In [7]:

    

from __future__ import print_function
import numpy as np
import tensorflow as tf
from six.moves import cPickle as pickle


    

In [8]:

    

pickle_file = '/Users/rringham/Projects/TensorFlow/tensorflow/tensorflow/examples/udacity/notMNIST.pickle'

with open(pickle_file, 'rb') as f:
  save = pickle.load(f)
  train_dataset = save['train_dataset']
  train_labels = save['train_labels']
  valid_dataset = save['valid_dataset']
  valid_labels = save['valid_labels']
  test_dataset = save['test_dataset']
  test_labels = save['test_labels']
  del save  # hint to help gc free up memory
  print('Training set', train_dataset.shape, train_labels.shape)
  print('Validation set', valid_dataset.shape, valid_labels.shape)
  print('Test set', test_dataset.shape, test_labels.shape)


    

    




Training set (200000, 28, 28) (200000,)
Validation set (10000, 28, 28) (10000,)
Test set (10000, 28, 28) (10000,)

In [9]:

    

image_size = 28
num_labels = 10

def reformat(dataset, labels):
  dataset = dataset.reshape((-1, image_size * image_size)).astype(np.float32)
  # Map 2 to [0.0, 1.0, 0.0 ...], 3 to [0.0, 0.0, 1.0 ...]
  labels = (np.arange(num_labels) == labels[:,None]).astype(np.float32)
  return dataset, labels
train_dataset, train_labels = reformat(train_dataset, train_labels)
valid_dataset, valid_labels = reformat(valid_dataset, valid_labels)
test_dataset, test_labels = reformat(test_dataset, test_labels)
print('Training set', train_dataset.shape, train_labels.shape)
print('Validation set', valid_dataset.shape, valid_labels.shape)
print('Test set', test_dataset.shape, test_labels.shape)


    

    




Training set (200000, 784) (200000, 10)
Validation set (10000, 784) (10000, 10)
Test set (10000, 784) (10000, 10)

In [10]:

    

def accuracy(predictions, labels):
  return (100.0 * np.sum(np.argmax(predictions, 1) == np.argmax(labels, 1))
          / predictions.shape[0])


    

In [19]:

    

import math as math

batch_size = 1
initial_learning_rate_value = 0.5
beta1 = 0.001
beta2 = 0.001
beta3 = 0.001
beta4 = 0.001
beta5 = 0.001

graph = tf.Graph()
with graph.as_default():

  # Input data. For the training data, we use a placeholder that will be fed
  # at run time with a training minibatch.
  tf_train_dataset = tf.placeholder(tf.float32, shape=(batch_size, image_size * image_size))
  tf_train_labels = tf.placeholder(tf.float32, shape=(batch_size, num_labels))
  tf_valid_dataset = tf.constant(valid_dataset)
  tf_test_dataset = tf.constant(test_dataset)    
    
  # learning rate decay
  global_step = tf.Variable(0)
  learning_rate = tf.train.exponential_decay(initial_learning_rate_value, global_step, 1, 0.9999)

  # Variables
  h1_size = 1024  
  h2_size = 1024
  h3_size = 305
  h4_size = 75
    
  weights_h1 = tf.Variable(tf.truncated_normal([image_size * image_size, h1_size], stddev=math.sqrt(2.0/(image_size*image_size))))
  biases_h1 = tf.Variable(tf.zeros([h1_size]))
    
  weights_h2 = tf.Variable(tf.truncated_normal([h1_size, h2_size], stddev=math.sqrt(2.0/h1_size)))
  biases_h2 = tf.Variable(tf.zeros([h2_size]))

  weights_h3 = tf.Variable(tf.truncated_normal([h2_size, h3_size], stddev=math.sqrt(2.0/h2_size)))
  biases_h3 = tf.Variable(tf.zeros([h3_size]))

  weights_h4 = tf.Variable(tf.truncated_normal([h3_size, h4_size], stddev=math.sqrt(2.0/h3_size)))
  biases_h4 = tf.Variable(tf.zeros([h4_size]))
    
  weights_o = tf.Variable(tf.truncated_normal([h4_size, num_labels], stddev=math.sqrt(2.0/h4_size)))
  biases_o = tf.Variable(tf.zeros([num_labels]))  

  # Model.
  def model(data):
    h1 = tf.nn.relu(tf.matmul(data, weights_h1) + biases_h1)
    h2 = tf.nn.relu(tf.matmul(h1, weights_h2) + biases_h2)
    h3 = tf.nn.relu(tf.matmul(h2, weights_h3) + biases_h3)
    h4 = tf.nn.relu(tf.matmul(h3, weights_h4) + biases_h4)
    return tf.matmul(h4, weights_o) + biases_o
    
  # Training computation.
  logits = model(tf_train_dataset)
  loss = tf.reduce_mean(
    tf.nn.softmax_cross_entropy_with_logits(logits, tf_train_labels)) + ((beta1 * tf.nn.l2_loss(weights_h1)) + (beta2 * tf.nn.l2_loss(weights_h2)) + (beta3 * tf.nn.l2_loss(weights_h3)) + (beta4 * tf.nn.l2_loss(weights_h4)) + (beta5 * tf.nn.l2_loss(weights_o)))
  
  # Optimizer.
  optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss, global_step=global_step)

  # Predictions for the training, validation, and test data.  
  train_prediction = tf.nn.softmax(model(tf_train_dataset))  
  valid_prediction = tf.nn.softmax(model(tf_valid_dataset))
  test_prediction = tf.nn.softmax(model(tf_test_dataset))


    

In [ ]:

    

num_steps = 50

with tf.Session(graph=graph) as session:
  tf.initialize_all_variables().run()
  print("Initialized")
  for step in range(num_steps):
    # Pick an offset within the training data, which has been randomized.
    # Note: we could use better randomization across epochs.
    offset = (step * batch_size) % (train_labels.shape[0] - batch_size)
    
    # Generate a minibatch.
    batch_data = train_dataset[offset:(offset + batch_size), :]
    batch_labels = train_labels[offset:(offset + batch_size), :]
    
    # Prepare a dictionary telling the session where to feed the minibatch.
    # The key of the dictionary is the placeholder node of the graph to be fed,
    # and the value is the numpy array to feed to it.
    feed_dict = {tf_train_dataset : batch_data, tf_train_labels : batch_labels}
    
    _, l, predictions = session.run(
      [optimizer, loss, train_prediction], feed_dict=feed_dict)
          
    print("Minibatch loss at step %d: %f" % (step, l))
    print("Learning rate at step %d: %f" % (step, learning_rate.eval()))
    print("Minibatch accuracy: %.1f%%" % accuracy(predictions, batch_labels))
    print("Validation accuracy: %.1f%%\n" % accuracy(valid_prediction.eval(), valid_labels))


#   for step in range(num_steps):
#     # Pick an offset within the training data, which has been randomized.
#     # Note: we could use better randomization across epochs.
#     offset = (step * batch_size) % (train_labels.shape[0] - batch_size)
    
#     # Generate a minibatch.
#     batch_data = train_dataset[offset:(offset + batch_size), :]
#     batch_labels = train_labels[offset:(offset + batch_size), :]
    
#     # Prepare a dictionary telling the session where to feed the minibatch.
#     # The key of the dictionary is the placeholder node of the graph to be fed,
#     # and the value is the numpy array to feed to it.
#     feed_dict = {tf_train_dataset : batch_data, tf_train_labels : batch_labels}
    
#     _, l, predictions = session.run(
#       [optimizer, loss, train_prediction], feed_dict=feed_dict)
    
#     if (step % 500 == 0):
#       print("Minibatch loss at step %d: %f" % (step, l))
#       print("Learning rate at step %d: %f" % (step, learning_rate.eval()))
#       print("Minibatch accuracy: %.1f%%" % accuracy(predictions, batch_labels))
#       print("Validation accuracy: %.1f%%\n" % accuracy(
#         valid_prediction.eval(), valid_labels))
        
#     if (step % 2000 == 0):
#       print("Test accuracy: %.1f%%\n" % accuracy(test_prediction.eval(), test_labels))
        
  print("Final test accuracy: %.1f%%\n" % accuracy(test_prediction.eval(), test_labels))


    

    




Initialized
Minibatch loss at step 0: 4.682405
Learning rate at step 0: 0.499950
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 1: 53.765388
Learning rate at step 1: 0.499900
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 2: 1385.730225
Learning rate at step 2: 0.499850
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 3: 8224031.500000
Learning rate at step 3: 0.499800
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 4: 10606649370738688.000000
Learning rate at step 4: 0.499750
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 5: nan
Learning rate at step 5: 0.499700
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 6: nan
Learning rate at step 6: 0.499650
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 7: nan
Learning rate at step 7: 0.499600
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 8: nan
Learning rate at step 8: 0.499550
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 9: nan
Learning rate at step 9: 0.499500
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 10: nan
Learning rate at step 10: 0.499450
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 11: nan
Learning rate at step 11: 0.499400
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 12: nan
Learning rate at step 12: 0.499350
Minibatch accuracy: 100.0%
Validation accuracy: 10.0%

Minibatch loss at step 13: nan
Learning rate at step 13: 0.499300
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 14: nan
Learning rate at step 14: 0.499250
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 15: nan
Learning rate at step 15: 0.499200
Minibatch accuracy: 100.0%
Validation accuracy: 10.0%

Minibatch loss at step 16: nan
Learning rate at step 16: 0.499151
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 17: nan
Learning rate at step 17: 0.499101
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 18: nan
Learning rate at step 18: 0.499051
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 19: nan
Learning rate at step 19: 0.499001
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 20: nan
Learning rate at step 20: 0.498951
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 21: nan
Learning rate at step 21: 0.498901
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 22: nan
Learning rate at step 22: 0.498851
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 23: nan
Learning rate at step 23: 0.498801
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 24: nan
Learning rate at step 24: 0.498751
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 25: nan
Learning rate at step 25: 0.498701
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 26: nan
Learning rate at step 26: 0.498652
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 27: nan
Learning rate at step 27: 0.498602
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 28: nan
Learning rate at step 28: 0.498552
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 29: nan
Learning rate at step 29: 0.498502
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 30: nan
Learning rate at step 30: 0.498452
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 31: nan
Learning rate at step 31: 0.498402
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 32: nan
Learning rate at step 32: 0.498352
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 33: nan
Learning rate at step 33: 0.498303
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 34: nan
Learning rate at step 34: 0.498253
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 35: nan
Learning rate at step 35: 0.498203
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

Minibatch loss at step 36: nan
Learning rate at step 36: 0.498153
Minibatch accuracy: 0.0%
Validation accuracy: 10.0%

In [ ]:

    




    

In [ ]: