[R] help! complete the reviewer's suggest: carry out GA+GP (gaussian process)!

bbslover dluthm at yeah.net
Fri Jan 21 08:40:00 CET 2011


Hello, all experts,

My major is computer-aied drug design ( main QSAR).

Now, my paper need be reviesed, and one reviewer ask me do genetic algorithm
coupled with gaussian process method (GA+GP).

my data:
training set: 191*106
    test set: 73*106

here, I need use GA+GP to do variable selection when building the model. 

In R, there are not GA package like in matlab
GA-toolbox(http://www.sheffield.ac.uk/acse/research/ecrg/gat.html) .

now, I just can use the matlab GA-tool box, however, I can not use
GP-toolbox in matlab. so I search the internet, find R package "genalg" can
do GA. and an example given is to do wavelength selection by GA+PLS, so I
think i certainly do the GA+GP. unfortunately, in this genalg package, i do
not know how to extract the selected variables, it seems likes there is not
such function. So I want to all friends help me to solve the reviewer's
suggestion: do GA+GP and extract the optimal variables and get the some
statistical parameters (i.e., cross-validation R2, pred R2 etc).

now, I can do GA+svm to do variable selection and build the models and get
some statistical paramets depicted above.

GA: matlab GA toolbox
(http://www.sheffield.ac.uk/acse/research/ecrg/gat.html)
svm: libsvm (http://www.csie.ntu.edu.tw/~cjlin/libsvm/)

now I want to know, how to get the predicted values :

In libsvm for example:
cmd = ['-v ',num2str(v),' -c',num2str(cgp(nind,1)),  '-g
',num2str(cgp(nind,2)),' -p ',num2str(cgp(nind,3)),' -s 3'];
model = svmtrain(train_y,train_data_best,cmd);
train_pred = svmpredict(train_y,train_data_best,model); % get the predicted
values for the training set

I can get the train_pred, likewise I can get the test_pred (tes_pred =
svmpredict(test_y,test_data_best,model);)

If I have the obsved train_y,test_y and the predicted train_pred and
test_pred, some statistical parameter can be calculated. 

But For GP, how can i get the predicted values?

(from GP website:  http://www.gaussianprocess.org/gpml/code/matlab/doc/)

prediction: [ymu ys2 fmu fs2   ] = gp(hyp, inf, mean, cov, lik, x, y, xs); 

here, the "ymu" are the predicted values that similar to "test_pred" in
libsvm?  

I hope all friends  can give me a hand, sincere there are little days i
should upload my revised manuscript, but until now this quest can not be
soved.

thanks for your help.

kevin




-- 
View this message in context: http://r.789695.n4.nabble.com/help-complete-the-reviewer-s-suggest-carry-out-GA-GP-gaussian-process-tp3229097p3229097.html
Sent from the R help mailing list archive at Nabble.com.



More information about the R-help mailing list