dae.md

Denoising Autoencoder

You can try the Denoising Autoencoder Example running on Mnist. Here we build two hidden layer encoding and decoding layers

from mozi.model import Sequential
from mozi.layers.linear import Linear
from mozi.layers.activation import *
from mozi.layers.noise import Gaussian
import theano.tensor as T

# build model
model = Sequential(input_var=T.matrix(), output_var=T.matrix())
# build encoder
model.add(Gaussian())
encode_layer1 = Linear(prev_dim=28*28, this_dim=200)
model.add(encode_layer1)
model.add(RELU())
encode_layer2 = Linear(prev_dim=200, this_dim=50)
model.add(encode_layer2)
model.add(Tanh())

# build decoder
decode_layer1 = Linear(prev_dim=50, this_dim=200, W=encode_layer2.W.T)
model.add(decode_layer1)
model.add(RELU())
decode_layer2 = Linear(prev_dim=200, this_dim=28*28, W=encode_layer1.W.T)
model.add(decode_layer2)
model.add(Sigmoid())

Next we prepare the mnist dataset such that the input X and output y is the same

from mozi.datasets.mnist import Mnist

# build dataset
data = Mnist(batch_size=64, train_valid_test_ratio=[5,1,1])
# for autoencoder, the output will be equal to input
data.set_train(X=data.get_train().X, y=data.get_train().X)
data.set_valid(X=data.get_valid().X, y=data.get_valid().X)

Finally build learning method and put everything in train object and run

from mozi.train_object import TrainObject
from mozi.cost import entropy
from mozi.learning_method import AdaGrad

learning_method = AdaGrad(learning_rate=0.01, momentum=0.9,
                          lr_decay_factor=0.9, decay_batch=10000)

train_object = TrainObject(model = model,
                           log = None,
                           dataset = data,
                           train_cost = entropy,
                           valid_cost = entropy,
                           learning_method = learning_method,
                           stop_criteria = {'max_epoch' : 10,
                                            'epoch_look_back' : 5,
                                            'percent_decrease' : 0.01}
                           )
# finally run the code
train_object.setup()
train_object.run()