James Edward Gray II’s Ruby Quiz #16 was to implement Rock, Paper, Scissors playing classes to compete on a playing field managed by a given Game class. Today we revisit this quiz for a bit of coding fun. We’ll implement some simple players and move on to some basic metaprogramming techniques and write players that manipulate each other or the game itself.

If you want to follow along:

1
2
3
4
5
6
7
8
9

% git clone git://github.com/sdball/ruby_quiz.git
% cd ruby_quiz/16
# ruby -I . rock_paper_scissors.rb [player files to run]
% ruby -I . rock_paper_scissors.rb players/
% ruby -I . rock_paper_scissors.rb players/always_rock.rb players/always_scissors.rb
AlwaysRock vs. AlwaysScissors
  AlwaysRock: 1000
  AlwaysScissors: 0
  AlwaysRock Wins

Simple Player

Right. So here’s what a very simple player class looks like.

1
2
3
4
5
6

# Michael Bluth
class AlwaysRock < Player
  def choose
    :rock
  end
end

The rock_paper_scissors.rb defines a Player class, a Game class, and provides a script to run the contest. Our own custom player classes just have to inherit from the given Player class and implement a choose method that returns :rock, :paper, or :scissors. Our players can optionally implement a result method that is used as a callback from the game to allow our players to see the result of their play. If our players have an initialize method it is called with the classname of the opponent.

So, back to Michael Bluth (or Poor Predictable Bart) who always chooses rock. His choose method just returns the symbol :rock. Easy. Let’s write a player to beat him. Also easy.

1
2
3
4
5
6

# GOB
class AlwaysPaper < Player
  def choose
    :paper
  end
end

GOB will beat Michael Bluth 100% of the time.

1
2
3
4
5

% ruby -I . rock_paper_scissors.rb players/always_rock.rb players/always_paper.rb
AlwaysRock vs. AlwaysPaper
  AlwaysRock: 0
  AlwaysPaper: 1000
  AlwaysPaper Wins

Ok, now let’s write a player to beat them both. Consistently.

Reactive Player

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

class Reactive < Player
  MOVE_THAT_BEATS = {
    rock: :paper,
    paper: :scissors,
    scissors: :rock
  }

  def choose
    @my_move || :paper
  end

  def result(my_move, opponent_move, outcome)
    @my_move = MOVE_THAT_BEATS[opponent_move]
  end
end

This player is kind of interesting, but still straightforward. Reactive has a constant MOVE_THAT_BEATS that maps moves to the move that would beat them. Reactive then uses that knowledge to play the move that beats the last move played by his opponent (or :paper). This strategy should prove extremely effective against a player who always makes the same move.

1
2
3
4
5
6
7
8
9

% ruby -I . rock_paper_scissors.rb players/always_rock.rb players/always_paper.rb players/reactive.rb
AlwaysRock vs. Reactive
  AlwaysRock: 0
  Reactive: 1000
  Reactive Wins
AlwaysPaper vs. Reactive
  AlwaysPaper: 0.5
  Reactive: 999.5
  Reactive Wins

Yep, Always Rock loses every game because Reactive plays paper from the start. Paper gets in one draw, then loses every game afterwards.

Now, let’s write a player that will beat all of the players thus far.

Random Player

Theoretically, the best strategy in Rock, Paper, Scissors is to be completely random. That’s a pretty easy strategy for a computer to play (although surprisingly difficult for a human) so let’s code that up next.

1
2
3
4
5

class RandomPlayer < Player
  def choose
    [:rock, :paper, :scissors].shuffle.first
  end
end

Well that was easy. Let’s pit Random against some of the players so far.

1
2

% ruby -I . rock_paper_scissors.rb players/always_rock.rb players/random.rb
# out of 10 runs, Random won 8 games

1
2

% ruby -I . rock_paper_scissors.rb players/reactive.rb players/random.rb
# out of 10 runs, Random won 8 games

Fun, but not all that interesting. How about we write a player who watches their opponent and builds a strategy to defeat them? This player is going to be a bit more complex than those we’ve written so far.

Pattern Matching Player

The plan: build up a pattern memory of player moves, opponent moves, and the next move the opponent played under those conditions. After every move:

• remember the moves that were just played
• record the previous game’s moves and the opponents move
• use that record of game patterns to determine the opponent’s next likely move and play the move that beats it

This pattern set will allow us to make observations such as “the last ten times I played rock and my opponent played paper, my opponent’s next move was scissors” or “out the last 97 times I played rock and my opponent played scissors my opponent played scissors next 96% of the time”.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56

class PatternMatching < Player
  MOVE_THAT_BEATS = {
    rock: :paper,
    paper: :scissors,
    scissors: :rock
  }
  def initialize(opponent)
    @first_game = true
    @patterns = {
      rock: {},
      paper: {},
      scissors: {}
    }
  end

  def choose
    @my_move || :rock
  end

  def result(mine, theirs, outcome)
    store_moves(mine, theirs)
    plan_next_move
    @first_game = false
  end

  private
  def store_moves(mine, theirs)
    unless @first_game
      store_pattern(mine, theirs)
    end
    @my_last_move = mine
    @their_last_move = theirs
  end

  def store_pattern(mine, theirs)
    if @patterns[@my_last_move][@their_last_move]
      @patterns[@my_last_move][@their_last_move] << theirs
    else
      @patterns[@my_last_move][@their_last_move] = [theirs]
    end
  end

  def plan_next_move
    @my_move = MOVE_THAT_BEATS[their_likely_next_move] || :rock
  end

  def their_likely_next_move
    their_moves = @patterns[@my_last_move][@their_last_move]
    return [:rock, :paper, :scissors].shuffle.first if their_moves.nil?
    count = Hash.new(0)
    their_moves.each do |move|
      count[move] += 1
    end
    count.sort_by {|key, value| value}.last.first
  end
end

This guy looks complicated, but the only tricky logic is around the whole “Is this the first game or not?” I’m not entirely happy with the way I’ve worked it out here, but it gets the job done. store_pattern and their_likely_next_move are the two key methods. The move patterns are stored as nested hashes [my move][opponent move] => [array of observed moves] e.g. `[:rock][:paper] => [:scissors, :paper, :scissors, :scissors, :scissors, :paper]

their_likely_next_move just looks into the pattern and counts up the data seen so far. If there’s no data yet, it just guesses randomly.

Let’s see how our pattern matcher fares!

1
2
3
4
5

% ruby -I . rock_paper_scissors.rb players/always_rock.rb players/pattern_matching.rb
AlwaysRock vs. PatternMatching
  AlwaysRock: 1.0
  PatternMatching: 999.0
  PatternMatching Wins

Not bad! Our pattern matcher correctly determines the simple pattern employed by Always Rock

1
2
3
4
5

% ruby -I . rock_paper_scissors.rb players/reactive.rb players/pattern_matching.rb
Reactive vs. PatternMatching
  Reactive: 3.5
  PatternMatching: 996.5
  PatternMatching Wins

Bam! Our pattern matcher handily figures out what our simple Reactive player is likely to do. Now for a real challenge.

1
2
3
4
5
6
7
8
9
10
11
12

% ruby -I . rock_paper_scissors.rb players/random.rb players/pattern_matching.rb
RandomPlayer vs. PatternMatching
  RandomPlayer: 491.0
  PatternMatching: 509.0
  PatternMatching Wins

% ruby -I . rock_paper_scissors.rb players/random.rb players/pattern_matching.rb
RandomPlayer vs. PatternMatching
  RandomPlayer: 518.5
  PatternMatching: 481.5
  RandomPlayer Wins
# it goes back and forth

So PatternMatching fares well against RandomPlayer but can’t consistently win. Let’s make a player who can consistently and completely defeat RandomPlayer. Let’s make a player who cheats.

Cheater

1
2
3
4
5
6
7
8
9

class Cheater < Player
  def initialize(opponent)
    Kernel.const_get(opponent).class_eval('def choose; :scissors; end')
  end

  def choose
    :rock
  end
end

Mwahaha! Cheater utilizes that so far unused ability to get the name of his opponent. Using that and some magic Cheater hypnotizes his opponent into always playing scissors.

Does it work? Absolutely.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23

% ruby -I . rock_paper_scissors.rb players/always_rock.rb players/cheater.rb
AlwaysRock vs. Cheater
  AlwaysRock: 0
  Cheater: 1000
  Cheater Wins

% ruby -I . rock_paper_scissors.rb players/reactive.rb players/cheater.rb
Reactive vs. Cheater
  Reactive: 0
  Cheater: 1000
  Cheater Wins

% ruby -I . rock_paper_scissors.rb players/pattern_matching.rb players/cheater.rb
PatternMatching vs. Cheater
  PatternMatching: 0
  Cheater: 1000
  Cheater Wins

% ruby -I . rock_paper_scissors.rb players/random.rb players/cheater.rb
RandomPlayer vs. Cheater
  RandomPlayer: 0
  Cheater: 1000
  Cheater Wins

So what’s going on here? To answer that, let’s dive into irb

1
2
3
4
5

% irb -I .
1.9.3p0 :001 > require 'rock_paper_scissors'
 => true
1.9.3p0 :002 > require 'players/always_rock'
 => true

Ok, we’re in irb and we’ve got our game and a player loaded. Let’s see what Cheater is up to.

1

Kernel.const_get(opponent).class_eval('def choose; :scissors; end')

1
2
3
4

1.9.3p0 :003 > Kernel.const_get('AlwaysRock')
 => AlwaysRock
1.9.3p0 :004 > Kernel.const_get('AlwaysRock').class
 => Class

const_get checks a module for a constant with the given name. Since a class is defined as a constant and Kernel sits over every class, we can ask Kernel to find our opponent’s class for us. Which is what we’re doing here.

Basically Kernel.const_get('AlwaysRock') is an easy way to turn the string “AlwaysRock” into a constant.

Now class_eval. This method just says to a class, “evaluate this string as if it were written as part of your code.”

1
2
3
4

1.9.3p0 :005 > String.class_eval('def magic; "magic!"; end')
 => nil
1.9.3p0 :006 > "hi".magic
 => "magic!"

Put those pieces together, and our Cheater:

1. Takes the string of his opponents name and turns it into a constant to get at his opponent’s class.
2. Rewrites his opponent with a new choose method that he controls.
3. Beats his opponents modified choose method.

1
2
3
4
5
6

1.9.3p0 :007 > AlwaysRock.new('Cheater').choose
 => :rock
1.9.3p0 :008 > Kernel.const_get('AlwaysRock').class_eval('def choose; :scissors; end')
=> nil
1.9.3p0 :009 > AlwaysRock.new('Cheater').choose
=> :scissors

Insidious! Even worse, if you’re running a contest with more than two players the cheater permanently modifies his opponents into always playing their redefined choose method.

Let’s level the playing field and force the cheater to play fair.

Level Playing Field

LevelPlayingField avoids being manipulated by the cheater by having a strong will to resist Cheater’s hypnotism. Cheater works by modifying the class of his opponents, so LevelPlayingField doesn’t have a choose method defined by his class. LevelPlayingField defines his own choose method on initialization.

1
2
3
4
5
6
7
8
9
10

class LevelPlayingField < Player
  def initialize(opponent)
    # prevent cheater from winning by waiting to pick a strategy
    self.class.class_eval do
      def choose
        [:rock, :paper, :scissors].shuffle.first
      end
    end
  end
end

Yes, that’s right. LevelPlayingField uses the same trick that Cheater does, but on himself to try and guarantee he’s playing his own strategy.

1
2
3
4
5
6
7
8

1.9.3p0 :010 > require 'players/level_playing_field'
 => true
1.9.3p0 :011 > LevelPlayingField.new('Cheater').choose
=> :paper
1.9.3p0 :012 > Kernel.const_get('LevelPlayingField').class_eval('def choose; :scissors; end')
=> nil
1.9.3p0 :011 > LevelPlayingField.new('Cheater').choose
=> :rock # LevelPlayingField is random so you actually might see :scissors here

1
2
3
4
5

% ruby -I . rock_paper_scissors.rb players/cheater.rb players/level_playing_field.rb
Cheater vs. LevelPlayingField
  Cheater: 496.5
  LevelPlayingField: 503.5
  LevelPlayingField Wins

Now this works, but there’s one flaw. Our LevelPlayingField player only succeeds when he’s initialized after the Cheater.

1
2
3
4
5

% ruby -I . rock_paper_scissors.rb players/level_playing_field.rb players/cheater.rb
LevelPlayingField vs. Cheater
  LevelPlayingField: 0
  Cheater: 1000
  Cheater Wins

If Cheater comes into play after LevelPlayingField has completed his trick to try and isolate his choose method, then the Cheater will still just override LevelPlayingField’s choose method. It’s a remarkably tough strategy to defeat. Any trick we use to cleverly hide our choose method from the Cheater can ultimately be sidestepped by the Cheater just supplying a new choose method.

Even if we write a player who removes the ability for the Cheater to cheat, if he’s loaded first: he wins.

1
2
3
4
5
6
7
8
9

class LevelPlayingField < Player
  def initialize(opponent)
    # rewrite class_eval to prevent cheating
    Module.class_eval('def class_eval(obj) end')
  end
  def choose
    [:rock, :paper, :scissors].shuffle.first
  end
end

So let’s bring in a player who always wins. A player who out-cheats the cheater. A player who modifies the game itself.

Batman

1
2
3
4
5
6
7
8
9
10
11
12

class Batman < Player
  def initialize(opponent)
    Game.class_eval do
      def play( match )
        match.times do
          next win @player1, :rock, :scissors if @player1.instance_of? Batman
          next win @player2, :rock, :scissors if @player2.instance_of? Batman
        end
      end
    end
  end
end

Batman’s playing strategy is to enter the Game class itself and ensure that he’s the winner no matter what’s been played. Not only that, but he completely alters the game for all other player matches. They all get recorded as draws, since only Batman can win.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17

% ruby -I . rock_paper_scissors.rb players/cheater.rb players/batman.rb
Cheater vs. Batman
  Cheater: 0
  Batman: 1000
  Batman Wins

% ruby -I . rock_paper_scissors.rb players/batman.rb players/cheater.rb
Batman vs. Cheater
  Batman: 1000
  Cheater: 0
  Batman Wins

% ruby -I . rock_paper_scissors.rb players/batman.rb players/reactive.rb
Batman vs. Reactive
  Batman: 1000
  Reactive: 0
  Batman Wins

Gosh, makes you almost feel bad for his opponents.

I hope you’ve enjoyed this bit of fun. On Thursday I plan to take this small glimpse into the world of metaprogramming a bit further.