#!/usr/bin/env ruby

require 'learning'

class Playback
	def initialize(limit = nil, outfile = 'playback.dat')
		learning_unit = 0
		bayes = Evaluator::Bayes::new(outfile)
		begin
			while (limit and learning_unit < limit)
				logs = []
				open("#{Learning::RAW_DATA_PREFIX}_#{sprintf('%03d', learning_unit)}.dat", 'r'){|file|
					logs = Marshal::load(file.read)
				}
				logs.each{|item|
					log, result = item
					rewards = [1]
					for j in 0...log.size
						next_reward = rewards[-1] * Learning::DISCOUNT_ALPHA
						if next_reward < 0.1 then next_reward = 0.1 end
						rewards << next_reward
					end
					rewards.reverse!
					
					if result[0] > result[1] then
						#黒が勝ち
						log.each_with_index{|state, j|
							bayes.train_black_win(state, rewards[j])
						}
					elsif result[0] < result[1] then
						#白が勝ち
						log.each_with_index{|state, j|
							bayes.train_white_win(state, rewards[j])
						}
					else
						#引き分け
						log.each_with_index{|state, j|
							bayes.train_black_win(state, rewards[j] * 0.3)
							bayes.train_white_win(state, rewards[j] * 0.3)
						}
					end
				}
				learning_unit += 1
			end
		rescue
			p $!
			p $@
		end
		puts "Playbacked #{learning_unit} units"
	end
end

if $0 == __FILE__ then
	case ARGV.size
	when 1; Playback::new(ARGV[0].to_i); return
	when 2; Playback::new(ARGV[0].to_i, ARGV[1]); return
	else; Playback::new; return
	end
end