Bad results obtained for LS-SVM regression. Any help?
3 views (last 30 days)
Show older comments
Hello. Although I read many times the tutorial for this LS-SVM toolbox http://www.esat.kuleuven.be/sista/lssvmlab/ , I'm still having very bad results when I use it for regression.
Here is the code I execute:
% Initialisation of the model
model = initlssvm(X,Y,'f',[],[],'RBF_kernel','o');
% Tuning of the model, optimizing the choice of gam and sig2
model = tunelssvm(model,'simplex','crossvalidatelssvm',{10,'mae'});
% Training
model = trainlssvm(model);
% Prediction of results
Yt=simlssvm(model,Xt)
The results obtained are far different from the initial ones. I really don't understand what I do wrong. Any one help me, please ? Thank you!
2 Comments
Gustavo
on 22 Oct 2013
When you use the 'o' in initlssvm you are not normalizing the data, which is quite important in many cases for regression. Can you try without it?
SANA
on 5 Apr 2018
My results are worse, but you can try giving different values of sig2 and gam in line: model = initlssvm(X,Y,'f',[gam],[sig2],'RBF_kernel','o'); This may solve the problem. I am getting a warning: *****> In crossvalidatelssvm (line 96) In tunelssvm>simanncostfun2 (line 485) In tunelssvm>@(x)simanncostfun2(x,model,costfun,costargs) (line 168) In csa (line 79) In tunelssvm (line 168) In tunelssvm (line 132) Warning: Matrix is close to singular or badly scaled. Results may be inaccurate. RCOND = 5.551115e-17. Any idea how to fix this?
Answers (0)
See Also
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!