class KNN
  require 'date'
  require_relative 'feature_loader.rb'

  attr_accessor :feature_loader

  def initialize(path)
    self.feature_loader = FeatureLoader.new(path)
  end

  def tests(k)
    @k = k
    correct = 0
    incorrect = 0

    for test in self.feature_loader.test_data
      predicted_class = classify(test)
      if test[0] == predicted_class
        correct += 1
      else
        incorrect += 1
        #puts "Expected Class #{test[0]} - Predicted Class #{predicted_class}"
        #puts "fail"
      end

    end

    puts "Correct: #{correct}"
    puts "Incorrect: #{incorrect}"
    puts "Percentage: #{correct.to_f/self.feature_loader.test_data.length}"
    correct
  end

  def classify(review)
    distances = self.feature_loader.train_data.map { |data|
        test_vector = review[1]
        train_vector = data[1]
        [data[0], euclid_distance(test_vector, train_vector)]
      }
    distances = distances.sort_by{ |x| x[1] }[0..@k].map{|x| x[0]}

    mode_class(distances)
  end

  def mode_class(distances)
    frequency = distances.inject(Hash.new(0)){ |h,v| h[v]+=1; h }
    max = distances.max_by { |v| frequency[v] }
    max
  end

  def euclid_distance(v1, v2)
    raise 'Vectors different length' if v1.length != v2.length
    dis = Math.sqrt(v1.each_with_index.map { |x,i| (v1[i]-v2[i])**2 }.inject(0) {|sum,x| sum + x } )
    dis
  end
end

knn = KNN.new('CA2data.txt')
result = knn.tests(9)
puts "k = #{9} - correct = #{result}"