a problem with implementing Gradient descent

2 次查看（过去 30 天）

Maria Sargsyan 2015-11-21

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/256767-a-problem-with-implementing-gradient-descent

评论： Walter Roberson 2015-11-22

I wrote the following programm implementation of gradient descent. But when I run the programm it works really slow . I guess it is because of the symbolic differentiation or maybe there is another problem in the implementaion . Can, you tell me what is wrong ?

function y = grad(x0,y0,z0,f,m1,m2)
syms a;
symsb;
symc c;
n=0;
if  (length(symvar(diff(f,a))==3)&& length(symvar(diff(f,b))==3) && length(symvar(diff(f,c))==3))
f_a = matlabFunction(diff(f,a),[a,b,c]);
f_b=matlabFunction(diff(f,b),[a,b,c]);
f_c=matlabFunction(diff(f,c),[a,b,c]);
else 
y= 0;    
return;
end
delta = -1*[f_a(x0,y0,z0), f_b(x0,y0,z0), f_c(x0,y0,z0)];
phi = @(t) ( f(x0+t*delta(1), y0+t*delta(2), z0+t*delta(3)) );
min = GRSMethod( phi,m1,m2 );
while ( abs ( min * ( delta(1)^2+delta(2)^2+delta(3)^2 ))> 10^ (-5) || n<100)
x0=x0+min*delta(1);
y0= y0+min*delta(2);
z0=z0+min*delta(3);
delta = -1*[f_a(x0,y0,z0), f_b(x0,y0,z0), f_c(x0,y0,z0)];
min = GRSMethod( phi,m1,m2 ); 
n=n+1;    
end 
y = [ x0,y0,z0];
end