I have read "A Practical Guide to Support Vector Classification", but I am
still confused about how to do homework 1.
In the guide, the proposed procedure is to use the RBF kernel (linear is fine
for homework right?) and use cross validation to find the best parameter C and
gamma.
The parts I don't understand are
1. The guide has this line - Each instance in the training set contains one
“target value”(class labels) and several“attributes”(features). Does this
mean that all instances in a training set should have the same label? Then the
following line in the guide - The goal of SVM is to produce a model which
predicts target value of data instances in the testing set which are given only
the attributes. Does that mean we need a model for each label? And if we have
multiple labels for an instance, each combination of labels needs a separate
model? ie. Label1 needs a model, Label1,3 needs another.
2. How do I interpret the result of kernel function. If I simply sub in xi and
xj to the kernel function, I get a number. What does that number mean?
3. How do I use k-nearest neighborhood to train the model? The guide suggests
that a grid search of C and gamma to identify the best C and gamma. What
k-nearest neighborhood should do here? If I am using linear model, there is no
parameter C and gamma.