from cle.cle.train import Training
from cle.cle.train.ext import (EpochCount, GradientClipping, Monitoring,
Picklize, EarlyStopping)
from cle.cle.train.opt import RMSProp
from cle.cle.utils import error, flatten, predict, OrderedDict
from cle.datasets.mnist import MNIST
# Set your dataset
data_path = '/home/junyoung/data/mnist/mnist.pkl'
save_path = '/home/junyoung/src/cle/saved/'
batch_size = 128
debug = 0
model = Model()
train_data = MNIST(name='train', path=data_path)
valid_data = MNIST(name='valid', path=data_path)
# Choose the random initialization method
init_W = InitCell('randn')
init_b = InitCell('zeros')
# Define nodes: objects
x, y = train_data.theano_vars()
# You must use THEANO_FLAGS="compute_test_value=raise" python -m ipdb
if debug:
x.tag.test_value = np.zeros((batch_size, 784), dtype=np.float32)
y.tag.test_value = np.zeros((batch_size, 1), dtype=np.float32)
h1 = FullyConnectedLayer(name='h1',
from cle.cle.train.opt import Adam
from cle.cle.utils import flatten
from cle.cle.utils.compat import OrderedDict
from cle.datasets.mnist import MNIST
datapath = '/home/junyoung/data/mnist/mnist_binarized_salakhutdinov.pkl'
savepath = '/home/junyoung/repos/cle/saved/'
batch_size = 100
input_dim = 784
latent_dim = 100
n_steps = 64
debug = 0
model = Model()
data = MNIST(name='train', unsupervised=1, path=datapath)
init_W = InitCell('rand')
init_U = InitCell('ortho')
init_b = InitCell('zeros')
init_b_sig = InitCell('const', mean=0.6)
x, _ = data.theano_vars()
if debug:
x.tag.test_value = np.zeros((batch_size, 784), dtype=np.float32)
error = ErrorLayer(name='error',
parent=['x'],
recurrent=['canvas'],
batch_size=batch_size)