import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data

#載入資料集
mnist = input_data.read_data_sets('MNIST_data',one_hot=True)

#每批次放進50個
batch_size = 50
#計算一共有多少批次
n_batch = mnist.train.num_examples // batch_size

#定義兩個placeholder 屬性的個數是固定的  樣本數是變化的
x = tf.placeholder(tf.float32,[None,784])
y = tf.placeholder(tf.float32,[None,10]) 
keep_prob = tf.placeholder(tf.float32)
L1_Node = 500
L2_Node = 300
L3_Node = 200
lr = tf.Variable(0.01,dtype = tf.float32)
#兩個None對應位置是一樣的

#建立第一個隱層神經網路
W_1 = tf.Variable(tf.truncated_normal([784,L1_Node],stddev=0.1))
b_1 = tf.Variable(tf.zeros([L1_Node])+0.1)
Wx_plus_b_1 = tf.matmul(x,W_1)+b_1
#雖然與ng講的是相反的 但是這裡維度是相對應乘起來的，所以沒啥問題
#行數是樣本數  一行裡面有784個feature  所以列是784 列    最後維度要對應上
prediction_L1 = tf.nn.tanh(Wx_plus_b_1)
L1_drop = tf.nn.dropout(prediction_L1,keep_prob)

#建立第二隱層
W_2 = tf.Variable(tf.truncated_normal([L1_Node,L2_Node],stddev=0.1))
b_2 = tf.Variable(tf.zeros([L2_Node])+0.1)
Wx_plus_b_2 = tf.matmul(prediction_L1,W_2)+b_2
prediction_L2 = tf.nn.tanh(Wx_plus_b_2)
L2_drop = tf.nn.dropout(prediction_L2,keep_prob)

#建立第三隱層
W_3 = tf.Variable(tf.truncated_normal([L2_Node,L3_Node],stddev=0.1))
b_3 = tf.Variable(tf.zeros([L3_Node])+0.1)
Wx_plus_b_3 = tf.matmul(prediction_L2,W_3)+b_3
prediction_L3 = tf.nn.tanh(Wx_plus_b_3)
L3_drop = tf.nn.dropout(prediction_L3,keep_prob)

#建立輸出層
W_4 = tf.Variable(tf.truncated_normal([L3_Node,10])*0.01)
b_4 = tf.Variable(tf.zeros([10])+0.1)
Wx_plus_b_4 = tf.matmul(prediction_L3,W_4)+b_4
prediction = tf.nn.softmax(Wx_plus_b_4)

#二次代價函式
#loss = tf.reduce_mean(tf.square(y-prediction))

#使用交叉熵損失
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels = y,logits= prediction))
#使用Adam優化器
train_step = tf.train.AdamOptimizer(lr).minimize(loss)

#初始化變數
init = tf.global_variables_initializer()

#
correct_prediction = tf.equal(tf.argmax(y,1),tf.argmax(prediction,1))
#tf.equal(x,y) 看x，y是否一樣  一樣就true
# argmax的0就是縱向的元素中最大的那個的索引
# 1 就是每行橫向的元素中最大的那個元素的索引
# 0軸就是豎向，1就是橫向 但是要分清元素還是層級
#arg會返回一個array
#最終返回布林型別列表

#求準確率
accuracy = tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
#tf.cast 是將布林型別轉化為浮點型

#建立會話
with tf.Session() as sess:
    sess.run(init)
    for epoch in range(51):
        sess.run(tf.assign(lr,0.001*(0.95**epoch)))
        for batch in range(n_batch):
            batch_xs,batch_ys = mnist.train.next_batch(batch_size)
            sess.run(train_step,feed_dict={x:batch_xs,y:batch_ys,keep_prob : 1.0})
        
        learning_rate = sess.run(lr)
            
        test_acc = sess.run(accuracy,feed_dict={x:mnist.test.images,y:mnist.test.labels,keep_prob : 1.0})
        train_acc = sess.run(accuracy,feed_dict={x:mnist.train.images,y:mnist.train.labels,keep_prob : 1.0})
        #利用訓練集上的資料來勁訓練，用測試級來進行
        print('Iter'+ str(epoch)+',testing accuracy'+str(test_acc)+',train acc'+str(train_acc)+',learning_rate'+str(learning_rate))
 

模型經過調參，修改層數，節點，優化器後 準確率得到提升，最終達到0.98以上