1 import tensorflow as tf 2 import numpy as np 3 # const = tf.constant(2.0, name='const') 4 # b = tf.placeholder(tf.float32, [None, 1], name='b') 5 # # b = tf.Variable(2.0, dtype=tf.float32, name='b') 6 # c = tf.Variable(1.0, dtype=tf.float32, name='c') 7 # 8 # d = tf.add(b, c, name='d') 9 # e = tf.add(c, const, name='e') 10 # a = tf.multiply(d, e, name='a') 11 # init = tf.global_variables_initializer() 12 # 13 # print(a) 14 # with tf.Session() as sess: 15 # sess.run(init) 16 # ans = sess.run(a, feed_dict={b: np.arange(0, 10)[:, np.newaxis]}) 17 # print(a) 18 # print(ans) 19 20 from tensorflow.examples.tutorials.mnist import input_data 21 mnist = input_data.read_data_sets("MNIST_data/", one_hot=True) # 載入數據集 22 23 learning_rate = 0.5 # 學習率 24 epochs = 10 # 訓練10次所有的樣本 25 batch_size = 100 # 每批訓練的樣本數 26 27 x = tf.placeholder(tf.float32, [None, 784]) # 為訓練集的特征提供占位符 28 y = tf.placeholder(tf.float32, [None, 10]) # 為訓練集的標簽提供占位符 29 30 W1 = tf.Variable(tf.random_normal([784, 300], stddev=0.03), name='W1') # 初始化隱藏層的W1參數 31 b1 = tf.Variable(tf.random_normal([300]), name='b1') # 初始化隱藏層的b1參數 32 W2 = tf.Variable(tf.random_normal([300, 10], stddev=0.03), name='W2') # 初始化全連接層的W1參數 33 b2 = tf.Variable(tf.random_normal([10]), name='b2') # 初始化全連接層的b1參數 34 35 hidden_out = tf.add(tf.matmul(x, W1), b1) # 定義隱藏層的第一步運算 36 hidden_out = tf.nn.relu(hidden_out) # 定義隱藏層經過激活函數后的運算 37 38 y_ = tf.nn.softmax(tf.add(tf.matmul(hidden_out, W2), b2)) # 定義全連接層的輸出運算 39 40 y_clipped = tf.clip_by_value(y_, 1e-10, 0.9999999) 41 cross_entropy = -tf.reduce_mean(tf.reduce_sum(y * tf.log(y_clipped) + (1 - y) * tf.log(1 - y_clipped), axis=1)) 42 # 交叉熵 43 44 optimizer = tf.train.GradientDescentOptimizer(learning_rate=learning_rate).minimize(cross_entropy) 45 # 梯度下降優化器，傳入的參數是交叉熵 46 47 init = tf.global_variables_initializer() # 所有參數初始化 48 49 correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1)) # 返回true|false 50 accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) # 將true轉化為1，false轉化為0 51 52 # 開始訓練 53 with tf.Session() as sess: 54 sess.run(init) 55 total_batch = int(len(mnist.train.labels) / batch_size) # 計算每個epoch要迭代幾次 56 for epoch in range(epochs): 57 avg_cost = 0 58 for i in range(total_batch): 59 batch_x, batch_y = mnist.train.next_batch(batch_size=batch_size) 60 _, c = sess.run([optimizer, cross_entropy], feed_dict={x: batch_x, y: batch_y}) 61 # 其實上面這一步只需要跑optimizer這個優化器就好了，因為交叉熵也會同時跑。 62 # 但是我們想要得到交叉熵的值來作為損失函數，所以還需要跑一個交叉熵。 63 avg_cost += c / total_batch 64 print("Epoch:", (epoch + 1), "cost = ", "{:.3f}".format(avg_cost)) # 這是每訓練完所有樣本得到的損失值 65 print(sess.run(accuracy, feed_dict={x: mnist.test.images, y: mnist.test.labels})) 66 # 因為之前的計算已經把中間參數計算出來了，所以這里只用最后的計算測試集就行了