0

I'm using svm rank that is implemented based on svm light, my data set consists of 50k samples with 2000 features. The svm_rank is running for 3 days and the iteration is over 5000 is this reasonable? I'm new to ml field and I need to use ranking algorithm. Do I need to change optimization parameters? I used svm_rank_learn -c 1 to train the classifier.

sshine
  • 15,635
  • 1
  • 41
  • 66
Samah
  • 15
  • 4
  • 1
    SVM ranking is unfortunately very expensive, as it transforms your problem to quadratic in size, so you are actually solving SVM classification task with 2 500 000 000 samples in 2000-dimensional space – lejlot Dec 18 '15 at 13:18
  • I used SVM ranking with 10,000 samples with feature space of 59D and it took 4 hours then when I experiment with the same size of samples but higher feature dimension , the run time is dramatically increased. that's what make me wondering. thank you for your clarification. – Samah Dec 18 '15 at 17:15

0 Answers0