1. 上次寫的一層神經網絡也都貼這裏了。
2. 我有點困，我先睡覺，完了我再修改
3. 這個代碼寫法不太符合工業代碼的規範，僅僅是用來學習的的。還望各位見諒

import sys,os
sys.path.append(os.pardir)
import numpy as np
from tensorflow.examples.tutorials.mnist import input_data
from PIL import Image
import tensorflow as tf

class mmmmser:
    id=""
    name=""

    def __int__(self,name,id):
        self.name = name
        self.id = id
    def setself(self,name,id):
        self.name = name
        self.id = id
    def setname(self,name):
        self.name = name
    def getname(self):
        return self.name
    def setid(self, id):
        self.id = id
    def getid(self):
        return self.id

#交叉熵損失函數
#值越小對應得越是預測的值
#預測的值越大，反而不是正確的預測
#減少了其他結果的影響
def cross_entropy_error(y, t):
    delta = 1e-7
    return -np.sum(t * np.log(y + delta))
#均方差損失函數
#均方誤差，對於的結果相減，平方求和。
#有其他結果的影響。
def mean_squared_error(y,t):
    return 0.5*np.sum((y-t)**2)
#圖像顯示函數
def showImg():
    batch_xs, batch_ys = mnist.train.next_batch (100)
    x=np.random.choice(batch_xs.shape[0],4)
    myx=batch_xs[0]
    myx=myx*255
    k=myx.reshape(28,28)
    print(myx)
    img=Image.fromarray(np.uint8(k))
    print(img)
    img.show()
def showImgTest(one_img):
    myx=one_img*255
    k=myx.reshape(28,28)
    img=Image.fromarray(np.uint8(k))
    img.show()
#獲取隨機的batch
def GetBatch(train_x,train_y):
    train_size = len(train_x)
    batch_size = 10   #每次訓練的batch數量
    batch_mask = np.random.choice (train_size, batch_size)
    # print (batch_mask)
    x_batch=[]
    t_batch=[]
    for i in range(len(batch_mask)):
        x_batch.append(np.array(train_x[batch_mask[i]]).reshape((-1,784)))
        t_batch.append(np.array(train_y[batch_mask[i]]).reshape((-1,10)))
    return np.array(x_batch).reshape((10,-1)),np.array(t_batch).reshape((10,-1))

#批量交叉熵
def cross_entropy_error_batch(y, t):
    batch_size = y.shape[0]
    return -np.sum(t * np.log(y + 1e-7)) / batch_size
def eachfile(filepath,arr,trr,k,a):
    print("loading face file..")
    pathdir=os.listdir(filepath)
    for allpath in pathdir:
        if(a>10 and k<1):
            break
        if k>=0:
            child=os.path.join(‘%s\%s‘%(filepath,allpath))
            if os.path.isfile(child):
                readfile(child,arr,trr,a)
            else:eachfile(child,arr,trr,a)
def getpeoplename(num):
    user1=mmmmser()
    user1.setself("tjl","9")
    user2=mmmmser()
    user2.setself("lw","8")
    user3=mmmmser()
    user3.setself("lt","0")
    user=[]
    user.append(user1)
    user.append(user2)
    user.append(user3)
    k=0
    for i in range(len(user)):
        if user[i].getid()==num:
            k=i
            break
        else:
            continue
    return user[k].getname()



def readfile(child,arr,trr,a):
    a+=1
    print(child)
    img1=Image.open(child)
    limg=img1.convert(‘L‘)
    k=limg.resize((28,28))
    if len(arr)==0:
        img = Image.fromarray (np.uint8(k))
        img.show()
    # print(np.array(k).shape)
    arr.append(k)
    trr.append([0,0,0,0,0,0,0,0,0,1])


    # fopen=open(filename,‘r‘)
    # fileread=fopen.read()
    # fopen.close()
def trainfaceoftian(arr,trr):
    print("train face of tianjingle...")
    x = tf.placeholder(tf.float32, [None, 784])
    y_ = tf.placeholder(tf.float32, [None, 10])
    trainImg=tf.reshape(x,[-1,28,28,1])
    #第一層網絡
    w1=tf.Variable(tf.truncated_normal([5,5,1,32], stddev=0.1),name="w1")
    b1=tf.Variable(tf.constant(0.1, shape=[32]),name="b1")
    h1=tf.nn.relu (tf.nn.conv2d(trainImg, w1, strides=[1, 1, 1, 1], padding=‘SAME‘) + b1)
    p1=tf.nn.max_pool(h1, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding=‘SAME‘)

    #第二層網絡
    w2=tf.Variable(tf.truncated_normal([5,5,32,64], stddev=0.1),name="w2")
    b2=tf.Variable(tf.constant(0.1, shape=[64]),name="b2")
    h2=tf.nn.relu (tf.nn.conv2d(p1, w2, strides=[1, 1, 1, 1], padding=‘SAME‘) + b2)
    p2=tf.nn.max_pool(h2, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding=‘SAME‘)

    #全連接層
    w3=tf.Variable(tf.truncated_normal([7*7*64,1024], stddev=0.1),name="w3")
    b3=tf.Variable(tf.constant(0.1, shape=[1024]),name="b3")

    #cnn輸出變形
    fp2=tf.reshape(p2,[-1,7*7*64])
    fp3=tf.nn.relu(tf.matmul(fp2, w3) + b3)

    #轉為【10】
    w4 = tf.Variable(tf.truncated_normal([1024, 10], stddev=0.14),name="w4")
    b4= tf.Variable(tf.constant(0.1,shape=[10]),name="b4")
    y_conv = tf.matmul(fp3, w4) + b4

    #定義交叉熵
    cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_,logits=y_conv))
    #定義訓練
    train_step = tf.train.AdamOptimizer(1e-4).minimize (cross_entropy)
    # 使用Dropout，keep_prob是一個占位符，訓練時為0.5，測試時為1
    keep_prob = tf.placeholder(tf.float32)
    # 定義測試的準確率
    correct_prediction = tf.equal(tf.argmax(y_conv, 1),tf.argmax(y_, 1))
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

    # 創建Session和變量初始化
    sess = tf.InteractiveSession()
    sess.run (tf.global_variables_initializer())

    #準確標誌
    maxaccuracy=0

    #模型存儲
    saver = tf.train.Saver(max_to_keep=1)
    # 訓練1000步
    for i in range (1000):
        # batch = mnist.train.next_batch(50)
        batch_xs, batch_ys = GetBatch (arr, trr)
        if i % 100 == 0:
            train_accuracy = accuracy.eval (feed_dict={x: batch_xs, y_: batch_ys, keep_prob: 1.0})
            print ("step %d, training accuracy %g" % (i, train_accuracy))
        train_step.run (feed_dict={x: batch_xs, y_: batch_ys, keep_prob: 0.5})
        if maxaccuracy<train_accuracy:
            maxaccuracy=train_accuracy
            saver.save (sess, ‘tface/tjlmodel.ckpt‘)
    print("Cnn model training with tjl end..")





def trainface():
    print("start train face...")
    filename="C:\\tianjingletianjingle"
    arr=[]
    trr=[]
    eachfile(filename,arr,trr,1,0)
    print("length of arr")
    print(len(arr))
    trainfaceoftian(arr,trr)

def loadface():
    filename="C:\\tianjingletianjingle"
    arr=[]
    trr=[]
    eachfile(filename,arr,trr,0,0)
    return GetBatch(arr,trr)

def predicttjl():
    print("start predict tjl...with cnn")
    meta_path = ‘tface/tjlmodel.ckpt.meta‘  #模型的結構
    model_path =‘tface/tjlmodel.ckpt‘       #模型的數據
    sess = tf.InteractiveSession()
    saver = tf.train.import_meta_graph (meta_path)
    saver.restore (sess, model_path)
    x = tf.placeholder (tf.float32, [None, 784])
    trainImg=tf.reshape(x,[-1,28,28,1])
    graph = tf.get_default_graph ()
    w1 = graph.get_tensor_by_name ("w1:0")
    b1 = graph.get_tensor_by_name ("b1:0")
    w2 = graph.get_tensor_by_name ("w2:0")
    b2 = graph.get_tensor_by_name ("b2:0")
    w3 = graph.get_tensor_by_name ("w3:0")
    b3 = graph.get_tensor_by_name ("b3:0")
    w4 = graph.get_tensor_by_name ("w4:0")
    b4 = graph.get_tensor_by_name ("b4:0")

    #第一層網絡
    # w1=tf.Variable(tf.truncated_normal([5,5,1,32], stddev=0.1),name="w1")
    # b1=tf.Variable(tf.constant(0.1, shape=[32]),name="b1")
    h1=tf.nn.relu (tf.nn.conv2d(trainImg, w1, strides=[1, 1, 1, 1], padding=‘SAME‘) + b1)
    p1=tf.nn.max_pool(h1, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding=‘SAME‘)
    #第二層網絡
    # w2=tf.Variable(tf.truncated_normal([5,5,32,64], stddev=0.1),name="w2")
    # b2=tf.Variable(tf.constant(0.1, shape=[64]),name="b2")
    h2=tf.nn.relu (tf.nn.conv2d(p1, w2, strides=[1, 1, 1, 1], padding=‘SAME‘) + b2)
    p2=tf.nn.max_pool(h2, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding=‘SAME‘)

    #全連接層
    # w3=tf.Variable(tf.truncated_normal([7*7*64,1024], stddev=0.1),name="w3")
    # b3=tf.Variable(tf.constant(0.1, shape=[1024]),name="b3")

    #cnn輸出變形
    fp2=tf.reshape(p2,[-1,7*7*64])
    fp3=tf.nn.relu(tf.matmul(fp2, w3) + b3)

    #轉為【10】
    # w4 = tf.Variable(tf.truncated_normal([1024, 10], stddev=0.14),name="w4")
    # b4= tf.Variable(tf.constant(0.1,shape=[10]),name="b4")
    y_conv = tf.matmul(fp3, w4) + b4
    keep_prob = tf.placeholder (tf.float32)


    one_img,one_title=loadface()
    # img1=tf.reshape(one_img[0],[-1,28,28,1])
    # print(img1.shape)
    img1=one_img[0].reshape((-1,784))
    # print(img1.reshape((0,784)).shape)
    temp=sess.run(y_conv,feed_dict={x: img1, keep_prob: 1.0})
    a=tf.arg_max(temp,1)
    b=tf.arg_max(one_title[0].reshape((-1,10)),1)
    if a.eval()==b.eval():
        print (‘success! his id is :%d his name is:%s‘%(b.eval()[0],getpeoplename(b.eval()[0])))
        showImgTest(one_img[0])
    else:
        print("error..")





def predictwithCnn():
    print("predict with 3 layer cnn..")
    meta_path = ‘cnn/model.ckpt.meta‘  #模型的結構
    model_path = ‘cnn/model.ckpt‘      #模型的數據
    sess = tf.InteractiveSession ()
    saver = tf.train.import_meta_graph (meta_path)
    saver.restore (sess, model_path)


    x = tf.placeholder (tf.float32, [None, 784])
    trainImg=tf.reshape(x,[-1,28,28,1])
    graph = tf.get_default_graph ()
    w1 = graph.get_tensor_by_name ("w1:0")
    b1 = graph.get_tensor_by_name ("b1:0")
    w2 = graph.get_tensor_by_name ("w2:0")
    b2 = graph.get_tensor_by_name ("b2:0")
    w3 = graph.get_tensor_by_name ("w3:0")
    b3 = graph.get_tensor_by_name ("b3:0")
    w4 = graph.get_tensor_by_name ("w4:0")
    b4 = graph.get_tensor_by_name ("b4:0")

    #第一層網絡
    # w1=tf.Variable(tf.truncated_normal([5,5,1,32], stddev=0.1),name="w1")
    # b1=tf.Variable(tf.constant(0.1, shape=[32]),name="b1")
    h1=tf.nn.relu (tf.nn.conv2d(trainImg, w1, strides=[1, 1, 1, 1], padding=‘SAME‘) + b1)
    p1=tf.nn.max_pool(h1, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding=‘SAME‘)
    #第二層網絡
    # w2=tf.Variable(tf.truncated_normal([5,5,32,64], stddev=0.1),name="w2")
    # b2=tf.Variable(tf.constant(0.1, shape=[64]),name="b2")
    h2=tf.nn.relu (tf.nn.conv2d(p1, w2, strides=[1, 1, 1, 1], padding=‘SAME‘) + b2)
    p2=tf.nn.max_pool(h2, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding=‘SAME‘)

    #全連接層
    # w3=tf.Variable(tf.truncated_normal([7*7*64,1024], stddev=0.1),name="w3")
    # b3=tf.Variable(tf.constant(0.1, shape=[1024]),name="b3")

    #cnn輸出變形
    fp2=tf.reshape(p2,[-1,7*7*64])
    fp3=tf.nn.relu(tf.matmul(fp2, w3) + b3)

    #轉為【10】
    # w4 = tf.Variable(tf.truncated_normal([1024, 10], stddev=0.14),name="w4")
    # b4= tf.Variable(tf.constant(0.1,shape=[10]),name="b4")
    y_conv = tf.matmul(fp3, w4) + b4
    keep_prob = tf.placeholder (tf.float32)
    one_img,one_title=mnist.train.next_batch (100)
    # img1=tf.reshape(one_img[0],[-1,28,28,1])
    # print(img1.shape)
    print(one_img.shape)
    img1=one_img[0].reshape((-1,784))
    print(img1.shape)
    temp=sess.run(y_conv,feed_dict={x: img1, keep_prob: 1.0})
    a=tf.arg_max(temp,1)
    b=tf.arg_max(one_title[0].reshape((-1,10)),1)
    print(a.eval())
    print(b.eval())
    if a.eval()==b.eval():
        print ("success! the num is :", (b.eval()[0]))
        showImgTest(img1)
    else:
        print("error..")


def CNNmodel():
    print("train with 3 layer cnn...")
    x = tf.placeholder(tf.float32, [None, 784])
    y_ = tf.placeholder(tf.float32, [None, 10])
    trainImg=tf.reshape(x,[-1,28,28,1])
    #第一層網絡
    w1=tf.Variable(tf.truncated_normal([5,5,1,32], stddev=0.1),name="w1")
    b1=tf.Variable(tf.constant(0.1, shape=[32]),name="b1")
    h1=tf.nn.relu (tf.nn.conv2d(trainImg, w1, strides=[1, 1, 1, 1], padding=‘SAME‘) + b1)
    p1=tf.nn.max_pool(h1, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding=‘SAME‘)

    #第二層網絡
    w2=tf.Variable(tf.truncated_normal([5,5,32,64], stddev=0.1),name="w2")
    b2=tf.Variable(tf.constant(0.1, shape=[64]),name="b2")
    h2=tf.nn.relu (tf.nn.conv2d(p1, w2, strides=[1, 1, 1, 1], padding=‘SAME‘) + b2)
    p2=tf.nn.max_pool(h2, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding=‘SAME‘)

    #全連接層
    w3=tf.Variable(tf.truncated_normal([7*7*64,1024], stddev=0.1),name="w3")
    b3=tf.Variable(tf.constant(0.1, shape=[1024]),name="b3")

    #cnn輸出變形
    fp2=tf.reshape(p2,[-1,7*7*64])
    fp3=tf.nn.relu(tf.matmul(fp2, w3) + b3)

    #轉為【10】
    w4 = tf.Variable(tf.truncated_normal([1024, 10], stddev=0.14),name="w4")
    b4= tf.Variable(tf.constant(0.1,shape=[10]),name="b4")
    y_conv = tf.matmul(fp3, w4) + b4

    #定義交叉熵
    cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_,logits=y_conv))
    #定義訓練
    train_step = tf.train.AdamOptimizer(1e-4).minimize (cross_entropy)
    # 使用Dropout，keep_prob是一個占位符，訓練時為0.5，測試時為1
    keep_prob = tf.placeholder(tf.float32)
    # 定義測試的準確率
    correct_prediction = tf.equal(tf.argmax(y_conv, 1),tf.argmax(y_, 1))
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

    # 創建Session和變量初始化
    sess = tf.InteractiveSession()
    sess.run (tf.global_variables_initializer())

    #準確標誌
    maxaccuracy=0

    #模型存儲
    saver = tf.train.Saver(max_to_keep=1)
    # 訓練1000步
    for i in range (400):
        batch = mnist.train.next_batch(50)
        print(batch[0].shape)
        if i % 100 == 0:
            train_accuracy = accuracy.eval (feed_dict={x: batch[0], y_: batch[1], keep_prob: 1.0})
            print ("step %d, training accuracy %g" % (i, train_accuracy))
        train_step.run (feed_dict={x: batch[0], y_: batch[1], keep_prob: 0.5})
        if maxaccuracy<train_accuracy:
            maxaccuracy=train_accuracy
            saver.save (sess, ‘cnn/model.ckpt‘)
    print("Cnn model training end..")
    # print("test accuracy %g" % accuracy.eval (feed_dict={x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0}))


def predict():
    meta_path = ‘ckpt/mnist.ckpt.meta‘
    model_path = ‘ckpt/mnist.ckpt‘
    sess = tf.InteractiveSession ()
    saver = tf.train.import_meta_graph (meta_path)
    saver.restore (sess, model_path)
    graph = tf.get_default_graph ()
    W = graph.get_tensor_by_name ("w:0")
    b = graph.get_tensor_by_name ("b:0")
    x = tf.placeholder (tf.float32, [None, 784])
    y = tf.nn.softmax (tf.matmul (x, W) + b)
    keep_prob = tf.placeholder (tf.float32)
    batch_xs, batch_ys=mnist.train.next_batch (100)
    one_img = batch_xs[0].reshape ((1, 784))
    one_num = batch_ys[0].reshape ((1, 10))
    temp = sess.run (y, feed_dict={x: one_img, keep_prob: 1.0})
    b = sess.run (tf.argmax (temp, 1))
    a = sess.run (tf.arg_max (one_num, 1))
    print(temp)
    print(one_num)
    if b == a:
        print ("success! the num is :", (b[0]))
        showImgTest(one_img)
    else:
        print ("mistakes predict.")



def trainNet():
    x = tf.placeholder (tf.float32, [None, 784])
    W = tf.Variable (tf.zeros ([784, 10]),name="w")
    b = tf.Variable (tf.zeros ([10]),name="b")
    y = tf.nn.softmax (tf.matmul (x, W) + b)
    y_ = tf.placeholder (tf.float32, [None, 10])
    keep_prob = tf.placeholder (tf.float32)
    # 定義測試的準確率
    correct_prediction = tf.equal (tf.argmax (y, 1), tf.argmax (y_, 1))
    accuracy = tf.reduce_mean (tf.cast (correct_prediction, tf.float32))
    #模型保存
    saver = tf.train.Saver (max_to_keep=1)
    #保存最優模型
    max_acc=0
    train_accuracy=0
    #交叉熵
    cross_entropy = tf.reduce_mean (-tf.reduce_sum (y_ * tf.log (y)))
    # cross_error=cross_entropy_error_batch(y,y_)
    train_step = tf.train.GradientDescentOptimizer (0.01).minimize (cross_entropy)
    sess = tf.InteractiveSession()
    tf.global_variables_initializer().run()
    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, keep_prob: 1.0})
        if i % 100 == 0:
            train_accuracy = accuracy.eval (feed_dict={x: batch_xs, y_: batch_ys, keep_prob: 1.0})
            print ("step %d, training accuracy %g" % (i, train_accuracy))
        if train_accuracy > max_acc:
            max_acc = train_accuracy
            saver.save (sess, ‘ckpt/mnist.ckpt‘)

if __name__ == ‘__main__‘:
    mnist = input_data.read_data_sets ("MNIST_data/", one_hot=True)
    choice="0"
    while choice == "0":
        print ("------------------------tensorflow--------------------------")
        print ("\t\t\t1\ttrain model..")
        print("\t\t\t2\tpredict model")
        print("\t\t\t3\tshow the first image")
        print("\t\t\t4\tCNN model")
        print("\t\t\t5\tpredict with cnn model")
        print("\t\t\t6\tface recognized train with cnn")
        print("\t\t\t7\tpredict tianjingle")
        print("-------------------------------------------------------------")
        print ("\t\t\t0\texit")
        choice = input ("please input your choice！")
        if choice == "1":
            print("start train...")
            trainNet()
        if choice=="2":
            predict()
        if choice=="3":
            showImg()
        if choice=="4":
            CNNmodel()
        if choice=="5":
            predictwithCnn()
        if choice=="6":
            trainface()
        if choice=="7":
            predicttjl()








    # train_x,train_y=mnist.train.next_batch(60000)
    # print(train_x.shape,train_y.shape)
    # img=train_x[1]
    # k=img.reshape(28,28)
    # #showImg(k)
    # y=[0.1,0.05,0.6,0.0,0.05,0.1,0.0,0.1,0.0,0.0]
    # t=[0,0,1,0,0,0,0,0,0,0]
    # print(mean_squared_error(np.array(y),np.array(t)))

#取圖像的個數

使用CNN做數字識別和人臉識別