This package implements Kernel SVM in julia. The quadratic optimisation problem is solved with Ipopt and
the following kernel functions are availables:
- Linear
- Polynomial
- Laplacian RBF
- RBF
Pkg.clone("https://github.com/remusao/KSVM.jl.git")
Pkg.update()using Vega
using KSVM
# Declare data
x = [[0 0 0 0 0 1 1 2 2 2 2 2 3 3 3 3 3 4 4 4 4 4 5 5 6 6 6 6 6],
[1 2 3 4 5 0 6 0 2 3 4 6 0 2 3 4 6 0 2 3 4 6 0 6 1 2 3 4 5]]
# Labels (-1 or +1)
y = [-1, -1, -1, -1, -1, -1, -1, -1, 1, 1, 1, -1, -1, 1, 1, 1, -1, -1, 1, 1, 1, -1, -1, -1, -1, -1, -1, -1, -1]
# Random data to test with unknown labels
data = rand(2, 1000) * 7
plot(x = reshape(x[1, :], size(x, 2)), y = reshape(x[2, :], size(x, 2)), group = y, kind = :scatter)
# Try with linear kernel
# This is the classic SVM algorithm
svm = train(x, y)
print_2Ddecision(svm, data)
# Try with a RBF kernel and different sigmas
# ex: 0.6 or 1.0
svm = train(x, y, rbf, sigma)
print_2Ddecision(svm, data)
