TensorFlow

myconvtensorflow.py

#!/bin/env python

import math
import tensorflow as tf

from tensorflow.examples.tutorials.mnist import input_data as mnist_data
print("Tensorflow version " + tf.__version__)

# https://stackoverflow.com/questions/47068709/your-cpu-supports-instructions-that-this-tensorflow-binary-was-not-compiled-to-u
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'


if __name__ == '__main__':
  mnist = mnist_data.read_data_sets("data", one_hot=True, reshape=False, validation_size=0)

  # input X: 28x28 grayscale images, the first dimension (None) will index the images in the mini-batch
  X = tf.placeholder(tf.float32, [None, 28, 28, 1])
  # correct answers will go here
  Y_ = tf.placeholder(tf.float32, [None, 10])

  # dropout
  pkeep = tf.placeholder(tf.float32)

  # weights   784=28*28
  # update 1: initialize using random numbers other than zeros
  W = tf.Variable(tf.truncated_normal([28, 28, 1, 4], stddev=0.1))
  # biases
  B = tf.Variable(tf.ones([4])/10) # 4 is the number of output channels

  # convolutional layers
  W1 = tf.Variable(tf.truncated_normal([14, 14, 4, 8], stddev=0.1))
  B1 = tf.Variable(tf.ones([8])/10)
  W2 = tf.Variable(tf.truncated_normal([7, 7, 8, 12], stddev=0.1))
  B2 = tf.Variable(tf.ones([12])/10)

  # connected layers
  W3 = tf.Variable(tf.truncated_normal([7 * 7 * 12, 200], stddev=0.1))
  B3 = tf.Variable(tf.ones([200])/10)
  W4 = tf.Variable(tf.truncated_normal([200, 10], stddev=0.1))
  B4 = tf.Variable(tf.ones([10])/10)

  Y1 = tf.nn.relu(tf.nn.conv2d(X, W, strides=[1, 1, 1, 1], padding='SAME') + B)
  Y2 = tf.nn.relu(tf.nn.conv2d(Y1, W1, strides=[1, 2, 2, 1], padding='SAME') + B1)
  Y3 = tf.nn.relu(tf.nn.conv2d(Y2, W2, strides=[1, 2, 2, 1], padding='SAME') + B2)

  YY = tf.reshape(Y3, shape=[-1, 7 * 7 * 12])
  Y4 = tf.nn.relu(tf.matmul(YY, W3) + B3)
  Ylogits = tf.matmul(Y4, W4) + B4
  Y = tf.nn.softmax(Ylogits)

  # cost function
  cross_entropy = tf.nn.softmax_cross_entropy_with_logits(logits=Ylogits, labels=Y_)
  cross_entropy = tf.reduce_mean(cross_entropy) * 100

  # optimizer and train
  # optimizer = tf.train.GradientDescentOptimizer(0.003)
  # train_step = optimizer.minimize(cross_entropy)
  lr = tf.placeholder(tf.float32)
  train_step = tf.train.AdamOptimizer(lr).minimize(cross_entropy)

  # accuracy calculation
  is_correct = tf.equal(tf.argmax(Y,1), tf.argmax(Y_,1))
  accuracy = tf.reduce_mean(tf.cast(is_correct, tf.float32))



  init = tf.global_variables_initializer()
  session = tf.Session()
  session.run(init)

  for i in range(1000):

    # adaptive learning rate
    max_learning_rate = 0.003
    min_learning_rate = 0.0001
    decay_speed = 2000.0 # 0.003-0.0001-2000=>0.9826 done in 5000 iterations
    learning_rate = min_learning_rate + (max_learning_rate - min_learning_rate) * math.exp(-i / decay_speed)

    batch_X, batch_Y = mnist.train.next_batch(100)
    train_data={X: batch_X, Y_: batch_Y, lr: learning_rate}

    session.run(train_step, feed_dict=train_data)

    a, c = session.run([accuracy, cross_entropy], feed_dict=train_data)
    print 'Training accuracy:', a

    if i % 100 == 0:
      test_data={X: mnist.test.images, Y_: mnist.test.labels}
      a, c = session.run([accuracy, cross_entropy], feed_dict=test_data)
      print 'Test accuracy:', a

mytensorflow.py

#!/bin/env python

import math
import tensorflow as tf

from tensorflow.examples.tutorials.mnist import input_data as mnist_data
print("Tensorflow version " + tf.__version__)

# https://stackoverflow.com/questions/47068709/your-cpu-supports-instructions-that-this-tensorflow-binary-was-not-compiled-to-u
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'


if __name__ == '__main__':
  mnist = mnist_data.read_data_sets("data", one_hot=True, reshape=False, validation_size=0)

  # input X: 28x28 grayscale images, the first dimension (None) will index the images in the mini-batch
  X = tf.placeholder(tf.float32, [None, 28, 28, 1])
  # correct answers will go here
  Y_ = tf.placeholder(tf.float32, [None, 10])

  # dropout
  pkeep = tf.placeholder(tf.float32)

  # weights   784=28*28
  # update 1: initialize using random numbers other than zeros
  W1 = tf.Variable(tf.truncated_normal([28 * 28, 200], stddev=0.1))
  # biases
  B1 = tf.Variable(tf.zeros([200]))

  W2 = tf.Variable(tf.truncated_normal([200, 100], stddev=0.1))
  B2 = tf.Variable(tf.zeros([100]))
  W3 = tf.Variable(tf.truncated_normal([100, 60], stddev=0.1))
  B3 = tf.Variable(tf.zeros([60]))
  W4 = tf.Variable(tf.truncated_normal([60, 30], stddev=0.1))
  B4 = tf.Variable(tf.zeros([30]))
  W5 = tf.Variable(tf.truncated_normal([30, 10], stddev=0.1))
  B5 = tf.Variable(tf.zeros([10]))

  XX = tf.reshape(X, [-1, 28 * 28])
  Y1 = tf.nn.relu(tf.matmul(XX, W1) + B1)
  Y2 = tf.nn.relu(tf.matmul(Y1, W2) + B2)
  Y3 = tf.nn.relu(tf.matmul(Y2, W3) + B3)
  Y4 = tf.nn.relu(tf.matmul(Y3, W4) + B4)
  Ylogits = tf.matmul(Y4, W5) + B5
  Y  = tf.nn.softmax(Ylogits)

  # cost function
  cross_entropy = tf.nn.softmax_cross_entropy_with_logits(logits=Ylogits, labels=Y_)
  cross_entropy = tf.reduce_mean(cross_entropy) * 100

  # optimizer and train
  # optimizer = tf.train.GradientDescentOptimizer(0.003)
  # train_step = optimizer.minimize(cross_entropy)
  lr = tf.placeholder(tf.float32)
  train_step = tf.train.AdamOptimizer(lr).minimize(cross_entropy)

  # accuracy calculation
  is_correct = tf.equal(tf.argmax(Y,1), tf.argmax(Y_,1))
  accuracy = tf.reduce_mean(tf.cast(is_correct, tf.float32))



  init = tf.global_variables_initializer()
  session = tf.Session()
  session.run(init)

  for i in range(5000):

    # adaptive learning rate
    max_learning_rate = 0.003
    min_learning_rate = 0.0001
    decay_speed = 2000.0 # 0.003-0.0001-2000=>0.9826 done in 5000 iterations
    learning_rate = min_learning_rate + (max_learning_rate - min_learning_rate) * math.exp(-i / decay_speed)

    batch_X, batch_Y = mnist.train.next_batch(100)
    train_data={X: batch_X, Y_: batch_Y, lr: learning_rate}

    session.run(train_step, feed_dict=train_data)

    a, c = session.run([accuracy, cross_entropy], feed_dict=train_data)
    print 'Training accuracy:', a

    test_data={X: mnist.test.images, Y_: mnist.test.labels}
    a, c = session.run([accuracy, cross_entropy], feed_dict=test_data)
    print 'Test accuracy:', a

https://github.com/GoogleCloudPlatform/tensorflow-without-a-phd

https://codelabs.developers.google.com/codelabs/cloud-tensorflow-mnist