#!/usr/bin/env python# 匯入mnist資料庫from tensorflow.examples.tutorials.mnist import input_datamnist = input_data.read_data_sets("MNIST_data/", one_hot=True)import tensorflow as tfimport os# 定義輸入變數x = tf.placeholder(tf.float32, [None, 784])# 定義引數W = tf.Variable(tf.zeros([784, 10]))b = tf.Variable(tf.zeros([10]))# 定義激勵函式y = tf.nn.softmax(tf.matmul(x, W) + b)# 定義輸出變數y_ = tf.placeholder(tf.float32, [None, 10])# 定義成本函式cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(y), reduction_indices=[1]))# 定義優化函式train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)# 初始化變數init = tf.global_variables_initializer()# 定義會話sess = tf.Session()# 執行初始化sess.run(init)# 定義模型儲存物件saver = tf.train.Saver()tf.add_to_collection('x', x)tf.add_to_collection('y', y)# 迴圈訓練1000次for i in range(1000): batch_xs, batch_ys = mnist.train.next_batch(100) sess.run(train_step, feed_dict={x: batch_xs, y_:batch_ys})print("訓練完成！")# 建立模型儲存目錄model_dir = "mnist_1"model_name = "ckp"if not os.path.exists(model_dir): os.mkdir(model_dir)# 儲存模型saver.save(sess, os.path.join(model_dir, model_name))print("儲存模型成功！")