應(yīng)用Tensorflow2.0的Eager模式是怎么快速構(gòu)建神經(jīng)網(wǎng)絡(luò)的

應(yīng)用Tensorflow2.0的Eager模式是怎么快速構(gòu)建神經(jīng)網(wǎng)絡(luò)的，相信很多沒有經(jīng)驗(yàn)的人對此束手無策，為此本文總結(jié)了問題出現(xiàn)的原因和解決方法，通過這篇文章希望你能解決這個(gè)問題。

import tensorflow as tf

    
    a = tf.constant(3.0)

    b = tf.placeholder(dtype = tf.float32)

    c = tf.add(a,b)

    sess = tf.Session() #創(chuàng)建會話對象

    init = tf.global_variables_initializer()

    sess.run(init) #初始化會話對象

    feed = {

        b: 2.0

    } #對變量b賦值

    c_res = sess.run(c, feed) #通過會話驅(qū)動(dòng)計(jì)算圖獲取計(jì)算結(jié)果

    print(c_res)

import tensorflow as tf

    import tensorflow.contrib.eager as tfe

    tf.enable_eager_execution()

  
def  add(num1, num2):

        a = tf.convert_to_tensor(num1) #將數(shù)值轉(zhuǎn)換為TF張量,這有利于加快運(yùn)算速度

        b = tf.convert_to_tensor(num2)

        c = a + b

        return c.numpy() #將張量轉(zhuǎn)換為數(shù)值
    
    add_res = add(3.0, 4.0)

    
    print(add_res)

from sklearn import datasets, preprocessing, model_selection

    data = datasets.load_iris() #加載數(shù)據(jù)到內(nèi)存

    x = preprocessing.MinMaxScaler(feature_range = (-1, 1)).fit_transform(data['data']) #將數(shù)據(jù)數(shù)值預(yù)處理到(-1,1)之間方便網(wǎng)絡(luò)識別
    
    #把不同分類的品種用向量表示，例如有三個(gè)不同品種，那么分別用(1,0,0),(0,1,0),(0,0,1)表示
    
    y = preprocessing.OneHotEncoder(sparse = False).fit_transform(data['target'].reshape(-1, 1))

    x_train, x_test, y_train, y_test = model_selection.train_test_split(x, y, test_size = 0.25, stratify = y) #將數(shù)據(jù)分成訓(xùn)練集合測試集

    print(len(x_train))

 class IrisClassifyModel(object):

        def  __init__(self, hidden_unit, output_unit):

            #這里只構(gòu)建兩層網(wǎng)絡(luò),第一層是輸入數(shù)據(jù)
   
            self.hidden_layer = tf.keras.layers.Dense(units = hidden_unit, activation = tf.nn.tanh, use_bias = True, name="hidden_layer")

            self.output_layer = tf.keras.layers.Dense(units = output_unit, activation = None, use_bias = True, name="output_layer")

        def  __call__(self, inputs):

            return self.output_layer(self.hidden_layer(inputs))

  
#構(gòu)造輸入數(shù)據(jù)檢驗(yàn)網(wǎng)絡(luò)是否正常運(yùn)行

    model = IrisClassifyModel(10, 3)

    train_dataset = tf.data.Dataset.from_tensor_slices((x_train, y_train))

 
    
    for x, y in tfe.Iterator(train_dataset.batch(32)):

        output = model(x)

        print(output.numpy())

        break
   

  
def  make_loss(model, inputs, labels):

        return  tf.reduce_sum(tf.nn.softmax_cross_entropy_with_logits_v2(logits = model(inputs), labels = labels))

    opt = tf.train.AdamOptimizer(learning_rate = 0.01)

    def train(model, x, y):

        opt.minimize(lambda:make_loss(model, x, y))

    accuracy = tfe.metrics.Accuracy()
    
    def  check_accuracy(model, x_batch, y_batch): #統(tǒng)計(jì)網(wǎng)絡(luò)判斷結(jié)果的準(zhǔn)確性

        accuracy(tf.argmax(model(tf.constant(x_batch)), axis = 1), tf.argmax(tf.constant(y_batch), axis = 1))

        return accuracy

import numpy as np

    model = IrisClassifyModel(10, 3)

    epochs = 50
    
    acc_history = np.zeros(epochs)

    for epoch in range(epochs):

    for (x_batch, y_batch) in tfe.Iterator(train_dataset.shuffle(1000).batch(32)):

    train(model, x_batch, y_batch)

     acc = check_accuracy(model, x_batch, y_batch)
   
        acc_history[epoch] = acc.result().numpy()

    import matplotlib.pyplot as plt
    
    plt.figure()
    
    plt.plot(acc_history)

    plt.xlabel('Epoch')

    plt.ylabel('Accuracy')

    plt.show()

  

上面代碼運(yùn)行后結(jié)果如下：  
 

看完上述內(nèi)容，你們掌握應(yīng)用Tensorflow2.0的Eager模式是怎么快速構(gòu)建神經(jīng)網(wǎng)絡(luò)的的方法了嗎？如果還想學(xué)到更多技能或想了解更多相關(guān)內(nèi)容，歡迎關(guān)注億速云行業(yè)資訊頻道，感謝各位的閱讀！