[对梯度迭代法与牛顿法求解极值问题的讨论.pptxVIP

梯度下降算法发散举例yx1x2?θ0=0, θ1=0, θ2=0, h(x(i))=0, x0=1y(1)=400, y(2)=330, y(3)=369, y(4)=232, y(5)=540x1(1)=2104, x1(2)=1600, x1(3)=2400, x1(4)=1416, x1(5)=3000x2(1)=3, x2(2)=3, x2(3)=3, x2(4)=2, x2(5)=4θ0=0+0.01×[(y(1)-h(x(1)))x0(1)+...+(y(5)-h(x(5)))x0(5)]θ1=0+0.01×[(y(1)-h(x(1)))x1(1)+...+(y(5)-h(x(5)))x1(5)]θ2=0+0.01×[(y(1)-h(x(1)))x2(1)+...+(y(5)-h(x(5)))x2(5)]θ0=0+0.01×[(y(1)-h(x(1)))x0(1)+...+(y(5)-h(x(5)))x0(5)]=0.01（400+330+369+232+540）=18.71θ1=0+0.01×[(y(1)-h(x(1)))x1(1)+...+(y(5)-h(x(5)))x1(5)]=0.01（400*2104+330*1600+…+540*3000)=42037.12θ2=0+0.01×[(y(1)-h(x(1)))x2(1)+...+(y(5)-h(x(5)))x2