Softmax Regression with MNIST

本文通過搭建Softmax Regression，並用MNIST數據集進行訓練以及測試，介紹tensorflow的最基礎使用方式。

MNIST數據集介紹以及Softmax迴歸介紹參考：http://wiki.jikexueyuan.com/project/tensorflow-zh/tutorials/mnist_beginners.html

1. MNIST數據集導入

   通過調用read_data_sets()，第一個參數填MNIST數據集存儲路徑，函數會自動判斷當前路徑下是否下載好數據，是否需要重新下載。

   import tensorflow.examples.tutorials.mnist.input_data as input_data
   mnist = input_data.read_data_sets("MNIST_data",one_hot=True)

2. Softmax迴歸模型搭建

   
   # Create the model
   
   
   #通過操作符號變量創建一個可交互的操作單元
   
   x = tf.placeholder(dtype=tf.float32, shape=[None, 784])
   
   #權重值和偏置量的創建
   
   w = tf.Variable(tf.zeros(shape=[784, 10]))
   b = tf.Variable(tf.zeros(shape=[10]))
   
   #Softmax模型創建
   
   y = tf.matmul(x, w) + b;
   
   # Define loss and optimizer
   
   y_ = tf.placeholder(tf.float32, [None, 10])

3. 訓練模型存儲

   
   #模型啓動
   
   sess = tf.InteractiveSession()
   saver=tf.train.Saver()
   def train():
     #交叉熵計算
     cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y))
     #執行反向傳播
     train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)
     tf.global_variables_initializer().run()
     # Train
     for _ in range(1000):
       batch_xs, batch_ys = mnist.train.next_batch(100)
       sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})
   
   train()
   
   #模型存儲，默認存儲路徑爲工程同目錄下文件夾
   
   saver.save(sess,save_path='./model/mnistmodel.ckpt')

4. 訓練模型載入

   載入模型時，必須先完整還原網絡結構的所有參數

   import tensorflow as tf
   import numpy as np
   import tensorflow.examples.tutorials.mnist.input_data as input_data
   
   mnist = input_data.read_data_sets("MNIST_data",one_hot=True)
   myGraph = tf.Graph()
   
   #還原網絡結構
   
   x = tf.placeholder(dtype=tf.float32, shape=[None, 784])
   w = tf.Variable(tf.zeros(shape=[784, 10]))
   b = tf.Variable(tf.zeros(shape=[10]))
   y = tf.matmul(x, w) + b;
   
   # Define loss and optimizer
   
   y_ = tf.placeholder(tf.float32, [None, 10])
   
   
   # Test trained model
   
   
   #提取變量
   
   saver = tf.train.Saver()
   with tf.Session() as sess:
       saver.restore(sess,'model/mnistmodel.ckpt')
       print('Weight:\n',sess.run(w))
       print('biases:\n',sess.run(b))
       #test
       correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
       accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
       print(sess.run(accuracy, feed_dict={x: mnist.test.images, y_: mnist.test.labels}))
   _: mnist.test.labels}))