import numpy as np
from sklearn.metrics import accuracy_score
def sigmod(x):
    return 1.0/(1.0+np.exp(-x))
def wangluo(l,m):#建立多隐层神经网络
    w1 = np.random.randn(2,l)
    b1 = np.zeros(( 1 ,l))
    w2 =np.random.randn(l,m)
    b2 = np.zeros(( 1 ,m))
    w3 = np.random.randn(m,2 )
    b3 = np.zeros(( 1 ,2 ))
    return w1,b1,w2,b2,w3,b3
t = np.zeros((200,2))
for i in range (200):
    t[i,y[i]]=1
epsilon = 0.0008
w1,b1,w2,b2,w3,b3 = wangluo(10,10)
for i in range(1000):#开始迭代
    a1 = sigmod(np.dot(x,w1)+b1)
    a2 = sigmod(np.dot(a1,w2)+b2)
    a3 = sigmod(np.dot(a2,w3)+b3)
    delta3 = a3*(1-a3)*(t-a3)
    delta2 = a2*(1-a2)*(np.dot(delta3,w3.T))
    delta1 = a1*(1-a1)*(np.dot(delta2,w2.T))
    w3 += epsilon * np.dot(a2.T, delta3)
    b3 += epsilon * np.sum(delta3, axis=0)
    w2 += epsilon * np.dot(a1.T, delta2)
    b2 += epsilon * np.sum(delta2, axis=0)
    w1 += epsilon * np.dot(x.T, delta1)
    b1 += epsilon * np.sum(delta1, axis=0)
    if (i+1)%100==0:
        L = np.sum((a3 - t)**2)
        y_pre = np.argmax(a3, axis=1)
        acc = accuracy_score(y, y_pre)
        print("每迭代100次的损失函数为：",L,"每迭代100次的准确率为：",acc)
    if i ==999:
        print(" ")
        print("最终的损失函数为：",L,"最终的准确率为：",acc)