梯度下降法:
x,w,b---->z=w1x1+w2x2+...+wnxn+b---->a=sigmoid(z)---->l(a,y)
逆序推导
1.l(a,y)求导
da=-y/a+(1-y)/(1-a)
2.z求导
dz=dl(a,y)/dz [第二个dz为z的变化量]
=(-y/a+(1-y)/(1-a))*a(1-a)
=a-y
3.w求导
dw1=dl/dw1=x1*dz
wn同理
db=dz
4.更新
w1=w1-alpha*dw1
b=b-alpha*db
综上 w+=alpha*x*(y-a)
w=w+alpha*xT*[yT-sigmoid(x*w)]
© stonechina0616 | Powered by LOFTER