def createSAEModels(self):
for inovelty in range(self.all_trgt_sparse.shape[1]):
# Initialize SAE objects for all novelties
self.SAE[inovelty] = StackedAutoEncoders(
parameters=self.parameters,
save_path=self.getBaseResultsPath(),
CVO=self.CVO,
inovelty=inovelty,
verbose=self.verbose)
return self.SAE
def createSAEModels(self):
self.SAE = {}
self.trn_data = {}
self.trn_trgt = {}
self.trn_trgt_sparse = {}
if (self.trn_params == None):
self.loadTrainParameters()
for inovelty in range(self.trgt_sparse.shape[1]):
self.trn_data[inovelty] = self.all_data[self.all_trgt != inovelty]
self.trn_trgt[inovelty] = self.all_trgt[self.all_trgt != inovelty]
self.trn_trgt[inovelty][
self.trn_trgt[inovelty] > inovelty] = self.trn_trgt[inovelty][
self.trn_trgt[inovelty] > inovelty] - 1
self.trn_trgt_sparse[inovelty] = np_utils.to_categorical(
self.trn_trgt[inovelty].astype(int))
# Initialize an SAE object for all novelties
self.SAE[inovelty] = StackedAutoEncoders(
params=self.trn_params,
development_flag=self.development_flag,
n_folds=self.n_folds,
save_path=self.getBaseResultsPath(),
CVO=self.CVO,
noveltyDetection=True,
inovelty=inovelty)
# Train Process
if len(sys.argv) < 2:
print '[-] Usage: %s <novelty class>' % sys.argv[0]
exit()
inovelty = int(sys.argv[2])
print '\nNovelty class to train: %i' % inovelty
# Choose neurons topology
# Create neurons vector to be used in multiprocessing.Pool()
SAE = StackedAutoEncoders(params=trn_params,
development_flag=development_flag,
n_folds=n_folds,
save_path=results_path,
CVO=CVO,
noveltyDetection=True,
inovelty=inovelty,
allow_change_weights=False)
n_folds = len(CVO[inovelty])
hidden_neurons = range(400, 0, -50) + [2]
print hidden_neurons
regularizer = "" #dropout / l1 / l2
regularizer_param = 0.5
trn_data = all_data[all_trgt != inovelty]
trn_trgt = all_trgt[all_trgt != inovelty]
trn_trgt[trn_trgt > inovelty] = trn_trgt[trn_trgt > inovelty] - 1
trn_params = trnparams.NeuralClassificationTrnParams()
trn_params.load(trn_params_folder)
# Choose how many fold to be used in Cross Validation
n_folds = 2
CVO = trnparams.ClassificationFolds(folder=results_path,
n_folds=2,
trgt=all_trgt,
dev=development_flag,
verbose=True)
print trn_params.get_params_str()
# Train Process
SAE = StackedAutoEncoders(params=trn_params,
development_flag=development_flag,
n_folds=n_folds,
save_path=results_path,
CVO=CVO)
# Choose layer to be trained
layer = 1
hidden_neurons = range(400, 0, -50) + [2]
regularizer = "dropout"
regularizer_param = 0.5
print hidden_neurons
# Functions defined to be used by multiprocessing.Pool()
def trainNeuron(ineuron):
for ifold in range(n_folds):