data[i]
a = (stamps_data[i, 2] + patch_width / 2,
stamps_data[i, 1] + patch_width / 2, stamps_data[i, 6],
stamps_data[i, 5])
plt.arrow(a[0], a[1], a[2], a[3], head_width=1.0, head_length=0.6)
# Outputs
plt.subplot(num2plot, 2, 2 * i + 2)
plt.imshow(outputs[i], cmap=cm.Greys_r)
plt.arrow(a[0],
a[1],
d_out[to_plot[i], 1],
d_out[to_plot[i], 0],
head_width=1.0,
head_length=0.6)
plt.show()
savestr = 'cifar_ppd.png'
plt.savefig(savestr)
