CODEWITHSUNDEEP
rsvmlibsvm
mir_nomer_nol
mir_nomer_nol

Reputation: 11

R e1071 predictions differ from libsvm

I'm trying to reproduce result that I get in R using pure libsvm.

First I've tried to export my model with write.svm function. I've got model file and scale file. I've scaled my test data with svm-scale -r model_file test_data and then applied svm-predict. But results that I get in R and in libsvm was different.

Then I tried to train libsvm with my train data and again I have predictions different from e1071 does.

Can someone help me?

P.S. I'm using eps-regression type of svm

Upvotes: 0

Views: 716

Answers (2)

mir_nomer_nol
mir_nomer_nol

Reputation: 11

The problem was in scaling, scale functions differ in R e1071 and libsvm. So I have descaled data in R and then scaled it back with libsvm. So it works fine.

Upvotes: 1

bright-star
bright-star

Reputation: 6427

Are you using a random sample for training? Is the e1071 libsvm implicitly using one? If so, does the seed match your sample?

Upvotes: 0

Related Questions