Keras
MNIST
Autoencoder
Backend

概要

kerasのbackendで、autoencoderやってみた。

実行結果

auto0.png

サンプルコード

from tensorflow.contrib.keras.python.keras import backend as K
from tensorflow.contrib.keras.python.keras.optimizers import SGD, Adam, RMSprop, Adagrad, Adadelta
from tensorflow.contrib.keras.python.keras.datasets import mnist
import numpy as np
import matplotlib.pyplot as plt

(x_train, _), (x_test, _) = mnist.load_data()
x_train = x_train.astype('float32') / 255.
x_test = x_test.astype('float32') / 255.
x_train = x_train.reshape((len(x_train), np.prod(x_train.shape[1 : ])))
x_test = x_test.reshape((len(x_test), np.prod(x_test.shape[1 : ])))

input_dim = 784
output_dim = 784
hidden_dim = 32
x = K.placeholder(shape = (None, input_dim), name = "x")
ytrue = K.placeholder(shape = (None, output_dim), name = "y")
W1 = K.random_uniform_variable((input_dim, hidden_dim), 0, 1, name = "W1")
W2 = K.random_uniform_variable((hidden_dim, output_dim), 0, 1, name = "W2")
b1 = K.random_uniform_variable((hidden_dim, ), 0, 1, name = "b1")
b2 = K.random_uniform_variable((output_dim, ), 0, 1, name = "b2")
params = [W1, b1, W2, b2]
hidden = K.relu(K.dot(x, W1) + b1)
ypred = K.relu(K.dot(hidden, W2) + b2)
loss = K.mean(K.square(ypred - ytrue), axis = -1)
opt = RMSprop()
updates = opt.get_updates(params, [], loss)
train = K.function(inputs = [x, ytrue], outputs = [loss], updates = updates)
test = K.function(inputs = [x], outputs = [ypred])
for ep in range(10000):
    for i in range(10):
        st = train([[x_train[i]], [x_train[i]]])
    if ep % 1000 == 0:
        print (ep, st[0])

n = 10
plt.figure(figsize = (20, 4))
for i in range(n):
    ax = plt.subplot(2, n, i + 1)
    plt.imshow(x_train[i].reshape(28, 28))
    ax = plt.subplot(2, n, i + 1 + n)
    img = test([[x_train[i]]])
    plt.imshow(img[0].reshape(28, 28))
plt.savefig("auto0.png")
plt.show()

以上。