在此项目中，你将对 CIFAR-10 数据集 中的图片进行分类。该数据集包含飞机、猫狗和其他物体。你需要预处理这些图片，然后用所有样本训练一个卷积神经网络。图片需要标准化（normalized），标签需要采用 one-hot 编码。你需要应用所学的知识构建卷积的、最大池化（max pooling）、丢弃（dropout）和完全连接（fully connected）的层。最后，你需要在样本图片上看到神经网络的预测结果。
请运行以下单元，以下载 CIFAR-10 数据集（Python版）。
In [17]:
"""
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)
该数据集分成了几部分／批次（batches），以免你的机器在计算时内存不足。CIFAR-10 数据集包含 5 个部分，名称分别为 data_batch_1、data_batch_2，以此类推。每个部分都包含以下某个类别的标签和图片：
了解数据集也是对数据进行预测的必经步骤。你可以通过更改 batch_id 和 sample_id 探索下面的代码单元。batch_id 是数据集一个部分的 ID（1 到 5）。sample_id 是该部分中图片和标签对（label pair）的 ID。
问问你自己：“可能的标签有哪些？”、“图片数据的值范围是多少？”、“标签是按顺序排列，还是随机排列的？”。思考类似的问题，有助于你预处理数据，并使预测结果更准确。
In [18]:
%matplotlib inline
%config InlineBackend.figure_format = 'retina'
import helper
import numpy as np
# Explore the dataset
batch_id = 1
sample_id = 5
helper.display_stats(cifar10_dataset_folder_path, batch_id, sample_id)
In [19]:
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)
In [20]:
from sklearn.preprocessing import OneHotEncoder
enc = OneHotEncoder()
each_label = np.array(list(range(10))).reshape(-1,1)
enc.fit(each_label)
print(enc.n_values_)
print(enc.feature_indices_)
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
X = np.array(x).reshape(-1, 1)
return enc.transform(X).toarray()
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_one_hot_encode(one_hot_encode)
In [21]:
"""
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)
In [22]:
"""
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'))
对于该神经网络，你需要将每层都构建为一个函数。你看到的大部分代码都位于函数外面。要更全面地测试你的代码，我们需要你将每层放入一个函数中。这样使我们能够提供更好的反馈，并使用我们的统一测试检测简单的错误，然后再提交项目。
注意：如果你觉得每周很难抽出足够的时间学习这门课程，我们为此项目提供了一个小捷径。对于接下来的几个问题，你可以使用 TensorFlow Layers 或 TensorFlow Layers (contrib) 程序包中的类来构建每个层级，但是“卷积和最大池化层级”部分的层级除外。TF Layers 和 Keras 及 TFLearn 层级类似，因此很容易学会。
但是，如果你想充分利用这门课程，请尝试自己解决所有问题，不使用 TF Layers 程序包中的任何类。你依然可以使用其他程序包中的类，这些类和你在 TF Layers 中的类名称是一样的！例如，你可以使用 TF Neural Network 版本的
conv2d类 tf.nn.conv2d，而不是 TF Layers 版本的conv2d类 tf.layers.conv2d。
我们开始吧！
神经网络需要读取图片数据、one-hot 编码标签和丢弃保留概率（dropout keep probability）。请实现以下函数：
neural_net_image_inputimage_shape 设置形状，部分大小设为 Nonename 参数对 TensorFlow 占位符 "x" 命名neural_net_label_inputn_classes 设置形状，部分大小设为 Nonename 参数对 TensorFlow 占位符 "y" 命名neural_net_keep_prob_inputname 参数对 TensorFlow 占位符 "keep_prob" 命名这些名称将在项目结束时，用于加载保存的模型。
注意：TensorFlow 中的 None 表示形状可以是动态大小。
In [23]:
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(tf.float32, shape=[None, image_shape[0], image_shape[1], image_shape[2]], name='x')
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(tf.float32, shape=[None, n_classes], name='y')
def neural_net_keep_prob_input():
"""
Return a Tensor for keep probability
: return: Tensor for keep probability.
"""
# TODO: Implement Function
return tf.placeholder(tf.float32, shape=None, name='keep_prob')
"""
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)
卷积层级适合处理图片。对于此代码单元，你应该实现函数 conv2d_maxpool 以便应用卷积然后进行最大池化：
conv_ksize、conv_num_outputs 和 x_tensor 的形状创建权重（weight）和偏置（bias）。conv_strides 对 x_tensor 应用卷积。pool_ksize 和 pool_strides 应用最大池化注意：对于此层，请勿使用 TensorFlow Layers 或 TensorFlow Layers (contrib)，但是仍然可以使用 TensorFlow 的 Neural Network 包。对于所有其他层，你依然可以使用快捷方法。
In [24]:
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_ksize: kernal size 2-D Tuple 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
weights = tf.Variable(tf.truncated_normal([conv_ksize[0], conv_ksize[1], x_tensor.get_shape().as_list()[-1], conv_num_outputs], stddev=0.1))
biases = tf.Variable(tf.zeros([conv_num_outputs]))
net = tf.nn.conv2d(x_tensor, weights, [1, conv_strides[0], conv_strides[1], 1], 'SAME')
net = tf.nn.bias_add(net, biases)
net = tf.nn.relu(net)
pool_kernel = [1, pool_ksize[0], pool_ksize[1], 1]
pool_strides = [1, pool_strides[0], pool_strides[1], 1]
net = tf.nn.max_pool(net, pool_kernel, pool_strides, 'VALID')
return net
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_con_pool(conv2d_maxpool)
实现 flatten 函数，将 x_tensor 的维度从四维张量（4-D tensor）变成二维张量。输出应该是形状（部分大小（Batch Size），扁平化图片大小（Flattened Image Size））。快捷方法：对于此层，你可以使用 TensorFlow Layers 或 TensorFlow Layers (contrib) 包中的类。如果你想要更大挑战，可以仅使用其他 TensorFlow 程序包。
In [25]:
import numpy as np
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
shape = x_tensor.get_shape().as_list()
dim = np.prod(shape[1:])
x_tensor = tf.reshape(x_tensor, [-1,dim])
return x_tensor
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_flatten(flatten)
实现 fully_conn 函数，以向 x_tensor 应用完全连接的层级，形状为（部分大小（Batch Size），num_outputs）。快捷方法：对于此层，你可以使用 TensorFlow Layers 或 TensorFlow Layers (contrib) 包中的类。如果你想要更大挑战，可以仅使用其他 TensorFlow 程序包。
In [26]:
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
weights = tf.Variable(tf.truncated_normal(shape=[x_tensor.get_shape().as_list()[-1], num_outputs], mean=0, stddev=1))
biases = tf.Variable(tf.zeros(shape=[num_outputs]))
net = tf.nn.bias_add(tf.matmul(x_tensor, weights), biases)
net = tf.nn.relu(net)
return net
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_fully_conn(fully_conn)
实现 output 函数，向 x_tensor 应用完全连接的层级，形状为（部分大小（Batch Size），num_outputs）。快捷方法：对于此层，你可以使用 TensorFlow Layers 或 TensorFlow Layers (contrib) 包中的类。如果你想要更大挑战，可以仅使用其他 TensorFlow 程序包。
注意：该层级不应应用 Activation、softmax 或交叉熵（cross entropy）。
In [27]:
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
weights = tf.Variable(tf.truncated_normal(shape=[x_tensor.get_shape().as_list()[-1], num_outputs], mean=0, stddev=1))
biases = tf.Variable(tf.zeros(shape=[num_outputs]))
net = tf.nn.bias_add(tf.matmul(x_tensor, weights), biases)
return net
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_output(output)
实现函数 conv_net， 创建卷积神经网络模型。该函数传入一批图片 x，并输出对数（logits）。使用你在上方创建的层创建此模型：
keep_prob 向模型中的一个或多个层应用 TensorFlow 的 Dropout
In [28]:
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)
# net = conv2d_maxpool(x, 32, (5,5), (1,1), (2,2), (2,2))
# net = tf.nn.dropout(net, keep_prob)
# net = conv2d_maxpool(net, 32, (5,5), (1,1), (2,2), (2,2))
# net = conv2d_maxpool(net, 64, (5,5), (1,1), (2,2), (2,2))
net = conv2d_maxpool(x, 32, (3,3), (1,1), (2,2), (2,2))
net = tf.nn.dropout(net, keep_prob)
net = conv2d_maxpool(net, 64, (3,3), (1,1), (2,2), (2,2))
net = tf.nn.dropout(net, keep_prob)
net = conv2d_maxpool(net, 64, (3,3), (1,1), (2,2), (2,2))
# TODO: Apply a Flatten Layer
# Function Definition from Above:
# flatten(x_tensor)
net = flatten(net)
# TODO: Apply 1, 2, or 3 Fully Connected Layers
# Play around with different number of outputs
# Function Definition from Above:
# fully_conn(x_tensor, num_outputs)
net = fully_conn(net, 64)
# TODO: Apply an Output Layer
# Set this to the number of classes
# Function Definition from Above:
# output(x_tensor, num_outputs)
net = output(net, enc.n_values_)
# TODO: return output
return net
"""
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)
In [29]:
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
_ = session.run([optimizer, cost, accuracy], 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)
In [30]:
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
valid_loss, valid_accuracy = session.run([cost, accuracy], feed_dict={x: valid_features, y: valid_labels, keep_prob: 1})
print("valid loss {:.3f}, accuracy {:.3f}".format(valid_loss, valid_accuracy))
In [31]:
# TODO: Tune Parameters
epochs = 100
batch_size = 256
keep_probability = 0.8
In [32]:
"""
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)
In [33]:
"""
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)
In [34]:
"""
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()
你可能想问，为何准确率不能更高了？首先，对于简单的 CNN 网络来说，50% 已经不低了。纯粹猜测的准确率为10%。但是，你可能注意到有人的准确率远远超过 80%。这是因为我们还没有介绍所有的神经网络知识。我们还需要掌握一些其他技巧。
提交项目时，确保先运行所有单元，然后再保存记事本。将 notebook 文件另存为“dlnd_image_classification.ipynb”，再在目录 "File" -> "Download as" 另存为 HTML 格式。请在提交的项目中包含 “helper.py” 和 “problem_unittests.py” 文件。