matlab - SVM LibSVM Ignore Feature 1,3,5 when Predicting -

this question libsvm or svms in general. wonder if possible categorize feature-vectors of different length same svm model.

let's train svm 1000 instances of following feature vector: [feature1 feature2 feature3 feature4 feature5]

now want predict test-vector has same length of 5. if probability receive poor, want check first subset of test-vector containing columns 2-5. want dismiss 1 feature.

my question is: possible tell svm check features 2-5 prediction (e.g. weights), or have train different svm models. 1 5 features, 4 features , on...?

thanks in advance...

marcus

you can remove features test points fiddling file, highly recommend not using such approach. svm model valid when features present. if using linear kernel, setting given feature 0 implicitly cause ignored (though should not this). when using other kernels, no no.

using different set of features predictions set used training not approach.

i suggest train new model subset of features wish use in prediction.