main_dir='hidden_layers/',
enc_act_func='sigmoid',
dec_act_func='sigmoid',
loss_func='mean_squared',
num_epochs=31,
batch_size=12,
dataset='cifar10',
xavier_init=1,
opt='adam',
learning_rate=0.001,
momentum=0.5,
corr_type='gaussian',
corr_frac=0.6,
verbose=1,
seed=-1)
dae.fit(trX, val_dict, teX, restore_previous_model=False)
dae.reset()
# =============================================================================
# Testing learning rates
# =============================================================================
elif arg == 'lrs':
feed_list = [0.00001, 0.0001, 0.001, 0.01, 0.1, 0.5, 1.0]
for i in feed_list:
print "\n Evaluating for lr=" + str(i)
t = arg + '=' + str(i)
dae = DAE(model_name=arg + '_model',
pickle_name=arg,
test_name=t,
n_components=256,
main_dir='hidden_layers/',
from LeNet5 import LeNet
from autoencoder import DenoisingAutoencoder as DAE
if sys.argv[1] =='dae':
dae = DAE(model_name='dae_svm', pickle_name='svm', test_name='svm',
n_components=256, main_dir='dae/',
enc_act_func='sigmoid', dec_act_func='none',
loss_func='mean_squared', num_epochs=50, batch_size=20,
dataset='cifar10', xavier_init=1, opt='adam',
learning_rate=0.0001, momentum=0.5, corr_type='gaussian',
corr_frac=0.5, verbose=1, seed=1)
trX, trY, teX, teY = getdata.load_cifar10_dataset('../cifar-10-batches-py/', mode='supervised')
val_dict = {}
dae.fit(trX, val_dict, teX, restore_previous_model=True)
#dae.load_model(256, 'models/dae/dae_svm')
dae_svm_train = dae.transform(trX, name='dae_svm_train_na', save=True)
dae_svm_test = dae.transform(teX, name='dae_svm_test_na', save=True)
elif sys.argv[1]=='cnn':
cifar_train = getdata.get_train()
cifar_test = getdata.get_test()
acc_list = []
cnn = LeNet(lr=0.001, epochs=101, batch_size=256, train_data=cifar_train,
test_data=cifar_test, wd=0.004, decay_lr=False,
decay_w=False, optimizer='rmsprop', seed=124,
model_name='lenet/rmsprop', init_dev=0.01, drop=0.2)
cnn.restore('models/lenet/rmsprop')