Last active
November 6, 2017 10:12
-
-
Save wedesoft/c2827ce7abedae2ebe04ad68909b8c84 to your computer and use it in GitHub Desktop.
MNIST autoencoder implemented using Tensorflow
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
#!/usr/bin/env python3 | |
# http://blog.aloni.org/posts/backprop-with-tensorflow/ | |
import random | |
import numpy as np | |
import tensorflow as tf | |
import pickle | |
import gzip | |
import cv2 | |
def show(title, img, wait=-1): | |
cv2.imshow(title, cv2.resize(img.reshape(28, 28), (280, 280))) | |
return cv2.waitKey(wait or 1) != 27 | |
if __name__ == '__main__': | |
training, validation, testing = pickle.load(gzip.open('mnist.pkl.gz', 'rb'), encoding='iso-8859-1') | |
with tf.Session() as sess: | |
saver = tf.train.import_meta_graph('auto.meta') | |
saver.restore(sess, 'auto') | |
image = np.random.rand(1, 784) | |
prediction = tf.get_collection('prediction')[0] | |
while True: | |
index = random.randrange(len(testing[0])) | |
noise = np.random.rand(1, 784) | |
image = np.clip(testing[0][index:index + 1] + (2 * noise - 1) ** 5, 0, 1) | |
show('input', image, False) | |
result = sess.run(prediction, feed_dict={'x:0': image}) | |
if not show('prediction', result, 300): | |
break |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
#!/usr/bin/env python3 | |
# https://github.com/aymericdamien/TensorFlow-Examples/blob/master/examples/3_NeuralNetworks/autoencoder.py | |
import math | |
import pickle | |
import gzip | |
from functools import reduce | |
from operator import add | |
import numpy as np | |
import tensorflow as tf | |
from tqdm import tqdm | |
import cv2 | |
def random_choice(count, size): | |
return np.random.choice(count, size, replace=False) | |
def random_selection(size, *arrays): | |
indices = random_choice(len(arrays[0]), size) | |
result = tuple(np.take(array, indices, axis=0) for array in arrays) | |
return result[0] if len(result) == 1 else result | |
def show(title, img, wait=True): | |
cv2.imshow(title, cv2.resize(img.reshape(28, 28), (280, 280))) | |
return cv2.waitKey(wait or 1) != 27 | |
if __name__ == '__main__': | |
# http://deeplearning.net/data/mnist/mnist.pkl.gz | |
training, validation, testing = pickle.load(gzip.open('mnist.pkl.gz', 'rb'), encoding='iso-8859-1') | |
n_iterations = 50000 | |
batch_size = 256 | |
n_hidden1 = 200 | |
n_hidden2 = 30 | |
alpha = 0.1 | |
x = tf.placeholder(tf.float32, [None, 28 * 28], name='x') | |
m1 = tf.Variable(tf.random_normal([28 * 28, n_hidden1], stddev=0.1)) | |
b1 = tf.Variable(tf.random_normal([n_hidden1])) | |
m2 = tf.Variable(tf.random_normal([n_hidden1, n_hidden2], stddev=0.1)) | |
b2 = tf.Variable(tf.random_normal([n_hidden2])) | |
m3 = tf.Variable(tf.random_normal([n_hidden2, n_hidden1], stddev=0.1)) | |
b3 = tf.Variable(tf.random_normal([n_hidden1])) | |
m4 = tf.Variable(tf.random_normal([n_hidden1, 28 * 28], stddev=0.1)) | |
b4 = tf.Variable(tf.random_normal([28 * 28])) | |
theta = [m1, b1, m2, b2, m3, b3, m4, b4] | |
a0 = x | |
z1 = tf.add(tf.matmul( x, m1), b1) | |
a1 = tf.sigmoid(z1) | |
z2 = tf.add(tf.matmul(a1, m2), b2) | |
a2 = tf.sigmoid(z2) | |
z3 = tf.add(tf.matmul(a2, m3), b3) | |
a3 = tf.sigmoid(z3) | |
z4 = tf.add(tf.matmul(a3, m4), b4) | |
a4 = tf.sigmoid(z4) | |
h = a4 | |
m = tf.to_float(tf.shape(x)[0]) | |
cost = tf.reduce_sum(tf.pow(x - h, 2)) / m | |
dtheta = tf.gradients(cost, theta) | |
step = [tf.assign(value, tf.subtract(value, tf.multiply(alpha, dvalue))) for value, dvalue in zip(theta, dtheta)] | |
saver = tf.train.Saver() | |
with tf.Session() as session: | |
train = {x: training[0]} | |
j_train = 0.5 * 768 | |
session.run(tf.global_variables_initializer()) | |
progress = tqdm(range(n_iterations)) | |
for i in progress: | |
selection = random_selection(batch_size, train[x]) | |
mini_batch = {x: selection} | |
j_train = 0.99 * j_train + 0.01 * session.run(cost, feed_dict=mini_batch) | |
progress.set_description('cost: %8.6f' % j_train) | |
if i % 50 == 0: | |
show('original', selection[0:1], False) | |
show('reconstruction', session.run(h, feed_dict={x: selection[0:1]}), 10) | |
session.run(step, feed_dict=mini_batch) | |
tf.add_to_collection('prediction', h) | |
saver.save(session, './auto') |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment