Presentation is loading. Please wait.

Presentation is loading. Please wait.

Selective Search in Games of Different Complexity Maarten Schadd.

Similar presentations


Presentation on theme: "Selective Search in Games of Different Complexity Maarten Schadd."— Presentation transcript:

1 Selective Search in Games of Different Complexity Maarten Schadd

2 Playing Chess

3 Computer vs. Human Intuition Feelings Only few variations Aggressive Pruning Selective Search Calculator Fast Examines most variations Safe Pruning Brute-Force Search

4 Problem Statement How can we improve selective-search methods in such a way that programs increase their performance in domains of different complexity ?

5 Domains of Different Complexity One-Player Game No Chance Perfect Information Two-Player Game No Chance Perfect Information Two-Player Game Chance or Imperfect Information Multi-Player Game No Chance Perfect Information

6 One-Player Game No Chance – Perfect Information Research question 1 How can we adapt Monte-Carlo Tree Search for a one-player game?

7 One player games No opponent! No uncertainty! Why not use all time at the beginning? Deviation on the score of moves

8 SameGame

9 Single-Player Monte Carlo Tree Search Selection Strategy – Expansion Strategy –Same Simulation Strategy –TabuColourRandom Policy Back-Propagation Strategy –Average Score, Sum of Squared Results and Best Result achieved so far

10 Experiments – Simulation Strategy 250 random positions 10 million nodes in memory

11 One search or several?

12 Parameter tuning

13 Highscores DBS 72,816 SP-MCTS(1)73,998 SP-MCTS(2)76,352 MC-RWS76,764 Nested MC77,934 SP-MCTS(3)78,012 Spurious AI84,414 HGSTS84,718

14 Position 1 – Move 0

15 Position 1 – Move 10

16 Position 1 – Move 20

17 Position 1 – Move 30

18 Position 1 – Move 40

19 Position 1 – Move 52

20 Position 1 – Move 53

21 Position 1 – Move 63

22 Two-Player Game No Chance – Perfect Information Research Question 2 How can we solve a two-player game by using Proof-Number Search in combination with endgame database?

23 Two-Player Game No Chance – Perfect Information Proof-Number SearchEndgame Databases

24 Fanorona

25 Average Branchin Factor

26 Average Number of Pieces

27 Endgame Database Statistics

28 Two-Player Game No Chance – Perfect Information 130,820,097,938 nodes Fanorona solved – Draw!

29 Two-Player Game Chance or Imperfect Information Research Question 3 How can we perform forward pruning at chance nodes in the expectimax framework? 0.90.1

30 Two-Player Game Chance or Imperfect Information ChanceProbCut Predictions based on shallow search

31 ChanceProbCut

32 Stratego

33 Predicting Stratego

34 Node Reduction

35 Performance gain

36 Multi-Player Game No Chance – Perfect Information Research Question 4 How can we improve search for multi-player games?

37 What games do you play?

38 Coalitions

39 Multi-Player Game No Chance – Perfect Information MaxN

40 Multi-Player Game No Chance – Perfect Information Paranoid

41 Max^n 1 22 3333 44444444 6,2,6,35,5,1,24,1,6,81,1,3,17,2,9,54,5,6,71,5,0,85,2,1,4 6,2,6,3 4,1,6,8 7,2,9,5 6,2,6,3 5,2,1,4 7,2,9,5

42 Paranoid 1 7,2,9,5 7-9

43 Paranoid 1 22 3333 44444444 -5-3-110-9-14-12-2 -5 -11 -14 -11<= -14 -11

44 Multi-Player Game No Chance – Perfect Information Best-Reply Search

45 Only 1 opponent plays Chose opponent –Strongest counter move Other opponents have to pass Long term planning Less paranoid Pruning possible

46 Best-Reply Search 1 2,3,4 -5 -11<= -14 -11 2,3,4 11111111111 -4-11-62-7-14-33-4-7 223344223344 1

47 Chinese Checkers

48 Focus

49 Rolit

50 Experiments 3 Players: 6 setups 4 Players: 14 setups 6 Players: 62 setups

51 Validation

52 Average Depth

53

54

55 BRS vs. Max^n

56

57

58 BRS vs. Paranoid

59

60

61 BRS vs. Max^n vs. Paranoid

62

63

64 Multi-Player Game No Chance – Perfect Information New Search Algorithm: Best-Reply Search Ignoring Opponents Long-Term Planning Illegal Positions don’t disturb Generally Stronger than Max^n and Paranoid

65 Conclusions We have investigated four ways to improve selective search methods –Single-Player Monte-Carlo Tree Search –Proof-Number search + endgame databases –ChanceProbCut –Best-Reply Search

66 Future Research Testing selective search methods in other domains –Other games in the same complexity level –Games of other complexity levels

67 Thank you for your attention!


Download ppt "Selective Search in Games of Different Complexity Maarten Schadd."

Similar presentations


Ads by Google