Calculating margin and bias for SVM's -
i apologise newbishness of question in advance stuck. trying solve question,

i can parts i)-1v) stuck on v. know calculate margin y, do
y=2/||w|| and know w normal hyperplane, don't know how calculate it. always
w=[1;1] ?
similarly, bias, w^t * x + b = 0
how find value x data points? thank help.

consider building svm on (very little) data set shown in picture example this, maximum margin weight vector parallel shortest line connecting points of 2 classes, is, line between , , giving weight vector of . optimal decision surface orthogonal line , intersects @ halfway point. therefore, passes through . so, svm decision boundary is:

working algebraically, standard constraint , seek minimize . happens when constraint satisfied equality 2 support vectors. further know solution . have that:

therefore a=2/5 , b=-11/5, , . optimal hyperplane given

, b= -11/5 . margin boundary
this answer can confirmed geometrically examining picture.
Comments
Post a Comment