perceptron.rb

def add_vectors(x, y); x.zip(y).map { |xj, yj| xj + yj }; end

def mult_scalar_to_vector(x, vec); vec.map { |v| v * x }; end

def dot_product(vec1, vec2)
  vec1.zip(vec2).map { |v1, v2| v1 * v2 }.reduce(:+)
end

# Hypothesis function
#
# @param [Array] w Array of weights
# @param [Array] x Array of data
#
# @return [Integer]
def h(w, x)
  dot_product(w, x) >= 0 ? 1 : 0
end

# Perceptron training algorithm.
#
# @param [Array] w     Array of weights
# @param [Array] x     Matrix of weights
# @param [Array] y     Array of classes
# @param [Float] alpha Learning rate
#
# @return [Array] Weights that correctly classify the input
def perceptron(w, x, y, alpha)
  iters     = 0
  converged = false

  until converged
    errors = 0
    iters  += 1

    x.each_with_index do |xj, j|
      hw = h(w, xj)
      if y[j] != hw
        w = add_vectors w, mult_scalar_to_vector(alpha * (y[j] - hw), xj)
        p w
        errors += 1
      end
    end

    break if errors == 0
  end

  [w, iters]
end

if $0 == __FILE__
  w = [0, 2, 1]
  x = [
    [1, 2, 1],
    [2, 2, 1],
    [4, 1, 1],
    [5, 1, 1],
  ]
  y = [1, 0, 1, 0]

  puts "Initial Weights = #{w.inspect}"
  wf, iters = perceptron(w, x, y, 0.5)
  puts "\nFinal Weight = #{wf.inspect}"
  puts "Converged after #{iters} iterations"
end