import tensorflow as tf
import tensorflow.examples.tutorials.mnist.input_data as input_data
import numpy as np
import matplotlib.pyplot as plt
mnist =  input_data.read_data_sets("data/",one_hot = True)
#導入Tensorflwo和mnist數據集

#構建輸入層
x = tf.placeholder(tf.float32,[None,784],name=‘X‘)
y = tf.placeholder(tf.float32,[None,10],name=‘
 
Y‘)

#隱藏層神經元數量
H1_NN = 256 #第一層神經元數量
W1 = tf.Variable(tf.random_normal([784,H1_NN])) #權重
b1 = tf.Variable(tf.zeros([H1_NN])) #偏置項
Y1 = tf.nn.relu(tf.matmul(x,W1)+b1) #第一層輸出
W2 = tf.Variable(tf.random_normal([H1_NN,10]))#權重
b2 = tf.Variable(tf.zeros(10))#偏置項

forward = tf.matmul(Y1,W2)+b2 #定義前向傳播
 

pred = tf.nn.softmax(forward) #激活函數輸出

#損失函數
#loss_function = tf.reduce_mean(-tf.reduce_sum(y*tf.log(pred),
#                                            reduction_indices=1))
#(log(0))超出範圍報錯

loss_function  = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(logits=forward,labels=y))

 
#訓練參數
train_epochs = 50 #訓練次數
batch_size = 50 #每次訓練多少個樣本
total_batch = int(mnist.train.num_examples/batch_size) #隨機抽取樣本
display_step = 1 #訓練情況輸出
learning_rate = 0.01 #學習率

#優化器
opimizer = tf.train.AdamOptimizer(learning_rate).minimize(loss_function)

#準確率函數
correct_prediction = tf.equal(tf.argmax(y,1),tf.argmax(pred,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction,tf.float32))

#記錄開始訓練時間
from time import time
startTime = time()
#初始化變量
sess =tf.Session()
init = tf.global_variables_initializer()
sess.run(init)
#訓練
for epoch in range(train_epochs):
    for batch in range(total_batch):
        xs,ys = mnist.train.next_batch(batch_size)#讀取批次數據
        sess.run(opimizer,feed_dict={x:xs,y:ys})#執行批次數據訓練
    
    #total_batch個批次訓練完成後，使用驗證數據計算誤差與準確率
    loss,acc=sess.run([loss_function,accuracy],
                     feed_dict={x:mnist.validation.images,
                                y:mnist.validation.labels})
    #輸出訓練情況
    if(epoch+1) % display_step == 0:
        print("Train Epoch:",‘%02d‘ % (epoch + 1),
               "Loss=","{:.9f}".format(loss),"Accuracy=","{:.4f}".format(acc))
duration = time()-startTime
print("Trian Finshed takes:","{:.2f}".format(duration))#顯示預測耗時

#由於pred預測結果是one_hot編碼格式，所以需要轉換0~9數字
prediction_resul =  sess.run(tf.argmax(pred,1),feed_dict={x:mnist.test.images})

prediction_resul[0:10]

#模型評估
accu_test = sess.run(accuracy,
                    feed_dict={x:mnist.test.images,y:mnist.test.labels})
print("Accuray:",accu_test)



compare_lists = prediction_resul == np.argmax(mnist.test.labels,1)
print(compare_lists)
err_lists = [i for i in range(len(mnist.test.labels)) if compare_lists[i] == False]
print(err_lists,len(err_lists))

index_list = []
def print_predct_errs(labels,#標簽列表
                        perdiction):#預測值列表
    count = 0
    compare_lists = (perdiction == np.argmax(labels,1))
    err_lists = [i for i in range(len(labels)) if compare_lists[i] == False]
    for x in err_lists:
        index_list.append(x)
        print("index="+str(x)+
        "標簽值=",np.argmax(labels[x]),
        "預測值=",perdiction[x])
        count = count+1
    print("總計:",count)
    return index_list

print_predct_errs(mnist.test.labels,prediction_resul) 

def plot_images_labels_prediction(images,labels,prediction,index,num=25):
    fig = plt.gcf() # 獲取當前圖片
    fig.set_size_inches(10,12)
    if num>=25:
        num=25 #最多顯示25張圖片
    for i in range(0,num):
        ax = plt.subplot(5,5, i+1) #獲取當前要處理的子圖
        
        ax.imshow(np.reshape(images[index],(28,28)),cmap=‘binary‘)#顯示第index個圖像
        title = ‘label=‘ + str(np.argmax(labels[index]))#構建該圖上要顯示的title
        if len(prediction)>0:
            title += ‘predict= ‘+str(prediction[index])
        
        ax.set_title(title,fontsize=10)
        ax.set_xticks([])
        ax.set_yticks([])
        index += 1
    plt.show()

plot_images_labels_prediction(mnist.test.images,mnist.test.labels,prediction_resul,index=index_list[100])

單純記錄一下個人代碼，很基礎的一個MNIST手寫識別使用Tensorflwo實現，算是入門的Hello world 了，有些奇怪的問題暫時沒有解決 訓練次數調成40 在訓練到第35次左右發生了梯度爆炸，原因未知，損失函數要使用帶softmax那個，不然也會發生梯度爆炸

使用tensorflow實現mnist手寫識別(單層神經網絡實現)