在筆記7中，和大家一起入門了 Tensorflow 的基本語(yǔ)法，并舉了一些實(shí)際的例子進(jìn)行了說(shuō)明，終于告別了使用 numpy 手動(dòng)搭建的日子。所以我們將繼續(xù)往下走，看看如何利用 Tensorflow 搭建神經(jīng)網(wǎng)絡(luò)模型。

盡管對(duì)于初學(xué)者而言使用 Tensorflow 看起來(lái)并不那么習(xí)慣，需要各種步驟，但簡(jiǎn)單來(lái)說(shuō)，Tensorflow 搭建模型實(shí)際就是兩個(gè)過(guò)程：創(chuàng)建計(jì)算圖和執(zhí)行計(jì)算圖。在 deeplearningai 課程中，NG和他的課程組給我們提供了 Signs Dataset （手勢(shì)）數(shù)據(jù)集，其中訓(xùn)練集包括1080張64x64像素的手勢(shì)圖片，并給定了 6 種標(biāo)注，測(cè)試集包括120張64x64的手勢(shì)圖片，我們需要對(duì)訓(xùn)練集構(gòu)建神經(jīng)網(wǎng)絡(luò)模型然后對(duì)測(cè)試集給出預(yù)測(cè)。

先來(lái)簡(jiǎn)單看一下數(shù)據(jù)集：

#LoadingthedatasetX_train_orig,Y_train_orig,X_test_orig,Y_test_orig,classes=load_dataset()#FlattenthetrainingandtestimagesX_train_flatten=X_train_orig.reshape(X_train_orig.shape[0],-1).T
X_test_flatten=X_test_orig.reshape(X_test_orig.shape[0],-1).T#NormalizeimagevectorsX_train=X_train_flatten/255.X_test=X_test_flatten/255.#ConverttrainingandtestlabelstoonehotmatricesY_train=convert_to_one_hot(Y_train_orig,6)
Y_test=convert_to_one_hot(Y_test_orig,6)print("numberoftrainingexamples="+str(X_train.shape[1]))print("numberoftestexamples="+str(X_test.shape[1]))print("X_trainshape:"+str(X_train.shape))print("Y_trainshape:"+str(Y_train.shape))print("X_testshape:"+str(X_test.shape))print("Y_testshape:"+str(Y_test.shape))

下面就根據(jù) NG 給定的找個(gè)數(shù)據(jù)集利用 Tensorflow 搭建神經(jīng)網(wǎng)絡(luò)模型。我們選擇構(gòu)建一個(gè)包含 2 個(gè)隱層的神經(jīng)網(wǎng)絡(luò)，網(wǎng)絡(luò)結(jié)構(gòu)大致如下：
LINEAR -> RELU -> LINEAR -> RELU -> LINEAR -> SOFTMAX
正如我們之前利用 numpy 手動(dòng)搭建一樣，搭建一個(gè)神經(jīng)網(wǎng)絡(luò)的主要步驟如下：
-定義網(wǎng)絡(luò)結(jié)構(gòu)
-初始化模型參數(shù)
-執(zhí)行前向計(jì)算/計(jì)算當(dāng)前損失/執(zhí)行反向傳播/權(quán)值更新

創(chuàng)建 placeholder

根據(jù) Tensorflow 的語(yǔ)法，我們首先創(chuàng)建輸入X 和輸出 Y 的占位符變量，這里需要注意 shape 參數(shù)的設(shè)置。

def create_placeholders(n_x, n_y):
  X = tf.placeholder(tf.float32, shape=(n_x, None), name='X')
  Y = tf.placeholder(tf.float32, shape=(n_y, None), name='Y')  
  return X, Y

初始化模型參數(shù)

其次就是初始化神經(jīng)網(wǎng)絡(luò)的模型參數(shù)，三層網(wǎng)絡(luò)包括六個(gè)參數(shù)，這里我們采用Xavier初始化方法：

def initialize_parameters(): 
  tf.set_random_seed(1)         
  W1 = tf.get_variable("W1", [25, 12288], initializer = tf.contrib.layers.xavier_initializer(seed = 1))
  b1 = tf.get_variable("b1", [25, 1], initializer = tf.zeros_initializer())
  W2 = tf.get_variable("W2", [12, 25], initializer = tf.contrib.layers.xavier_initializer(seed = 1))
  b2 = tf.get_variable("b2", [12, 1], initializer = tf.zeros_initializer())
  W3 = tf.get_variable("W3", [6, 12], initializer = tf.contrib.layers.xavier_initializer(seed = 1))
  b3 = tf.get_variable("b3", [6,1], initializer = tf.zeros_initializer())

  parameters = {"W1": W1,         
         "b1": b1,  
         "W2": W2, 
         "b2": b2, 
         "W3": W3, 
         "b3": b3}  
  return parameters

執(zhí)行前向傳播

defforward_propagation(X,parameters):"""
Implementstheforwardpropagationforthemodel:LINEAR->RELU->LINEAR->RELU->LINEAR->SOFTMAX
"""

W1=parameters['W1']
b1=parameters['b1']
W2=parameters['W2']
b2=parameters['b2']
W3=parameters['W3']
b3=parameters['b3']

Z1=tf.add(tf.matmul(W1,X),b1)
A1=tf.nn.relu(Z1)
Z2=tf.add(tf.matmul(W2,A1),b2)
A2=tf.nn.relu(Z2)
Z3=tf.add(tf.matmul(W3,A2),b3)
returnZ3

計(jì)算損失函數(shù)

在 Tensorflow 中損失函數(shù)的計(jì)算要比手動(dòng)搭建時(shí)方便很多，一行代碼即可搞定：

def compute_cost(Z3, Y):
  logits = tf.transpose(Z3)
  labels = tf.transpose(Y)

  cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits = logits, labels = labels))  
  return cost

代碼整合：執(zhí)行反向傳播和權(quán)值更新

跟計(jì)算損失函數(shù)類似，Tensorflow 中執(zhí)行反向傳播的梯度優(yōu)化非常簡(jiǎn)便，兩行代碼即可搞定，定義完整的神經(jīng)網(wǎng)絡(luò)模型如下：

def model(X_train, Y_train, X_test, Y_test, learning_rate = 0.0001,
     num_epochs = 1500, minibatch_size = 32, print_cost = True):
  ops.reset_default_graph()          
  tf.set_random_seed(1)             
  seed = 3                     
  (n_x, m) = X_train.shape            
  n_y = Y_train.shape[0]             
  costs = []                  

  # Create Placeholders of shape (n_x, n_y)
  X, Y = create_placeholders(n_x, n_y)  # Initialize parameters
  parameters = initialize_parameters()  # Forward propagation: Build the forward propagation in the tensorflow graph

  Z3 = forward_propagation(X, parameters)  # Cost function: Add cost function to tensorflow graph
  cost = compute_cost(Z3, Y)  # Backpropagation: Define the tensorflow optimizer. Use an AdamOptimizer.
  optimizer = tf.train.GradientDescentOptimizer(learning_rate = learning_rate).minimize(cost)  # Initialize all the variables
  init = tf.global_variables_initializer()  # Start the session to compute the tensorflow graph
  with tf.Session() as sess:    # Run the initialization
    sess.run(init)    # Do the training loop
    for epoch in range(num_epochs):
      epoch_cost = 0.          
      num_minibatches = int(m / minibatch_size) 
      seed = seed + 1
      minibatches = random_mini_batches(X_train, Y_train, minibatch_size, seed)      
      for minibatch in minibatches:        # Select a minibatch
        (minibatch_X, minibatch_Y) = minibatch
        _ , minibatch_cost = sess.run([optimizer, cost], feed_dict={X: minibatch_X, Y: minibatch_Y})
        epoch_cost += minibatch_cost / num_minibatches      # Print the cost every epoch
    if print_cost == True and epoch % 100 == 0:        
      print ("Cost after epoch %i: %f" % (epoch, epoch_cost))      
    if print_cost == True and epoch % 5 == 0:
        costs.append(epoch_cost)    # plot the cost
    plt.plot(np.squeeze(costs))
    plt.ylabel('cost')
    plt.xlabel('iterations (per tens)')
    plt.title("Learning rate =" + str(learning_rate))
    plt.show()    # lets save the parameters in a variable
    parameters = sess.run(parameters)    
    print ("Parameters have been trained!")    # Calculate the correct predictions
    correct_prediction = tf.equal(tf.argmax(Z3), tf.argmax(Y))    # Calculate accuracy on the test set
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))    
    print ("Train Accuracy:", accuracy.eval({X: X_train, Y: Y_train}))    
    print ("Test Accuracy:", accuracy.eval({X: X_test, Y: Y_test}))    
    return parameters

執(zhí)行模型：

parameters=model(X_train,Y_train,X_test,Y_test)

根據(jù)模型的訓(xùn)練誤差和測(cè)試誤差可以看到：模型整體效果雖然沒有達(dá)到最佳，但基本也能達(dá)到預(yù)測(cè)效果。

總結(jié)

• Tensorflow 語(yǔ)法中兩個(gè)基本的對(duì)象類是 Tensor 和 Operator.

• Tensorflow 執(zhí)行計(jì)算的基本步驟為

  • 創(chuàng)建計(jì)算圖（張量、變量和占位符變量等）

  • 創(chuàng)建會(huì)話

  • 初始化會(huì)話

  • 在計(jì)算圖中執(zhí)行會(huì)話

可以看到的是，在 Tensorflow 中編寫神經(jīng)網(wǎng)絡(luò)要比我們手動(dòng)搭建要方便的多，這也正是深度學(xué)習(xí)框架存在的意義之一。功能強(qiáng)大的深度學(xué)習(xí)框架能夠幫助我們快速的搭建起復(fù)雜的神經(jīng)網(wǎng)絡(luò)模型，在經(jīng)歷了手動(dòng)搭建神經(jīng)網(wǎng)絡(luò)的思維訓(xùn)練過(guò)程之后，這對(duì)于我們來(lái)說(shuō)就不再困難了。

本文由《自興動(dòng)腦人工智能》項(xiàng)目部 凱文 投稿。