13-PracticalMachineLearning

Which kernel parameter values which value for c try

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Classifier/ JSupportVectorApplet.html Tips and Tricks •  SVMs are not scale invariant •  Check if your library normalizes by default •  Normalize your data –  mean: 0 , std: 1 –  map to [0,1] or [- 1,1] •  Normalize test set in same way! Tips and Tricks •  RBF kernel is a good default •  For parameters try exponen%al sequences •  Read: Chih- Wei Hsu et al., “A Prac%cal Guide to Support Vector Classifica%on”, Bioinforma%cs (2010) Parameter Tuning •  Given a classifica%on task •  Which kernel ? •  Which kernel parameter values? •  Which value for C? Try different combina%ons and take the best. Grid Search Zang et al., “Iden%fica%on of heparin samples that contain impuri%es or contaminants by chemometric pa4ern recogni%on analysis of proton NMR spectral data”, Anal Bioanal Chem (2011) Mul% Class One vs. All One vs All •  Train n classifier for n classes •  Take classifica%on with greatest posi%ve margin •  Slow training Mul% Class One vs. One One vs One •  Train n(n- 1)/2 classifiers •  Take majority vote •  Fast training Decision Tree no aSer 10 pm? call...
View Full Document

Ask a homework question - tutors are online