from da import DenoisingAutoencoder
from dataset import Dataset
import theano.tensor as T
import numpy
if __name__ == "__main__":
fname = "/data/lisa/data/mnist/mnist_all.pickle"
# fname = "/data/lisa/data/pentomino/"
ds = Dataset()
ds.setup_dataset(data_path=fname, train_split_scale=0.8)
x_data = ds.Xtrain
input = T.dmatrix("x_input")
weights_file = "../out/dae_mnist_weights.npy"
recons_file = "../out/dae_mnist_recons.npy"
rnd = numpy.random.RandomState(1231)
dae = DenoisingAutoencoder(input, nvis=28 * 28, nhid=600, rnd=rnd)
dae.fit(learning_rate=0.1, data=x_data, weights_file=weights_file, n_epochs=100, recons_img_file=recons_file)
nhid_l2 = 800
dae_l1 = DenoisingAutoencoder(input,
nvis=28*28,
nhid=nhid_l1,
L1_reg=9*1e-5,
L2_reg=7*1e-4,
rnd=rnd)
#std_data = standardize(x_data)
std_data = numpy.asarray(x_data, dtype=theano.config.floatX)
dae_l1.fit(learning_rate=9.96*1e-3,
shuffle_data=True,
data=std_data,
weights_file=weights_file_l1,
recons_img_file=None,
corruption_level=0.095,
batch_size=40,
n_epochs=1400)
dae_l1_obj_out = open("dae_l1_obj.pkl", "wb")
pkl.dump(dae_l1, dae_l1_obj_out, protocol=pkl.HIGHEST_PROTOCOL)
dae_l1_out = dae_l1.encode(input)
dae_l1_h = dae_l1.encode(std_data)
dae_l1_h_fn = theano.function([], dae_l1_h)
dae_l2_in = dae_l1_h_fn()
dae_l2_in = numpy.asarray(dae_l2_in, dtype=theano.config.floatX)
dae_l2 = DenoisingAutoencoder(dae_l1_out,
from da import DenoisingAutoencoder
from dataset import Dataset
import theano.tensor as T
import numpy
if __name__ == "__main__":
fname = "/data/lisa/data/pentomino/pento64x64_40k_seed_5365102867_64patches.npy"
ds = Dataset()
ds.setup_dataset(data_path=fname, train_split_scale=0.4)
x_data = ds.Xtrain
input = T.dmatrix("x_input")
rnd = numpy.random.RandomState(1231)
dae = DenoisingAutoencoder(input, nvis=64 * 64, nhid=1500, rnd=rnd)
dae.fit(data=x_data,
learning_rate=0.04,
n_epochs=32,
weights_file="out/dae_weights_pento.npy")
from da import DenoisingAutoencoder
from dataset import Dataset
import theano.tensor as T
import numpy
if __name__ == "__main__":
fname = "/data/lisa/data/mnist/mnist_all.pickle"
#fname = "/data/lisa/data/pentomino/"
ds = Dataset()
ds.setup_dataset(data_path=fname, train_split_scale=0.8)
x_data = ds.Xtrain
input = T.dmatrix("x_input")
weights_file = "../out/dae_mnist_weights.npy"
recons_file = "../out/dae_mnist_recons.npy"
rnd = numpy.random.RandomState(1231)
dae = DenoisingAutoencoder(input, nvis=28 * 28, nhid=600, rnd=rnd)
dae.fit(learning_rate=0.1,
data=x_data,
weights_file=weights_file,
n_epochs=100,
recons_img_file=recons_file)
from da import DenoisingAutoencoder
from dataset import Dataset
import theano.tensor as T
import numpy
if __name__=="__main__":
fname = "/data/lisa/data/pentomino/pento64x64_40k_seed_5365102867_64patches.npy"
ds = Dataset()
ds.setup_dataset(data_path=fname, train_split_scale=0.4)
x_data = ds.Xtrain
input = T.dmatrix("x_input")
rnd = numpy.random.RandomState(1231)
dae = DenoisingAutoencoder(input,
nvis=64*64,
nhid=1500,
rnd=rnd)
dae.fit(data=x_data,
learning_rate=0.04,
n_epochs=32,
weights_file="out/dae_weights_pento.npy")