classifier_simulation.rb

class Mention
  attr_accessor :topic, :prominence, :topic_sentiment, :other_sentiment
  def initialize(topic, prominence, topic_sentiment, other_sentiment)
    @topic = topic
    @prominence = prominence
    if(topic == "Apples")
      topic_sentiment = topic_sentiment ** 3
    elsif topic == "Bananas"
      topic_sentiment = Math.sqrt topic_sentiment
    end
    @topic_sentiment = topic_sentiment
    @other_sentiment = other_sentiment
  end
  def other_prominence
    1.0 - prominence
  end
  def overall_sentiment
    topic_sentiment * prominence + other_sentiment * other_prominence
  end
  def discover_probability(factor)
    prob = prominence ** (1 / factor)
  end
end

def pretty(k,v)

  display = ""
  hash.each do |k,v|
    display << "%10s " % k
    display << "%02.1d" % v
  end
end

Topics = [ "Apples", "Oranges", "Bananas" ]
Items = [ 10, 100, 1000, 10000 ]
Factor = [ 0.25, 0.5, 1, 2, 4 ]

Items.each do |items|
  Factor.each do |factor|
    mentions = []
    items.times do
      mentions << Mention.new(Topics[rand(3)], rand, rand, rand)
    end
    correct = Hash.new
    Topics.each { |t| correct[t] = 0.0 }
    numCorrect = Hash.new
    Topics.each { |t| numCorrect[t] = 0.0 }
    approximate = Hash.new
    Topics.each { |t| approximate[t] = 0.0 }
    numApprox = Hash.new
    Topics.each { |t| numApprox[t] = 0.0 }
    randomized = Hash.new
    Topics.each { |t| randomized[t] = 0.0 }
    numRandom = Hash.new
    Topics.each { |t| numRandom[t] = 0.0 }
    
    mentions.each do |m|
      correct[m.topic] = correct[m.topic] + m.prominence * m.topic_sentiment
      numCorrect[m.topic] += m.prominence
      approximate[m.topic] = approximate[m.topic] +
        m.discover_probability(factor) * m.overall_sentiment
      numApprox[m.topic] += m.discover_probability(factor)
      if rand < m.discover_probability(factor) then
        randomized[m.topic] += m.overall_sentiment
        numRandom[m.topic] += 1.0
      end
    end

    puts ""
    puts "Calculating for #{items} items, factor of #{factor}:"
    
    maxCorrect = correct.values.inject { |m,x| m > x ? m : x }
    scaled = []
    Topics.each_with_index do |v, i|
      scaled[i] = "%10s %2.0f%%  " % [v, 100 * correct[v] / numCorrect[v] ]
    end
    puts "Correct #{scaled.to_s}"
    maxApprox = approximate.values.inject { |m,x| m > x ? m : x}
    maxCorrect = correct.values.inject { |m,x| m > x ? m : x }
    Topics.each_with_index do |v, i|
      scaled[i] = "%10s %2.0f%%  " % [v, 100 * approximate[v] / numApprox[v] ]
    end
    puts "Approx  #{scaled.to_s}"
    maxRand = randomized.values.inject { |m,x| m > x ? m : x}
    maxCorrect = correct.values.inject { |m,x| m > x ? m : x }
    Topics.each_with_index do |v, i|
      scaled[i] = "%10s %2.0f%%  " % [v, 100 * randomized[v] / numRandom[v] ]
    end
    puts "Random  #{scaled.to_s}"
  end
end