Online Learning Algorithms

OnlineLearning.jl

type Perceptron
	w::Vector{Float64}
	Perceptron(d::Int) = new(zeros(d))
	Perceptron(w::Vector{Float64}) = new(w)
end

type Winnow
	w::Vector{Float64}
	η::Float64
	Perceptron(d::Int, η::Float64) = new(ones(d)./d, η)
	Perceptron(w::Vector{Float64}, η::Float64) = new(w, η)
end

predict(p::Perceptron, x::Vector{Float64}) = sign(p.w ⋅ x)
function learn!(p::Perceptron, x::Vector{Float64}, y::Int)
	yy = p.w ⋅ x
	if yy * y <= 0
		p.w += y*x
	end
	sign(yy)
end

predict(p::Winnow, x::Vector{Float64}) = sign(2(p.w ⋅ x) - 1)
function learn!(p::Winnow, x::Vector{Float64}, y::Int)
	yy = 2(p.w ⋅ x) - 1
	if yy * y <= 0
		p.w .*= exp(-p.η * 2y.*x)
	end
	sign(yy)
end