Image Classification

In this project, you'll classify images from the CIFAR-10 dataset. The dataset consists of airplanes, dogs, cats, and other objects. You'll preprocess the images, then train a convolutional neural network on all the samples. The images need to be normalized and the labels need to be one-hot encoded. You'll get to apply what you learned and build a convolutional, max pooling, dropout, and fully connected layers. At the end, you'll get to see your neural network's predictions on the sample images.

Get the Data

Run the following cell to download the CIFAR-10 dataset for python.


In [1]:
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
from urllib.request import urlretrieve
from os.path import isfile, isdir
from tqdm import tqdm
import problem_unittests as tests
import tarfile

cifar10_dataset_folder_path = 'cifar-10-batches-py'

# Use Floyd's cifar-10 dataset if present
floyd_cifar10_location = '/input/cifar-10/python.tar.gz'
if isfile(floyd_cifar10_location):
    tar_gz_path = floyd_cifar10_location
else:
    tar_gz_path = 'cifar-10-python.tar.gz'

class DLProgress(tqdm):
    last_block = 0

    def hook(self, block_num=1, block_size=1, total_size=None):
        self.total = total_size
        self.update((block_num - self.last_block) * block_size)
        self.last_block = block_num

if not isfile(tar_gz_path):
    with DLProgress(unit='B', unit_scale=True, miniters=1, desc='CIFAR-10 Dataset') as pbar:
        urlretrieve(
            'https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz',
            tar_gz_path,
            pbar.hook)

if not isdir(cifar10_dataset_folder_path):
    with tarfile.open(tar_gz_path) as tar:
        tar.extractall()
        tar.close()


tests.test_folder_path(cifar10_dataset_folder_path)


All files found!

Explore the Data

The dataset is broken into batches to prevent your machine from running out of memory. The CIFAR-10 dataset consists of 5 batches, named data_batch_1, data_batch_2, etc.. Each batch contains the labels and images that are one of the following:

  • airplane
  • automobile
  • bird
  • cat
  • deer
  • dog
  • frog
  • horse
  • ship
  • truck

Understanding a dataset is part of making predictions on the data. Play around with the code cell below by changing the batch_id and sample_id. The batch_id is the id for a batch (1-5). The sample_id is the id for a image and label pair in the batch.

Ask yourself "What are all possible labels?", "What is the range of values for the image data?", "Are the labels in order or random?". Answers to questions like these will help you preprocess the data and end up with better predictions.


In [2]:
%matplotlib inline
%config InlineBackend.figure_format = 'retina'

import helper
import numpy as np

# Explore the dataset
batch_id = 5
sample_id = 5
helper.display_stats(cifar10_dataset_folder_path, batch_id, sample_id)


Stats of batch 5:
Samples: 10000
Label Counts: {0: 1014, 1: 1014, 2: 952, 3: 1016, 4: 997, 5: 1025, 6: 980, 7: 977, 8: 1003, 9: 1022}
First 20 Labels: [1, 8, 5, 1, 5, 7, 4, 3, 8, 2, 7, 2, 0, 1, 5, 9, 6, 2, 0, 8]

Example of Image 5:
Image - Min Value: 1 Max Value: 255
Image - Shape: (32, 32, 3)
Label - Label Id: 7 Name: horse

Implement Preprocess Functions

Normalize

In the cell below, implement the normalize function to take in image data, x, and return it as a normalized Numpy array. The values should be in the range of 0 to 1, inclusive. The return object should be the same shape as x.


In [3]:
def normalize(x):
    """
    Normalize a list of sample image data in the range of 0 to 1
    : x: List of image data.  The image shape is (32, 32, 3)
    : return: Numpy array of normalize data
    """
    # TODO: Implement Function
    return x/255.


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_normalize(normalize)


Tests Passed

One-hot encode

Just like the previous code cell, you'll be implementing a function for preprocessing. This time, you'll implement the one_hot_encode function. The input, x, are a list of labels. Implement the function to return the list of labels as One-Hot encoded Numpy array. The possible values for labels are 0 to 9. The one-hot encoding function should return the same encoding for each value between each call to one_hot_encode. Make sure to save the map of encodings outside the function.

Hint: Don't reinvent the wheel.


In [4]:
label_to_index = {}
def one_hot_encode(x):
    """
    One hot encode a list of sample labels. Return a one-hot encoded vector for each label.
    : x: List of sample Labels
    : return: Numpy array of one-hot encoded labels
    """
    # TODO: Implement Function
    one_hot = []
    for label in x:
        if label not in label_to_index:
            label_to_index[label] = len(label_to_index)
        one_hot.append(np.zeros(10))
        one_hot[-1][label_to_index[label]] = 1
    return np.array(one_hot)


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_one_hot_encode(one_hot_encode)


Tests Passed

Randomize Data

As you saw from exploring the data above, the order of the samples are randomized. It doesn't hurt to randomize it again, but you don't need to for this dataset.

Preprocess all the data and save it

Running the code cell below will preprocess all the CIFAR-10 data and save it to file. The code below also uses 10% of the training data for validation.


In [5]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
# Preprocess Training, Validation, and Testing Data
helper.preprocess_and_save_data(cifar10_dataset_folder_path, normalize, one_hot_encode)

Check Point

This is your first checkpoint. If you ever decide to come back to this notebook or have to restart the notebook, you can start from here. The preprocessed data has been saved to disk.


In [6]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import pickle
import problem_unittests as tests
import helper

# Load the Preprocessed Validation data
valid_features, valid_labels = pickle.load(open('preprocess_validation.p', mode='rb'))

Build the network

For the neural network, you'll build each layer into a function. Most of the code you've seen has been outside of functions. To test your code more thoroughly, we require that you put each layer in a function. This allows us to give you better feedback and test for simple mistakes using our unittests before you submit your project.

If you're finding it hard to dedicate enough time for this course a week, we've provided a small shortcut to this part of the project. In the next couple of problems, you'll have the option to use TensorFlow Layers or TensorFlow Layers (contrib) to build each layer, except "Convolutional & Max Pooling" layer. TF Layers is similar to Keras's and TFLearn's abstraction to layers, so it's easy to pickup.

If you would like to get the most of this course, try to solve all the problems without TF Layers. Let's begin!

Input

The neural network needs to read the image data, one-hot encoded labels, and dropout keep probability. Implement the following functions

  • Implement neural_net_image_input
    • Return a TF Placeholder
    • Set the shape using image_shape with batch size set to None.
    • Name the TensorFlow placeholder "x" using the TensorFlow name parameter in the TF Placeholder.
  • Implement neural_net_label_input
    • Return a TF Placeholder
    • Set the shape using n_classes with batch size set to None.
    • Name the TensorFlow placeholder "y" using the TensorFlow name parameter in the TF Placeholder.
  • Implement neural_net_keep_prob_input
    • Return a TF Placeholder for dropout keep probability.
    • Name the TensorFlow placeholder "keep_prob" using the TensorFlow name parameter in the TF Placeholder.

These names will be used at the end of the project to load your saved model.

Note: None for shapes in TensorFlow allow for a dynamic size.


In [7]:
import tensorflow as tf

def neural_net_image_input(image_shape):
    """
    Return a Tensor for a batch of image input
    : image_shape: Shape of the images
    : return: Tensor for image input.
    """
    # TODO: Implement Function
    return tf.placeholder(shape=[None, *image_shape], name='x', dtype=tf.float32)


def neural_net_label_input(n_classes):
    """
    Return a Tensor for a batch of label input
    : n_classes: Number of classes
    : return: Tensor for label input.
    """
    # TODO: Implement Function
    return tf.placeholder(shape=[None, n_classes], name='y', dtype=tf.float32)


def neural_net_keep_prob_input():
    """
    Return a Tensor for keep probability
    : return: Tensor for keep probability.
    """
    # TODO: Implement Function
    return tf.placeholder(name="keep_prob", dtype=tf.float32)


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tf.reset_default_graph()
tests.test_nn_image_inputs(neural_net_image_input)
tests.test_nn_label_inputs(neural_net_label_input)
tests.test_nn_keep_prob_inputs(neural_net_keep_prob_input)


Image Input Tests Passed.
Label Input Tests Passed.
Keep Prob Tests Passed.

Convolution and Max Pooling Layer

Convolution layers have a lot of success with images. For this code cell, you should implement the function conv2d_maxpool to apply convolution then max pooling:

  • Create the weight and bias using conv_ksize, conv_num_outputs and the shape of x_tensor.
  • Apply a convolution to x_tensor using weight and conv_strides.
    • We recommend you use same padding, but you're welcome to use any padding.
  • Add bias
  • Add a nonlinear activation to the convolution.
  • Apply Max Pooling using pool_ksize and pool_strides.
    • We recommend you use same padding, but you're welcome to use any padding.

Note: You can't use TensorFlow Layers or TensorFlow Layers (contrib) for this layer. You're free to use any TensorFlow package for all the other layers.


In [8]:
def conv2d_maxpool(x_tensor, conv_num_outputs, conv_ksize, conv_strides, pool_ksize, pool_strides):
    """
    Apply convolution then max pooling to x_tensor
    :param x_tensor: TensorFlow Tensor
    :param conv_num_outputs: Number of outputs for the convolutional layer
    :param conv_strides: Stride 2-D Tuple for convolution
    :param pool_ksize: kernal size 2-D Tuple for pool
    :param pool_strides: Stride 2-D Tuple for pool
    : return: A tensor that represents convolution and max pooling of x_tensor
    """
    # TODO: Implement Function
    depth = x_tensor.get_shape().as_list()[-1]
    F_W = tf.Variable(tf.truncated_normal((*conv_ksize, depth, conv_num_outputs), stddev=0.1))
    F_b = tf.Variable(tf.zeros(conv_num_outputs))
    strides = [1, *conv_strides, 1]
    padding = 'SAME'
    conv_layer = tf.nn.conv2d(x_tensor, F_W, strides=strides, padding=padding)
    conv_layer = tf.nn.bias_add(conv_layer, F_b)
    
    conv_layer = tf.nn.relu(conv_layer)
    conv_layer = tf.nn.max_pool(
        conv_layer,
        ksize=[1, *pool_ksize, 1],
        strides=[1, *pool_strides, 1],
        padding='SAME')

    return conv_layer

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_con_pool(conv2d_maxpool)


Tests Passed

Flatten Layer

Implement the flatten function to change the dimension of x_tensor from a 4-D tensor to a 2-D tensor. The output should be the shape (Batch Size, Flattened Image Size). You can use TensorFlow Layers or TensorFlow Layers (contrib) for this layer.


In [9]:
def flatten(x_tensor):
    """
    Flatten x_tensor to (Batch Size, Flattened Image Size)
    : x_tensor: A tensor of size (Batch Size, ...), where ... are the image dimensions.
    : return: A tensor of size (Batch Size, Flattened Image Size).
    """
    # TODO: Implement Function
    dim = x_tensor.get_shape().as_list()
    return tf.reshape(x_tensor, shape=[-1, dim[1]*dim[2]*dim[3]])


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_flatten(flatten)


Tests Passed

Fully-Connected Layer

Implement the fully_conn function to apply a fully connected layer to x_tensor with the shape (Batch Size, num_outputs). You can use TensorFlow Layers or TensorFlow Layers (contrib) for this layer.


In [10]:
def fully_conn(x_tensor, num_outputs):
    """
    Apply a fully connected layer to x_tensor using weight and bias
    : x_tensor: A 2-D tensor where the first dimension is batch size.
    : num_outputs: The number of output that the new tensor should be.
    : return: A 2-D tensor where the second dimension is num_outputs.
    """
    # TODO: Implement Function
    dim = x_tensor.get_shape().as_list()
    W = tf.Variable(tf.truncated_normal((dim[1], num_outputs), stddev=0.1))
    b = tf.Variable(tf.zeros(num_outputs))
    return tf.nn.relu(tf.add(tf.matmul(x_tensor, W), b))

#     or the easier version
#     return tf.layers.dense(x_tensor, num_outputs)


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_fully_conn(fully_conn)


Tests Passed

Output Layer

Implement the output function to apply a fully connected layer to x_tensor with the shape (Batch Size, num_outputs). You can use TensorFlow Layers or TensorFlow Layers (contrib) for this layer.

Note: Activation, softmax, or cross entropy shouldn't be applied to this.


In [11]:
def output(x_tensor, num_outputs):
    """
    Apply a output layer to x_tensor using weight and bias
    : x_tensor: A 2-D tensor where the first dimension is batch size.
    : num_outputs: The number of output that the new tensor should be.
    : return: A 2-D tensor where the second dimension is num_outputs.
    """
    # TODO: Implement Function
    dim = x_tensor.get_shape().as_list()
    W = tf.Variable(tf.truncated_normal((dim[1], num_outputs)))
    b = tf.Variable(tf.zeros(num_outputs))
    return tf.add(tf.matmul(x_tensor, W), b)


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_output(output)


Tests Passed

Create Convolutional Model

Implement the function conv_net to create a convolutional neural network model. The function takes in a batch of images, x, and outputs logits. Use the layers you created above to create this model:

  • Apply 1, 2, or 3 Convolution and Max Pool layers
  • Apply a Flatten Layer
  • Apply 1, 2, or 3 Fully Connected Layers
  • Apply an Output Layer
  • Return the output
  • Apply TensorFlow's Dropout to one or more layers in the model using keep_prob.

In [12]:
def conv_net(x, keep_prob):
    """
    Create a convolutional neural network model
    : x: Placeholder tensor that holds image data.
    : keep_prob: Placeholder tensor that hold dropout keep probability.
    : return: Tensor that represents logits
    """
    # TODO: Apply 1, 2, or 3 Convolution and Max Pool layers
    #    Play around with different number of outputs, kernel size and stride
    # Function Definition from Above:
    #    conv2d_maxpool(x_tensor, conv_num_outputs, conv_ksize, conv_strides, pool_ksize, pool_strides)
    
    # First conv layer
    conv_num_outputs = 32
    conv_ksize = [3, 3]
    conv_strides = [1, 1]
    pool_ksize = [2, 2]
    pool_strides = [2, 2]
    x_tensor = conv2d_maxpool(x, conv_num_outputs, conv_ksize, conv_strides, pool_ksize, pool_strides)

    # Second conv layer
    conv_num_outputs = 64
    conv_ksize = [3, 3]
    conv_strides = [1, 1]
    pool_ksize = [2, 2]
    pool_strides = [2, 2]
    x_tensor = conv2d_maxpool(x_tensor, conv_num_outputs, conv_ksize, conv_strides, pool_ksize, pool_strides)

    # Third conv layer
    conv_num_outputs = 128
    conv_ksize = [3, 3]
    conv_strides = [1, 1]
    pool_ksize = [2, 2]
    pool_strides = [2, 2]
    x_tensor = conv2d_maxpool(x_tensor, conv_num_outputs, conv_ksize, conv_strides, pool_ksize, pool_strides)
    
    # Apply a Flatten Layer
    x_tensor = flatten(x_tensor)
    
    # Apply 1, 2, or 3 Fully Connected Layers
    #    Play around with different number of outputs
    
    # First fully connected layer with dropout
    x_tensor = fully_conn(x_tensor, 64)
    x_tensor = tf.nn.dropout(x_tensor, keep_prob)
    
    # Apply an Output Layer
    #    Set this to the number of classes    
    x_tensor = output(x_tensor, len(label_to_index))

    return x_tensor


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""

##############################
## Build the Neural Network ##
##############################

# Remove previous weights, bias, inputs, etc..
tf.reset_default_graph()

# Inputs
x = neural_net_image_input((32, 32, 3))
y = neural_net_label_input(10)
keep_prob = neural_net_keep_prob_input()

# Model
logits = conv_net(x, keep_prob)

# Name logits Tensor, so that is can be loaded from disk after training
logits = tf.identity(logits, name='logits')

# Loss and Optimizer
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logits, labels=y))
optimizer = tf.train.AdamOptimizer().minimize(cost)

# Accuracy
correct_pred = tf.equal(tf.argmax(logits, 1), tf.argmax(y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32), name='accuracy')

tests.test_conv_net(conv_net)


Neural Network Built!

Train the Neural Network

Single Optimization

Implement the function train_neural_network to do a single optimization. The optimization should use optimizer to optimize in session with a feed_dict of the following:

  • x for image input
  • y for labels
  • keep_prob for keep probability for dropout

This function will be called for each batch, so tf.global_variables_initializer() has already been called.

Note: Nothing needs to be returned. This function is only optimizing the neural network.


In [13]:
def train_neural_network(session, optimizer, keep_probability, feature_batch, label_batch):
    """
    Optimize the session on a batch of images and labels
    : session: Current TensorFlow session
    : optimizer: TensorFlow optimizer function
    : keep_probability: keep probability
    : feature_batch: Batch of Numpy image data
    : label_batch: Batch of Numpy label data
    """
    # TODO: Implement Function
#     image_input = neural_net_image_input(feature_batch.shape)
#     label_input = neural_net_label_input(len(label_batch))
#     keep_prob = neural_net_keep_prob_input()
    session.run(
        optimizer, 
        feed_dict={
            x: feature_batch, 
            y: label_batch, 
            keep_prob: keep_probability
        }
    )


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_train_nn(train_neural_network)


Tests Passed

Show Stats

Implement the function print_stats to print loss and validation accuracy. Use the global variables valid_features and valid_labels to calculate validation accuracy. Use a keep probability of 1.0 to calculate the loss and validation accuracy.


In [14]:
def print_stats(session, feature_batch, label_batch, cost, accuracy):
    """
    Print information about loss and validation accuracy
    : session: Current TensorFlow session
    : feature_batch: Batch of Numpy image data
    : label_batch: Batch of Numpy label data
    : cost: TensorFlow cost function
    : accuracy: TensorFlow accuracy function
    """
    # TODO: Implement Function
    batch_accuracy = session.run(
        accuracy, 
        feed_dict={
            x: feature_batch,
            y: label_batch,
            keep_prob: 1.
        }
    )
    batch_cost = session.run(
        cost, 
        feed_dict={
            x: feature_batch,
            y: label_batch,
            keep_prob: 1.
        }
    )
    
    valid_accuracy = session.run(
        accuracy, 
        feed_dict={
            x: valid_features,
            y: valid_labels,
            keep_prob: 1.
        }
    )
    valid_cost = session.run(
        cost, 
        feed_dict={
            x: valid_features,
            y: valid_labels,
            keep_prob: 1.
        }
    )
    print(
        "batch acc: {:.4f}, batch cost: {:.4f}, valid acc: {:.4f}, valid cost: {:.4f}".format(
            batch_accuracy, batch_cost, valid_accuracy, valid_cost
        )
    )

Hyperparameters

Tune the following parameters:

  • Set epochs to the number of iterations until the network stops learning or start overfitting
  • Set batch_size to the highest number that your machine has memory for. Most people set them to common sizes of memory:
    • 64
    • 128
    • 256
    • ...
  • Set keep_probability to the probability of keeping a node using dropout

In [15]:
# TODO: Tune Parameters
epochs = 25
batch_size = 512
keep_probability = 0.8

Train on a Single CIFAR-10 Batch

Instead of training the neural network on all the CIFAR-10 batches of data, let's use a single batch. This should save time while you iterate on the model to get a better accuracy. Once the final validation accuracy is 50% or greater, run the model on all the data in the next section.


In [16]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
print('Checking the Training on a Single Batch...')
with tf.Session() as sess:
    # Initializing the variables
    sess.run(tf.global_variables_initializer())
    
    # Training cycle
    for epoch in range(epochs):
        batch_i = 1
        for batch_features, batch_labels in helper.load_preprocess_training_batch(batch_i, batch_size):
            train_neural_network(sess, optimizer, keep_probability, batch_features, batch_labels)
        print('Epoch {:>2}, CIFAR-10 Batch {}:  '.format(epoch + 1, batch_i), end='')
        print_stats(sess, batch_features, batch_labels, cost, accuracy)


Checking the Training on a Single Batch...
Epoch  1, CIFAR-10 Batch 1:  batch acc: 0.1554, batch cost: 2.2374, valid acc: 0.1542, valid cost: 2.2448
Epoch  2, CIFAR-10 Batch 1:  batch acc: 0.2128, batch cost: 2.1670, valid acc: 0.2072, valid cost: 2.1557
Epoch  3, CIFAR-10 Batch 1:  batch acc: 0.2804, batch cost: 2.0350, valid acc: 0.2644, valid cost: 2.0186
Epoch  4, CIFAR-10 Batch 1:  batch acc: 0.3041, batch cost: 1.9552, valid acc: 0.3132, valid cost: 1.9335
Epoch  5, CIFAR-10 Batch 1:  batch acc: 0.3378, batch cost: 1.8668, valid acc: 0.3548, valid cost: 1.8464
Epoch  6, CIFAR-10 Batch 1:  batch acc: 0.3750, batch cost: 1.7721, valid acc: 0.3890, valid cost: 1.7834
Epoch  7, CIFAR-10 Batch 1:  batch acc: 0.4088, batch cost: 1.6958, valid acc: 0.4002, valid cost: 1.7423
Epoch  8, CIFAR-10 Batch 1:  batch acc: 0.4459, batch cost: 1.6094, valid acc: 0.4286, valid cost: 1.6734
Epoch  9, CIFAR-10 Batch 1:  batch acc: 0.4865, batch cost: 1.5105, valid acc: 0.4330, valid cost: 1.6242
Epoch 10, CIFAR-10 Batch 1:  batch acc: 0.5236, batch cost: 1.4403, valid acc: 0.4620, valid cost: 1.5581
Epoch 11, CIFAR-10 Batch 1:  batch acc: 0.5405, batch cost: 1.3478, valid acc: 0.4804, valid cost: 1.5039
Epoch 12, CIFAR-10 Batch 1:  batch acc: 0.5473, batch cost: 1.2910, valid acc: 0.4862, valid cost: 1.4807
Epoch 13, CIFAR-10 Batch 1:  batch acc: 0.5912, batch cost: 1.2382, valid acc: 0.4958, valid cost: 1.4580
Epoch 14, CIFAR-10 Batch 1:  batch acc: 0.6284, batch cost: 1.1702, valid acc: 0.5072, valid cost: 1.4208
Epoch 15, CIFAR-10 Batch 1:  batch acc: 0.6419, batch cost: 1.1431, valid acc: 0.5084, valid cost: 1.4302
Epoch 16, CIFAR-10 Batch 1:  batch acc: 0.6554, batch cost: 1.0792, valid acc: 0.5206, valid cost: 1.3929
Epoch 17, CIFAR-10 Batch 1:  batch acc: 0.6588, batch cost: 1.0473, valid acc: 0.5254, valid cost: 1.3708
Epoch 18, CIFAR-10 Batch 1:  batch acc: 0.6926, batch cost: 1.0389, valid acc: 0.5238, valid cost: 1.3767
Epoch 19, CIFAR-10 Batch 1:  batch acc: 0.7095, batch cost: 0.9672, valid acc: 0.5352, valid cost: 1.3257
Epoch 20, CIFAR-10 Batch 1:  batch acc: 0.7230, batch cost: 0.9339, valid acc: 0.5374, valid cost: 1.3237
Epoch 21, CIFAR-10 Batch 1:  batch acc: 0.7365, batch cost: 0.8768, valid acc: 0.5454, valid cost: 1.3087
Epoch 22, CIFAR-10 Batch 1:  batch acc: 0.7331, batch cost: 0.8383, valid acc: 0.5466, valid cost: 1.2973
Epoch 23, CIFAR-10 Batch 1:  batch acc: 0.7669, batch cost: 0.8066, valid acc: 0.5484, valid cost: 1.2881
Epoch 24, CIFAR-10 Batch 1:  batch acc: 0.7635, batch cost: 0.7613, valid acc: 0.5538, valid cost: 1.2664
Epoch 25, CIFAR-10 Batch 1:  batch acc: 0.7939, batch cost: 0.7374, valid acc: 0.5478, valid cost: 1.2761

Fully Train the Model

Now that you got a good accuracy with a single CIFAR-10 batch, try it with all five batches.


In [17]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
save_model_path = './image_classification'

print('Training...')
with tf.Session() as sess:
    # Initializing the variables
    sess.run(tf.global_variables_initializer())
    
    # Training cycle
    for epoch in range(epochs):
        # Loop over all batches
        n_batches = 5
        for batch_i in range(1, n_batches + 1):
            for batch_features, batch_labels in helper.load_preprocess_training_batch(batch_i, batch_size):
                train_neural_network(sess, optimizer, keep_probability, batch_features, batch_labels)
            print('Epoch {:>2}, CIFAR-10 Batch {}:  '.format(epoch + 1, batch_i), end='')
            print_stats(sess, batch_features, batch_labels, cost, accuracy)
            
    # Save Model
    saver = tf.train.Saver()
    save_path = saver.save(sess, save_model_path)


Training...
Epoch  1, CIFAR-10 Batch 1:  batch acc: 0.1318, batch cost: 2.2811, valid acc: 0.1220, valid cost: 2.2870
Epoch  1, CIFAR-10 Batch 2:  batch acc: 0.1115, batch cost: 2.2732, valid acc: 0.1384, valid cost: 2.2548
Epoch  1, CIFAR-10 Batch 3:  batch acc: 0.2162, batch cost: 2.1584, valid acc: 0.2148, valid cost: 2.1571
Epoch  1, CIFAR-10 Batch 4:  batch acc: 0.3007, batch cost: 1.9760, valid acc: 0.2742, valid cost: 2.0449
Epoch  1, CIFAR-10 Batch 5:  batch acc: 0.3682, batch cost: 1.9331, valid acc: 0.3104, valid cost: 1.9671
Epoch  2, CIFAR-10 Batch 1:  batch acc: 0.3209, batch cost: 1.9236, valid acc: 0.3240, valid cost: 1.9092
Epoch  2, CIFAR-10 Batch 2:  batch acc: 0.3209, batch cost: 1.8746, valid acc: 0.3450, valid cost: 1.8450
Epoch  2, CIFAR-10 Batch 3:  batch acc: 0.4020, batch cost: 1.7507, valid acc: 0.3656, valid cost: 1.7930
Epoch  2, CIFAR-10 Batch 4:  batch acc: 0.4223, batch cost: 1.6844, valid acc: 0.3990, valid cost: 1.7570
Epoch  2, CIFAR-10 Batch 5:  batch acc: 0.4561, batch cost: 1.5814, valid acc: 0.4078, valid cost: 1.6643
Epoch  3, CIFAR-10 Batch 1:  batch acc: 0.4561, batch cost: 1.6379, valid acc: 0.4232, valid cost: 1.6598
Epoch  3, CIFAR-10 Batch 2:  batch acc: 0.4865, batch cost: 1.5398, valid acc: 0.4236, valid cost: 1.5836
Epoch  3, CIFAR-10 Batch 3:  batch acc: 0.5372, batch cost: 1.4362, valid acc: 0.4394, valid cost: 1.5641
Epoch  3, CIFAR-10 Batch 4:  batch acc: 0.5000, batch cost: 1.4176, valid acc: 0.4746, valid cost: 1.5035
Epoch  3, CIFAR-10 Batch 5:  batch acc: 0.5473, batch cost: 1.3618, valid acc: 0.4814, valid cost: 1.4474
Epoch  4, CIFAR-10 Batch 1:  batch acc: 0.5439, batch cost: 1.4155, valid acc: 0.4920, valid cost: 1.4338
Epoch  4, CIFAR-10 Batch 2:  batch acc: 0.5709, batch cost: 1.3284, valid acc: 0.5040, valid cost: 1.3881
Epoch  4, CIFAR-10 Batch 3:  batch acc: 0.6081, batch cost: 1.2163, valid acc: 0.5042, valid cost: 1.3930
Epoch  4, CIFAR-10 Batch 4:  batch acc: 0.5203, batch cost: 1.2567, valid acc: 0.5176, valid cost: 1.3854
Epoch  4, CIFAR-10 Batch 5:  batch acc: 0.5912, batch cost: 1.2143, valid acc: 0.5246, valid cost: 1.3249
Epoch  5, CIFAR-10 Batch 1:  batch acc: 0.5946, batch cost: 1.2879, valid acc: 0.5402, valid cost: 1.3257
Epoch  5, CIFAR-10 Batch 2:  batch acc: 0.6284, batch cost: 1.1878, valid acc: 0.5318, valid cost: 1.3216
Epoch  5, CIFAR-10 Batch 3:  batch acc: 0.6351, batch cost: 1.0733, valid acc: 0.5508, valid cost: 1.2764
Epoch  5, CIFAR-10 Batch 4:  batch acc: 0.5574, batch cost: 1.1119, valid acc: 0.5560, valid cost: 1.2721
Epoch  5, CIFAR-10 Batch 5:  batch acc: 0.6182, batch cost: 1.1529, valid acc: 0.5532, valid cost: 1.2781
Epoch  6, CIFAR-10 Batch 1:  batch acc: 0.6182, batch cost: 1.1678, valid acc: 0.5668, valid cost: 1.2395
Epoch  6, CIFAR-10 Batch 2:  batch acc: 0.6791, batch cost: 1.0909, valid acc: 0.5688, valid cost: 1.2392
Epoch  6, CIFAR-10 Batch 3:  batch acc: 0.6757, batch cost: 0.9888, valid acc: 0.5650, valid cost: 1.2227
Epoch  6, CIFAR-10 Batch 4:  batch acc: 0.6453, batch cost: 1.0159, valid acc: 0.5766, valid cost: 1.2125
Epoch  6, CIFAR-10 Batch 5:  batch acc: 0.6655, batch cost: 1.0451, valid acc: 0.5828, valid cost: 1.1856
Epoch  7, CIFAR-10 Batch 1:  batch acc: 0.6520, batch cost: 1.0494, valid acc: 0.5860, valid cost: 1.1666
Epoch  7, CIFAR-10 Batch 2:  batch acc: 0.7095, batch cost: 0.9879, valid acc: 0.5924, valid cost: 1.1794
Epoch  7, CIFAR-10 Batch 3:  batch acc: 0.7027, batch cost: 0.9430, valid acc: 0.5830, valid cost: 1.1929
Epoch  7, CIFAR-10 Batch 4:  batch acc: 0.6791, batch cost: 0.9563, valid acc: 0.5978, valid cost: 1.1613
Epoch  7, CIFAR-10 Batch 5:  batch acc: 0.6757, batch cost: 0.9687, valid acc: 0.6062, valid cost: 1.1281
Epoch  8, CIFAR-10 Batch 1:  batch acc: 0.6588, batch cost: 0.9843, valid acc: 0.6004, valid cost: 1.1388
Epoch  8, CIFAR-10 Batch 2:  batch acc: 0.7196, batch cost: 0.9326, valid acc: 0.6060, valid cost: 1.1305
Epoch  8, CIFAR-10 Batch 3:  batch acc: 0.7061, batch cost: 0.8942, valid acc: 0.5930, valid cost: 1.1604
Epoch  8, CIFAR-10 Batch 4:  batch acc: 0.7196, batch cost: 0.8656, valid acc: 0.6128, valid cost: 1.0981
Epoch  8, CIFAR-10 Batch 5:  batch acc: 0.6993, batch cost: 0.9012, valid acc: 0.6242, valid cost: 1.0848
Epoch  9, CIFAR-10 Batch 1:  batch acc: 0.7196, batch cost: 0.9048, valid acc: 0.6238, valid cost: 1.0734
Epoch  9, CIFAR-10 Batch 2:  batch acc: 0.7162, batch cost: 0.8951, valid acc: 0.6048, valid cost: 1.1349
Epoch  9, CIFAR-10 Batch 3:  batch acc: 0.7399, batch cost: 0.8254, valid acc: 0.6022, valid cost: 1.1238
Epoch  9, CIFAR-10 Batch 4:  batch acc: 0.7196, batch cost: 0.8403, valid acc: 0.6184, valid cost: 1.1030
Epoch  9, CIFAR-10 Batch 5:  batch acc: 0.7196, batch cost: 0.8244, valid acc: 0.6282, valid cost: 1.0450
Epoch 10, CIFAR-10 Batch 1:  batch acc: 0.7162, batch cost: 0.8519, valid acc: 0.6246, valid cost: 1.0630
Epoch 10, CIFAR-10 Batch 2:  batch acc: 0.7399, batch cost: 0.8243, valid acc: 0.6070, valid cost: 1.1076
Epoch 10, CIFAR-10 Batch 3:  batch acc: 0.7669, batch cost: 0.7373, valid acc: 0.6306, valid cost: 1.0500
Epoch 10, CIFAR-10 Batch 4:  batch acc: 0.7635, batch cost: 0.8008, valid acc: 0.6336, valid cost: 1.0541
Epoch 10, CIFAR-10 Batch 5:  batch acc: 0.7432, batch cost: 0.8182, valid acc: 0.6416, valid cost: 1.0332
Epoch 11, CIFAR-10 Batch 1:  batch acc: 0.7230, batch cost: 0.8147, valid acc: 0.6540, valid cost: 1.0013
Epoch 11, CIFAR-10 Batch 2:  batch acc: 0.7669, batch cost: 0.7608, valid acc: 0.6204, valid cost: 1.0948
Epoch 11, CIFAR-10 Batch 3:  batch acc: 0.7973, batch cost: 0.7016, valid acc: 0.6416, valid cost: 1.0342
Epoch 11, CIFAR-10 Batch 4:  batch acc: 0.7804, batch cost: 0.7036, valid acc: 0.6546, valid cost: 0.9894
Epoch 11, CIFAR-10 Batch 5:  batch acc: 0.7601, batch cost: 0.7502, valid acc: 0.6530, valid cost: 1.0091
Epoch 12, CIFAR-10 Batch 1:  batch acc: 0.7432, batch cost: 0.7386, valid acc: 0.6596, valid cost: 0.9717
Epoch 12, CIFAR-10 Batch 2:  batch acc: 0.7838, batch cost: 0.7046, valid acc: 0.6312, valid cost: 1.0601
Epoch 12, CIFAR-10 Batch 3:  batch acc: 0.8007, batch cost: 0.6370, valid acc: 0.6552, valid cost: 0.9989
Epoch 12, CIFAR-10 Batch 4:  batch acc: 0.7838, batch cost: 0.6891, valid acc: 0.6478, valid cost: 0.9908
Epoch 12, CIFAR-10 Batch 5:  batch acc: 0.7872, batch cost: 0.6751, valid acc: 0.6574, valid cost: 0.9652
Epoch 13, CIFAR-10 Batch 1:  batch acc: 0.7872, batch cost: 0.7202, valid acc: 0.6598, valid cost: 0.9977
Epoch 13, CIFAR-10 Batch 2:  batch acc: 0.7973, batch cost: 0.6707, valid acc: 0.6510, valid cost: 1.0097
Epoch 13, CIFAR-10 Batch 3:  batch acc: 0.8041, batch cost: 0.6169, valid acc: 0.6620, valid cost: 0.9889
Epoch 13, CIFAR-10 Batch 4:  batch acc: 0.8108, batch cost: 0.6174, valid acc: 0.6662, valid cost: 0.9510
Epoch 13, CIFAR-10 Batch 5:  batch acc: 0.7905, batch cost: 0.6381, valid acc: 0.6494, valid cost: 0.9839
Epoch 14, CIFAR-10 Batch 1:  batch acc: 0.7804, batch cost: 0.6724, valid acc: 0.6628, valid cost: 0.9702
Epoch 14, CIFAR-10 Batch 2:  batch acc: 0.7736, batch cost: 0.6746, valid acc: 0.6498, valid cost: 1.0090
Epoch 14, CIFAR-10 Batch 3:  batch acc: 0.8243, batch cost: 0.5789, valid acc: 0.6676, valid cost: 0.9862
Epoch 14, CIFAR-10 Batch 4:  batch acc: 0.7973, batch cost: 0.6091, valid acc: 0.6616, valid cost: 0.9715
Epoch 14, CIFAR-10 Batch 5:  batch acc: 0.8176, batch cost: 0.6042, valid acc: 0.6746, valid cost: 0.9443
Epoch 15, CIFAR-10 Batch 1:  batch acc: 0.8108, batch cost: 0.6273, valid acc: 0.6754, valid cost: 0.9398
Epoch 15, CIFAR-10 Batch 2:  batch acc: 0.8277, batch cost: 0.5794, valid acc: 0.6666, valid cost: 0.9650
Epoch 15, CIFAR-10 Batch 3:  batch acc: 0.8142, batch cost: 0.5467, valid acc: 0.6640, valid cost: 0.9776
Epoch 15, CIFAR-10 Batch 4:  batch acc: 0.7905, batch cost: 0.6184, valid acc: 0.6564, valid cost: 0.9788
Epoch 15, CIFAR-10 Batch 5:  batch acc: 0.8243, batch cost: 0.5576, valid acc: 0.6792, valid cost: 0.9201
Epoch 16, CIFAR-10 Batch 1:  batch acc: 0.8108, batch cost: 0.6187, valid acc: 0.6728, valid cost: 0.9407
Epoch 16, CIFAR-10 Batch 2:  batch acc: 0.8345, batch cost: 0.5474, valid acc: 0.6792, valid cost: 0.9215
Epoch 16, CIFAR-10 Batch 3:  batch acc: 0.8412, batch cost: 0.5255, valid acc: 0.6620, valid cost: 0.9947
Epoch 16, CIFAR-10 Batch 4:  batch acc: 0.8209, batch cost: 0.5341, valid acc: 0.6820, valid cost: 0.9258
Epoch 16, CIFAR-10 Batch 5:  batch acc: 0.8311, batch cost: 0.5238, valid acc: 0.6816, valid cost: 0.9184
Epoch 17, CIFAR-10 Batch 1:  batch acc: 0.8277, batch cost: 0.5396, valid acc: 0.6844, valid cost: 0.9174
Epoch 17, CIFAR-10 Batch 2:  batch acc: 0.8615, batch cost: 0.5054, valid acc: 0.6872, valid cost: 0.8944
Epoch 17, CIFAR-10 Batch 3:  batch acc: 0.8480, batch cost: 0.4828, valid acc: 0.6668, valid cost: 0.9655
Epoch 17, CIFAR-10 Batch 4:  batch acc: 0.8412, batch cost: 0.4843, valid acc: 0.6814, valid cost: 0.9111
Epoch 17, CIFAR-10 Batch 5:  batch acc: 0.8818, batch cost: 0.4598, valid acc: 0.6918, valid cost: 0.8984
Epoch 18, CIFAR-10 Batch 1:  batch acc: 0.8378, batch cost: 0.5221, valid acc: 0.6866, valid cost: 0.9119
Epoch 18, CIFAR-10 Batch 2:  batch acc: 0.8682, batch cost: 0.4899, valid acc: 0.6886, valid cost: 0.9006
Epoch 18, CIFAR-10 Batch 3:  batch acc: 0.8682, batch cost: 0.4322, valid acc: 0.6812, valid cost: 0.9305
Epoch 18, CIFAR-10 Batch 4:  batch acc: 0.8480, batch cost: 0.4522, valid acc: 0.6834, valid cost: 0.9214
Epoch 18, CIFAR-10 Batch 5:  batch acc: 0.8615, batch cost: 0.4283, valid acc: 0.6960, valid cost: 0.8866
Epoch 19, CIFAR-10 Batch 1:  batch acc: 0.8514, batch cost: 0.5004, valid acc: 0.6872, valid cost: 0.9147
Epoch 19, CIFAR-10 Batch 2:  batch acc: 0.8716, batch cost: 0.4777, valid acc: 0.6888, valid cost: 0.9038
Epoch 19, CIFAR-10 Batch 3:  batch acc: 0.8615, batch cost: 0.4280, valid acc: 0.6860, valid cost: 0.9195
Epoch 19, CIFAR-10 Batch 4:  batch acc: 0.8682, batch cost: 0.4328, valid acc: 0.6874, valid cost: 0.9045
Epoch 19, CIFAR-10 Batch 5:  batch acc: 0.8818, batch cost: 0.4199, valid acc: 0.6898, valid cost: 0.9021
Epoch 20, CIFAR-10 Batch 1:  batch acc: 0.8547, batch cost: 0.4892, valid acc: 0.6858, valid cost: 0.9088
Epoch 20, CIFAR-10 Batch 2:  batch acc: 0.8649, batch cost: 0.4489, valid acc: 0.6912, valid cost: 0.9098
Epoch 20, CIFAR-10 Batch 3:  batch acc: 0.8851, batch cost: 0.4130, valid acc: 0.6804, valid cost: 0.9331
Epoch 20, CIFAR-10 Batch 4:  batch acc: 0.8682, batch cost: 0.4163, valid acc: 0.6912, valid cost: 0.9030
Epoch 20, CIFAR-10 Batch 5:  batch acc: 0.9020, batch cost: 0.3806, valid acc: 0.6936, valid cost: 0.9004
Epoch 21, CIFAR-10 Batch 1:  batch acc: 0.8818, batch cost: 0.4660, valid acc: 0.6924, valid cost: 0.9076
Epoch 21, CIFAR-10 Batch 2:  batch acc: 0.8818, batch cost: 0.4247, valid acc: 0.6882, valid cost: 0.9222
Epoch 21, CIFAR-10 Batch 3:  batch acc: 0.8986, batch cost: 0.3790, valid acc: 0.6838, valid cost: 0.9212
Epoch 21, CIFAR-10 Batch 4:  batch acc: 0.8750, batch cost: 0.3958, valid acc: 0.6908, valid cost: 0.9138
Epoch 21, CIFAR-10 Batch 5:  batch acc: 0.9189, batch cost: 0.3333, valid acc: 0.7048, valid cost: 0.8782
Epoch 22, CIFAR-10 Batch 1:  batch acc: 0.8682, batch cost: 0.4475, valid acc: 0.6880, valid cost: 0.9334
Epoch 22, CIFAR-10 Batch 2:  batch acc: 0.8784, batch cost: 0.4072, valid acc: 0.6886, valid cost: 0.9196
Epoch 22, CIFAR-10 Batch 3:  batch acc: 0.9088, batch cost: 0.3643, valid acc: 0.6882, valid cost: 0.9201
Epoch 22, CIFAR-10 Batch 4:  batch acc: 0.8885, batch cost: 0.3889, valid acc: 0.6956, valid cost: 0.8964
Epoch 22, CIFAR-10 Batch 5:  batch acc: 0.9020, batch cost: 0.3460, valid acc: 0.7056, valid cost: 0.8855
Epoch 23, CIFAR-10 Batch 1:  batch acc: 0.8851, batch cost: 0.4219, valid acc: 0.6968, valid cost: 0.9056
Epoch 23, CIFAR-10 Batch 2:  batch acc: 0.8716, batch cost: 0.3840, valid acc: 0.6906, valid cost: 0.9242
Epoch 23, CIFAR-10 Batch 3:  batch acc: 0.9122, batch cost: 0.3409, valid acc: 0.6880, valid cost: 0.9367
Epoch 23, CIFAR-10 Batch 4:  batch acc: 0.8818, batch cost: 0.4125, valid acc: 0.6714, valid cost: 0.9692
Epoch 23, CIFAR-10 Batch 5:  batch acc: 0.9189, batch cost: 0.3433, valid acc: 0.6968, valid cost: 0.9158
Epoch 24, CIFAR-10 Batch 1:  batch acc: 0.8986, batch cost: 0.3913, valid acc: 0.6976, valid cost: 0.9051
Epoch 24, CIFAR-10 Batch 2:  batch acc: 0.8784, batch cost: 0.3899, valid acc: 0.6868, valid cost: 0.9545
Epoch 24, CIFAR-10 Batch 3:  batch acc: 0.9257, batch cost: 0.3276, valid acc: 0.6868, valid cost: 0.9413
Epoch 24, CIFAR-10 Batch 4:  batch acc: 0.8682, batch cost: 0.3819, valid acc: 0.6828, valid cost: 0.9303
Epoch 24, CIFAR-10 Batch 5:  batch acc: 0.9020, batch cost: 0.3487, valid acc: 0.6920, valid cost: 0.9070
Epoch 25, CIFAR-10 Batch 1:  batch acc: 0.8818, batch cost: 0.4046, valid acc: 0.7000, valid cost: 0.8851
Epoch 25, CIFAR-10 Batch 2:  batch acc: 0.8986, batch cost: 0.3389, valid acc: 0.6996, valid cost: 0.9224
Epoch 25, CIFAR-10 Batch 3:  batch acc: 0.9189, batch cost: 0.3051, valid acc: 0.6828, valid cost: 0.9734
Epoch 25, CIFAR-10 Batch 4:  batch acc: 0.8615, batch cost: 0.4027, valid acc: 0.6790, valid cost: 0.9527
Epoch 25, CIFAR-10 Batch 5:  batch acc: 0.9155, batch cost: 0.3000, valid acc: 0.6960, valid cost: 0.8887

Checkpoint

The model has been saved to disk.

Test Model

Test your model against the test dataset. This will be your final accuracy. You should have an accuracy greater than 50%. If you don't, keep tweaking the model architecture and parameters.


In [18]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
%config InlineBackend.figure_format = 'retina'

import tensorflow as tf
import pickle
import helper
import random

# Set batch size if not already set
try:
    if batch_size:
        pass
except NameError:
    batch_size = 64

save_model_path = './image_classification'
n_samples = 4
top_n_predictions = 3

def test_model():
    """
    Test the saved model against the test dataset
    """

    test_features, test_labels = pickle.load(open('preprocess_test.p', mode='rb'))
    loaded_graph = tf.Graph()

    with tf.Session(graph=loaded_graph) as sess:
        # Load model
        loader = tf.train.import_meta_graph(save_model_path + '.meta')
        loader.restore(sess, save_model_path)

        # Get Tensors from loaded model
        loaded_x = loaded_graph.get_tensor_by_name('x:0')
        loaded_y = loaded_graph.get_tensor_by_name('y:0')
        loaded_keep_prob = loaded_graph.get_tensor_by_name('keep_prob:0')
        loaded_logits = loaded_graph.get_tensor_by_name('logits:0')
        loaded_acc = loaded_graph.get_tensor_by_name('accuracy:0')
        
        # Get accuracy in batches for memory limitations
        test_batch_acc_total = 0
        test_batch_count = 0
        
        for test_feature_batch, test_label_batch in helper.batch_features_labels(test_features, test_labels, batch_size):
            test_batch_acc_total += sess.run(
                loaded_acc,
                feed_dict={loaded_x: test_feature_batch, loaded_y: test_label_batch, loaded_keep_prob: 1.0})
            test_batch_count += 1

        print('Testing Accuracy: {}\n'.format(test_batch_acc_total/test_batch_count))

        # Print Random Samples
        random_test_features, random_test_labels = tuple(zip(*random.sample(list(zip(test_features, test_labels)), n_samples)))
        random_test_predictions = sess.run(
            tf.nn.top_k(tf.nn.softmax(loaded_logits), top_n_predictions),
            feed_dict={loaded_x: random_test_features, loaded_y: random_test_labels, loaded_keep_prob: 1.0})
        helper.display_image_predictions(random_test_features, random_test_labels, random_test_predictions)


test_model()


Testing Accuracy: 0.6953584551811218

Why 50-80% Accuracy?

You might be wondering why you can't get an accuracy any higher. First things first, 50% isn't bad for a simple CNN. Pure guessing would get you 10% accuracy. However, you might notice people are getting scores well above 80%. That's because we haven't taught you all there is to know about neural networks. We still need to cover a few more techniques.

Submitting This Project

When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_image_classification.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "helper.py" and "problem_unittests.py" files in your submission.