Aseman <email@example.com> wrote in message <firstname.lastname@example.org>... > hi > > Consider a robust regression problem like this > x = (-1:0.02:1)'; > y = x+0.9*normrnd(0,0.1,length(x),1)+0.1*normrnd(4,0.1,length(x),1); > brob = robustfit(x,y)
Try replacing the first term of y with a*x+b
> I belive that both regress and robustfit employ mean square error. How can I used a different error criterion to solve the same problem?
If either of these do not have a weighting option, consider
help lscov doc lscov
Otherwise consider a neural network with 0 (or more) hidden nodes. Options include combinations of weighting, MSE, SSE, MAE and SAE.