NCA feature selection method

14 次查看（过去 30 天）
Alaa Almazroey
Alaa Almazroey 2019-9-11
评论： Alexis 2020-9-29
i tried to use NCA feature selection method to select the most relitive features from 16483 features, but i always got all features because the bestlambda and bestloss are always 0. how can work in this problem.
another question how can i set the lambdavals, is there a rule or something for linspace () parameter? as well how can i select the best value for 'tol' ?
cvx=cvpartition(size(Features,1),'kfold',5);
numvalidsets = cvx.NumTestSets;
n = cvx.TrainSize(1);
lambdavals=(linspace(0,20,20)./n;
lossvals = zeros(length(lambdavals),numvalidsets);
for w = 1:length(lambdavals)
for p =1:numvalidsets
train=1;
test=1;
indextrain=training(cvx,p);
for i=1:size(Features,1)
if indextrain(i)==1
XTrain(train,:)=Features(i,:);
YTrain(train)=label(i);
train=train+1;
else
XTest(test,:)=Features(i,:);
YTest(test)=label(i);
test=test+1;
end
end
TrainData= XTrain,YTrain;
TestData =XTest,YTest;
nca = fscnca(XTrain,YTrain,'FitMethod','exact', ...
'Solver','sgd','Lambda',lambdavals(w), ...
'IterationLimit',1,'Standardize',true);
lossvals(w,p) = loss(nca,XTest,YTest,'LossFunction','classiferror');
end
end
%%
meanloss = mean(lossvals,2);
[~,idx] = min(meanloss) % Find the index
bestlambda = lambdavals(idx) % Find the best lambda value
bestloss = meanloss(idx)
nca = fscnca(XTrain,YTrain,'FitMethod','exact','Solver','sgd',...
'Lambda',bestlambda,'Standardize',true,'Verbose',1);
tol = 0.55;
selidx = find(nca.FeatureWeights > tol*max(1,max(nca.FeatureWeights)))
Best_Features_train = XTrain(:,selidx);
  1 个评论
Alexis
Alexis 2020-9-29
I have this same problem. Without an error message or warning it's not clear to me where to start. I have 14 features and over 5,000 observations.

请先登录，再进行评论。

请先登录，再回答此问题。

回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Fit Postprocessing 的更多信息

标签

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by