machine learning - should I add derivable features to a feature vector? -
for supervised learning (like classification), idea add new derivable features given feature set hope of improving accuracy?
for example: - if "unit_price" , "no_of_units" 2 given features, make sense create new feature "amount" (which unit_price*no_of_units)? - if "standard_deviation" given feature, idea create new feature "variance" (which standard_deviation^2)?
is there theoretical guideline this, or matter of trial-and-error?
thank you.
imho yes ,you can add new features that..look @ 'kernel' in svm can add feature xy if have 2 features x , y.
Comments
Post a Comment