Udacity deep learning course assignment 3 problem 4

code.py

# download the file @ https://www.dropbox.com/s/urqmc4jgt66hbef/notMNIST.pickle?dl=0
pickle_file = 'notMNIST.pickle'

from time import strftime
from math import sqrt
from __future__ import print_function
import numpy as np
import tensorflow as tf
from six.moves import cPickle as pickle

#load data
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)
  
# reformat
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)

# accuracy calculation function
def accuracy(predictions, labels):
  return (100.0 * np.sum(np.argmax(predictions, 1) == np.argmax(labels, 1))
          / predictions.shape[0])
          
batch_size = 128
weights = []
biases = []
dims = [image_size*image_size, 1024, 305, 75, num_labels]

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, dims[0]))
  tf_train_labels = tf.placeholder(tf.float32, shape=(batch_size, dims[-1]))
  tf_valid_dataset = tf.constant(valid_dataset)
  tf_test_dataset = tf.constant(test_dataset)
  
  # variables
  for i in range(len(dims) - 1):
    stddev = sqrt(2/dims[i])
    weights.append(tf.Variable(tf.truncated_normal([dims[i], dims[i+1]], stddev=stddev)))
    biases.append(tf.Variable(tf.zeros([dims[i+1]])))
  
  keep_prob = tf.placeholder(tf.float32)
  # prediction
  def predict(_dataset, _weights, _biases, keep_prob=1.0):
    _output = tf.matmul(_dataset, _weights[0]) + _biases[0]
    for _w, _b in zip(_weights[1:], _biases[1:]):
      _output = tf.matmul(tf.nn.dropout(tf.nn.relu(_output), keep_prob), _w) + _b
    return _output
  
  train_prediction = predict(tf_train_dataset, weights, biases, keep_prob=keep_prob)
  loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(train_prediction, tf_train_labels))
  
  # regularization with l2 loss
#   beta = .002
#   l2_loss = 0
#   for w in weights:
#     l2_loss += tf.nn.l2_loss(w)

#   loss += beta * l2_loss
  
  # Optimizer.
  global_step = tf.Variable(0, trainable=False)  # count the number of steps taken.
  learning_rate = tf.train.exponential_decay(0.1, global_step, 3000, 0.96)
  optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss, global_step=global_step)
  
  # Predictions for the validation and test data.
  valid_prediction = tf.nn.softmax(predict(tf_valid_dataset, weights, biases))
  test_prediction = tf.nn.softmax(predict(tf_test_dataset, weights, biases))

num_epochs = 128
num_steps = num_epochs * train_labels.shape[0] // batch_size + 1

with tf.Session(graph=graph) as session:
  tf.initialize_all_variables().run()
  print("Initialized. Total steps: %d\n" % num_steps)
  print("    time    step   offset   minibatch loss   minibatch accuracy   validation accuracy")
  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, keep_prob: 0.75}
    _, l, predictions = session.run([optimizer, loss, train_prediction], feed_dict=feed_dict)
    if (step % (num_steps//20) == 0):  
      print("%s  %6d   %6d   %14.1f   %17.1f%%   %18.1f%%" % 
            (strftime("%H:%M:%S"), step, offset, l, accuracy(predictions, batch_labels),
            accuracy(valid_prediction.eval(feed_dict={keep_prob:1.0}), valid_labels)))
  
  print("Test accuracy: %.1f%%" % 
        accuracy(test_prediction.eval(feed_dict={keep_prob:1.0}), test_labels))