I'm using the svmtrain function of libsvm for a 3 Features and 2 class problem. For other reasons i want to build a lookup table based on the decision boundaries from the svmtrain results.
The LUT has 3 dimensions - one for each feature and each entry is either -1 or 1 (--> 2 class problem).
Does anybody have an idea how the training data could be used to build a lookup table?
Not sure what you try to accomplish with this lookup table. SVM classifiers construct a separating hyperplane. A point's label is determined by the side of the hyperplane it resides on.
A hyperplane in n-dimensional space always separates it into two. You cannot determine which side a point is on by using less than n dimensions, in general. Your lookup table approach does not work.
Consider the following hyperplane in two dimensions: x1+x2=0 and you want to 'classify' a point based on one dimension, e.g. (x1val, ?) -> x1val + ? > 0. This is impossible.
Edit:
Still not sure what you are trying to do. If you simply want to tabulate your training set, just predict it with your model and structure the results the way you want them.
I can't see any reason to do this, though, so I guess I don't understand what you want to do.
