#!/usr/bin/env ruby

# -*- coding: utf-8 -*-

# Reference

# * http://d.hatena.ne.jp/echizen_tm/20110606/1307378609

# * http://d.hatena.ne.jp/lettas0726/20111005/1317803286

module MachineLearning

DEFAULT_BIAS = 1

class Perceptron

def initialize(w = {}, bias = DEFAULT_BIAS)

@w = w.has_key?(:bias) ? w : w.merge({bias: bias})

end

def predict(vector)

vector.inject(0) { |sum, (key, value)|

@w.has_key?(key) ? (sum + @w[key] * value) : sum

}

end

def train(vector, fact)

vector[:bias] = DEFAULT_BIAS unless vector.has_key?(:bias)

sum = predict(vector)

return unless (sum * fact < 0)

vector.each do |key, value|

@w[key] = 0 unless @w.has_key?(key)

@w[key] += fact * value

end

end

end

end

if __FILE__ == $0

train_data = [

[{r: 255, g: 0, b: 0}, 1],

[{r: 0, g: 255, b: 255}, -1],

[{r: 0, g: 255, b: 0}, -1],

[{r: 255, g: 0, b: 255}, 1],

[{r: 0, g: 0, b: 255}, -1],

[{r: 255, g: 255, b: 0}, 1],

]

machine = MachineLearning::Perceptron.new({r: 0, g: 0, b: 0})

100.times do

train_data.each do |vector, fact|

machine.train(vector, fact)

end

end

puts "Input color code. (ex) 102 204 255"

begin

while (print "> "; input = gets) do

input = input.chomp.strip

next unless input =~ /^\d{1,3} +\d{1,3} +\d{1,3}$/

r, g, b = input.split(' ').map { |s| s.to_i }

x = {r: r, g: g, b: b, bias: 1}

t = machine.predict(x)

if t >= 0

puts ">> warm: #{t}"

else

puts ">> cold: #{t}"

end

end

rescue Interrupt

puts

puts "Bye."

end

end