I use lasso to perform regularized regression on the weights of the output layer of an ELM.
[OutputWeight FitInfo] = lasso(H',T','Lambda',[2 1 0.5 0.3],'Alpha',0.1);
[minMSE, minLidx] = min(FitInfo.MSE) ;
OutputWeight = OutputWeight(:,minLidx);
However the MSE I get from the lasso function is different (much less, around 2-3 orders of magnitude) than the MSE when I perform
Y=(H' * OutputWeight)';
Tr_error = mse(T-Y)
Any idea why this happens?
In the above notations, T is tha targets, and Y the output of the network
Hello, I cannot find your paper by looking at your publications , can you point me to it?
Thanks in advance,
Nikolaos Bismpikos
Hi, Nikolaos
"The Matroska Feature Selection for Microarray data" tells us the structure of microarray data completely that consist several linealy separable gene subspaces.
There is no merit for feature selection by LASSO.
The book preface titled "New Theory of Discriminant Analysis after R. Fisher" discuss about LASSO that will be published in October from Springer.
- Badges
- Science topic