import tensorflow as tf
import numpy as np
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
from tensorflow.examples.tutorials.mnist import input_data
 
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)
trX, trY, teX, teY = mnist.train.images, mnist.train.labels, mnist.test.images, mnist.test.labels
#把上述trX和teX的形狀變?yōu)閇-1,28,28,1]，-1表示不考慮輸入圖片的數(shù)量，28×28是圖片的長和寬的像素數(shù)，
# 1是通道（channel）數(shù)量，因為MNIST的圖片是黑白的，所以通道是1，如果是RGB彩色圖像，通道是3。
trX = trX.reshape(-1, 28, 28, 1) # 28x28x1 input img
teX = teX.reshape(-1, 28, 28, 1) # 28x28x1 input img
 
X = tf.placeholder("float", [None, 28, 28, 1])
Y = tf.placeholder("float", [None, 10])
#初始化權(quán)重與定義網(wǎng)絡(luò)結(jié)構(gòu)。
# 這里，我們將要構(gòu)建一個擁有3個卷積層和3個池化層，隨后接1個全連接層和1個輸出層的卷積神經(jīng)網(wǎng)絡(luò)
def init_weights(shape):
 return tf.Variable(tf.random_normal(shape, stddev=0.01))
 
w = init_weights([3, 3, 1, 32])   # patch大小為3×3，輸入維度為1，輸出維度為32
w2 = init_weights([3, 3, 32, 64])   # patch大小為3×3，輸入維度為32，輸出維度為64
w3 = init_weights([3, 3, 64, 128])   # patch大小為3×3，輸入維度為64，輸出維度為128
w4 = init_weights([128 * 4 * 4, 625])  # 全連接層，輸入維度為 128 × 4 × 4,是上一層的輸出數(shù)據(jù)又三維的轉(zhuǎn)變成一維， 輸出維度為625
w_o = init_weights([625, 10]) # 輸出層，輸入維度為 625, 輸出維度為10，代表10類(labels)
# 神經(jīng)網(wǎng)絡(luò)模型的構(gòu)建函數(shù)，傳入以下參數(shù)
# X：輸入數(shù)據(jù)
# w：每一層的權(quán)重
# p_keep_conv，p_keep_hidden：dropout要保留的神經(jīng)元比例
 
def model(X, w, w2, w3, w4, w_o, p_keep_conv, p_keep_hidden):
 # 第一組卷積層及池化層，最后dropout一些神經(jīng)元
 l1a = tf.nn.relu(tf.nn.conv2d(X, w, strides=[1, 1, 1, 1], padding='SAME'))
 # l1a shape=(?, 28, 28, 32)
 l1 = tf.nn.max_pool(l1a, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
 # l1 shape=(?, 14, 14, 32)
 l1 = tf.nn.dropout(l1, p_keep_conv)
 
 # 第二組卷積層及池化層，最后dropout一些神經(jīng)元
 l2a = tf.nn.relu(tf.nn.conv2d(l1, w2, strides=[1, 1, 1, 1], padding='SAME'))
 # l2a shape=(?, 14, 14, 64)
 l2 = tf.nn.max_pool(l2a, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
 # l2 shape=(?, 7, 7, 64)
 l2 = tf.nn.dropout(l2, p_keep_conv)
 # 第三組卷積層及池化層，最后dropout一些神經(jīng)元
 l3a = tf.nn.relu(tf.nn.conv2d(l2, w3, strides=[1, 1, 1, 1], padding='SAME'))
 # l3a shape=(?, 7, 7, 128)
 l3 = tf.nn.max_pool(l3a, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
 # l3 shape=(?, 4, 4, 128)
 l3 = tf.reshape(l3, [-1, w4.get_shape().as_list()[0]]) # reshape to (?, 2048)
 l3 = tf.nn.dropout(l3, p_keep_conv)
 # 全連接層，最后dropout一些神經(jīng)元
 l4 = tf.nn.relu(tf.matmul(l3, w4))
 l4 = tf.nn.dropout(l4, p_keep_hidden)
 # 輸出層
 pyx = tf.matmul(l4, w_o)
 return pyx #返回預(yù)測值
 
#我們定義dropout的占位符——keep_conv，它表示在一層中有多少比例的神經(jīng)元被保留下來。生成網(wǎng)絡(luò)模型，得到預(yù)測值
p_keep_conv = tf.placeholder("float")
p_keep_hidden = tf.placeholder("float")
py_x = model(X, w, w2, w3, w4, w_o, p_keep_conv, p_keep_hidden) #得到預(yù)測值
#定義損失函數(shù)，這里我們?nèi)匀徊捎胻f.nn.softmax_cross_entropy_with_logits來比較預(yù)測值和真實值的差異，并做均值處理；
# 定義訓(xùn)練的操作（train_op），采用實現(xiàn)RMSProp算法的優(yōu)化器tf.train.RMSPropOptimizer，學(xué)習(xí)率為0.001，衰減值為0.9，使損失最??；
# 定義預(yù)測的操作（predict_op）
cost = tf.reduce_mean(tf.nn. softmax_cross_entropy_with_logits(logits=py_x, labels=Y))
train_op = tf.train.RMSPropOptimizer(0.001, 0.9).minimize(cost)
predict_op = tf.argmax(py_x, 1)
#定義訓(xùn)練時的批次大小和評估時的批次大小
batch_size = 128
test_size = 256
#在一個會話中啟動圖，開始訓(xùn)練和評估
# Launch the graph in a session
with tf.Session() as sess:
 # you need to initialize all variables
 tf. global_variables_initializer().run()
 for i in range(100):
  training_batch = zip(range(0, len(trX), batch_size),
        range(batch_size, len(trX)+1, batch_size))
  for start, end in training_batch:
   sess.run(train_op, feed_dict={X: trX[start:end], Y: trY[start:end],
           p_keep_conv: 0.8, p_keep_hidden: 0.5})
 
  test_indices = np.arange(len(teX)) # Get A Test Batch
  np.random.shuffle(test_indices)
  test_indices = test_indices[0:test_size]
 
  print(i, np.mean(np.argmax(teY[test_indices], axis=1) ==
       sess.run(predict_op, feed_dict={X: teX[test_indices],
               p_keep_conv: 1.0,
               p_keep_hidden: 1.0})))