w r assumptions
play

W R Assumptions Since: 1 The starting state of each player is - PowerPoint PPT Presentation

WAR with Auction Rounds W R Assumptions Since: 1 The starting state of each player is known to all other players 2 Every player knows what all other players played We may assume: Perfect Information Principle Since every player is capable of


  1. WAR with Auction Rounds W R

  2. Assumptions Since: 1 The starting state of each player is known to all other players 2 Every player knows what all other players played We may assume: Perfect Information Principle Since every player is capable of “counting cards”, every player is theoretically capable of knowing the cards in every other player i ’s hand ( H i ) and wins pile ( W i ). W R

  3. WAR as a Repeated Static Game WAR can be represented as a repeated static game; that is, the outcome of a player’s move depends on the other players’ moves, but state is maintained between moves to formulate the outcome of game. W R

  4. Problem Statement You already know how WAR works. Given a game of WAR with N players, devise a strategy to “do well” at WAR. W R

  5. Problem Statement You already know how WAR works. Given a game of WAR with N players, devise a strategy to “do well” at WAR. W R

  6. How well does DP do? When playing against N other DP’s, a DP will have a N Dead Simple Strategy: DummiePlayer (DP) DummiePlayer i (DP) : To play a normal move, select randomly from H i last chance of winning. To play a tie, choose 4 cards randomly from H i and discard the fjrst 3, play the W R

  7. Dead Simple Strategy: DummiePlayer (DP) DummiePlayer i (DP) : To play a normal move, select randomly from H i last N chance of winning. To play a tie, choose 4 cards randomly from H i and discard the fjrst 3, play the How well does DP do? When playing against N other DP’s, a DP will have a 1 W R

  8. An Intelligent Player: SimpleMindedPlayer (SMP) SimpleMindedPlayer i (SMP) : To determine a normal play: the corresponding H ic . To play a tie, follow strategy similar to above, but remove 3 smallest cards from hand fjrst to be discarded. 1 Compute S = { H ic | ∀ j , d [ H ic > H jd ] } . 2 If S ̸ = ∅ , play min ( S ) . 3 Otherwise if, | H i | ⩾ 5 , ∃ j , c [ H ic = max ( H j ) ∧ max 2 ( H i ) ⩾ max 2 ( H j )] , then play 4 Otherwise, play min ( H i ) . W R

  9. H ic . H Q H H H 1 Compute S Player 1 is a SMP. Players 2, 3, 4 are opponents of unknown strategy. j d H ic H jd Q . 2 S , so we play min S Example: SMP Move Computation ( S ̸ = ∅ ) W R

  10. 1 Compute S Player 1 is a SMP. Players 2, 3, 4 are opponents of unknown strategy. H ic j d H ic H jd Q . 2 S , so we play min S . Example: SMP Move Computation ( S ̸ = ∅ ) H 1 = { 8 , 9 , 10 , 10 , Q } H 2 = { 4 , 5 , 7 , 7 , 9 } H 3 = { 2 , 3 } H 4 = { 5 , 9 } W R

  11. Player 1 is a SMP. Players 2, 3, 4 are opponents of unknown strategy. 2 S , so we play min S . Example: SMP Move Computation ( S ̸ = ∅ ) H 1 = { 8 , 9 , 10 , 10 , Q } H 2 = { 4 , 5 , 7 , 7 , 9 } H 3 = { 2 , 3 } H 4 = { 5 , 9 } 1 Compute S = { H ic | ∀ j , d [ H ic > H jd ] } = { 10 , 10 , Q } . W R

  12. Player 1 is a SMP. Players 2, 3, 4 are opponents of unknown strategy. Example: SMP Move Computation ( S ̸ = ∅ ) H 1 = { 8 , 9 , 10 , 10 , Q } H 2 = { 4 , 5 , 7 , 7 , 9 } H 3 = { 2 , 3 } H 4 = { 5 , 9 } 1 Compute S = { H ic | ∀ j , d [ H ic > H jd ] } = { 10 , 10 , Q } . 2 S ̸ = ∅ , so we play min ( S ) = 10 . W R

  13. H j ? Yes, j 1 , so we cannot play from S . max max H j j c H ic 2 ? Yes. H i Player 1 is a SMP. Players 2, 3, 4 are opponents of unknown strategy. 3 See if we can play high: 2 S max . H jd j d H ic H ic 1 Compute S , H jc Q . 3 Play Q . H i Example: SMP Move Computation ( S = ∅ ) H 1 = { 8 , 9 , 10 , 10 , Q } H 2 = { 4 , 5 , 7 , 7 , 9 , Q } H 3 = { 2 , 3 } H 4 = { 5 , 9 } W R

  14. H j ? Yes, j ? Yes. 1 max H j j c H ic 2 Player 1 is a SMP. Players 2, 3, 4 are opponents of unknown strategy. H i , so we cannot play from S . 3 See if we can play high: H i 2 S max , H jc Q . 3 Play Q . max Example: SMP Move Computation ( S = ∅ ) H 1 = { 8 , 9 , 10 , 10 , Q } H 2 = { 4 , 5 , 7 , 7 , 9 , Q } H 3 = { 2 , 3 } H 4 = { 5 , 9 } 1 Compute S = { H ic | ∀ j , d [ H ic > H jd ] } = ∅ . W R

  15. H j ? Yes, j 2 max 3 Play Q . Q . , H jc max H i 3 See if we can play high: 1 H i ? Yes. Player 1 is a SMP. Players 2, 3, 4 are opponents of unknown strategy. j c H ic max H j Example: SMP Move Computation ( S = ∅ ) H 1 = { 8 , 9 , 10 , 10 , Q } H 2 = { 4 , 5 , 7 , 7 , 9 , Q } H 3 = { 2 , 3 } H 4 = { 5 , 9 } 1 Compute S = { H ic | ∀ j , d [ H ic > H jd ] } = ∅ . 2 S = ∅ , so we cannot play from S . W R

  16. H j ? Yes, j 2 max 3 Play Q . Q . , H jc max H i 3 See if we can play high: 1 H i ? Yes. Player 1 is a SMP. Players 2, 3, 4 are opponents of unknown strategy. j c H ic max H j Example: SMP Move Computation ( S = ∅ ) H 1 = { 8 , 9 , 10 , 10 , Q } H 2 = { 4 , 5 , 7 , 7 , 9 , Q } H 3 = { 2 , 3 } H 4 = { 5 , 9 } 1 Compute S = { H ic | ∀ j , d [ H ic > H jd ] } = ∅ . 2 S = ∅ , so we cannot play from S . W R

  17. H j ? Yes, j j c H ic max H j 3 Play Q . Q . , H jc max H i 3 See if we can play high: max 2 Player 1 is a SMP. Players 2, 3, 4 are opponents of unknown strategy. Example: SMP Move Computation ( S = ∅ ) H 1 = { 8 , 9 , 10 , 10 , Q } H 2 = { 4 , 5 , 7 , 7 , 9 , Q } H 3 = { 2 , 3 } H 4 = { 5 , 9 } 1 Compute S = { H ic | ∀ j , d [ H ic > H jd ] } = ∅ . 2 S = ∅ , so we cannot play from S . 1 | H i | ⩾ 5 ? Yes. W R

  18. 3 See if we can play high: Player 1 is a SMP. Players 2, 3, 4 are opponents of unknown strategy. 3 Play Q . Example: SMP Move Computation ( S = ∅ ) H 1 = { 8 , 9 , 10 , 10 , Q } H 2 = { 4 , 5 , 7 , 7 , 9 , Q } H 3 = { 2 , 3 } H 4 = { 5 , 9 } 1 Compute S = { H ic | ∀ j , d [ H ic > H jd ] } = ∅ . 2 S = ∅ , so we cannot play from S . 1 | H i | ⩾ 5 ? Yes. 2 ∃ j , c [ H ic = max ( H j ) ∧ max 2 ( H i ) ⩾ max 2 ( H j )] ? Yes, j = 2 , H jc = Q . W R

  19. 3 See if we can play high: Player 1 is a SMP. Players 2, 3, 4 are opponents of unknown strategy. 3 Play Q . Example: SMP Move Computation ( S = ∅ ) H 1 = { 8 , 9 , 10 , 10 , Q } H 2 = { 4 , 5 , 7 , 7 , 9 , Q } H 3 = { 2 , 3 } H 4 = { 5 , 9 } 1 Compute S = { H ic | ∀ j , d [ H ic > H jd ] } = ∅ . 2 S = ∅ , so we cannot play from S . 1 | H i | ⩾ 5 ? Yes. 2 ∃ j , c [ H ic = max ( H j ) ∧ max 2 ( H i ) ⩾ max 2 ( H j )] ? Yes, j = 2 , H jc = Q . W R

  20. SMP Preformance How well does a SMP do against N other SMPs? A SMP will always die against a fellow SMP, and the game will result with no winners. So an SMP stands a 0% chance here. How well does a SMP do against N DPs? Too hard to fjgure out mathematically, so we wrote a computer simulation to answer this. W R

  21. SMP Preformance How well does a SMP do against N other SMPs? A SMP will always die against a fellow SMP, and the game will result with no winners. So an SMP stands a 0% chance here. How well does a SMP do against N DPs? Too hard to fjgure out mathematically, so we wrote a computer simulation to answer this. W R

  22. Computer Simulation https://github.com/psattiza/war 100,000 simulated games A single SMP battles N DPs, for all N ∈ { 1 , . . . , 100 } . W R

  23. Computer Simulation Number of DPs ( N ) SMP Win Ratio SMP vs. N DPs: Win Ratio 0 . 9 0 . 85 0 . 8 0 . 75 0 20 40 60 80 100 W R

  24. Number of DPs ( N ) Computer Simulation Mean Rounds Played SMP vs. N DPs: Mean Rounds Played 1 , 000 500 0 0 20 40 60 80 100 W R

  25. Computer Simulation SMP vs. N DPs: Variance of Rounds Played Number of DPs ( N ) 500 400 300 200 σ (Rounds) 100 0 0 20 40 60 80 100 W R

  26. Extensions When an SMP determines it cannot win, it chooses it’s lowest card, which may have a high probability of causing a tie against DPs. Design a player good for defeating large quantities of DPs that takes this into account. Throw the game at a ML algorithm and see what happens? More simulation runs! SMP does not take into account W i for either themselves, or other players. Design a CMP (ComplexMindedPlayer) which adds in factors from W i . W R

  27. Extensions When an SMP determines it cannot win, it chooses it’s lowest card, which may have a high probability of causing a tie against DPs. Design a player good for defeating large quantities of DPs that takes this into account. Throw the game at a ML algorithm and see what happens? More simulation runs! SMP does not take into account W i for either themselves, or other players. Design a CMP (ComplexMindedPlayer) which adds in factors from W i . W R

  28. Extensions When an SMP determines it cannot win, it chooses it’s lowest card, which may have a high probability of causing a tie against DPs. Design a player good for defeating large quantities of DPs that takes this into account. Throw the game at a ML algorithm and see what happens? More simulation runs! SMP does not take into account W i for either themselves, or other players. Design a CMP (ComplexMindedPlayer) which adds in factors from W i . W R

  29. Extensions When an SMP determines it cannot win, it chooses it’s lowest card, which may have a high probability of causing a tie against DPs. Design a player good for defeating large quantities of DPs that takes this into account. Throw the game at a ML algorithm and see what happens? More simulation runs! SMP does not take into account W i for either themselves, or other players. Design a CMP (ComplexMindedPlayer) which adds in factors from W i . W R

  30. Questions? W R

Recommend


More recommend