Presentation is loading. Please wait.

Presentation is loading. Please wait.

Artificial Intelligence for Games Game playing Patrick Olivier

Similar presentations


Presentation on theme: "Artificial Intelligence for Games Game playing Patrick Olivier"— Presentation transcript:

1 Artificial Intelligence for Games Game playing Patrick Olivier

2 Games & search “unpredictable" opponent –specifying a move for every possible reply time limits: unlikely to find goal, approximate

3 Brief history of game playing… Computer considers possible lines of play (Babbage, 1846) Algorithm for perfect play (Zermelo, 1912; Von Neumann, 1944) Finite horizon, approximate evaluation (Zuse, 1945; Wiener, 1948; Shannon, 1950) First chess program (Turing, 1951) Machine learning to improve evaluation accuracy (Samuel, ) Pruning to allow deeper search (McCarthy, 1956)

4 Game tree: 2-player/deterministic/turns

5 Minimax Perfect play for deterministic, perfect-information games Choose position with highest minimax value –best achievable payoff against best play

6 Minimax algorithm

7 Example

8 Solution

9 Properties of minimax Complete –Yes, if tree is finite (chess has specific rules for this) Optimal –Yes, against an optimal opponent. Time complexity –O(b m ) Space complexity –O(bm) (depth-first exploration) for chess, b ≈ 35, m ≈ 100 for “reasonable” games exact solution completely infeasible

10 3  2 2 Optimisation using α - β pruning  3 3 2   5 2

11 α - β pruning algorithm

12 Example

13 Solution <=3 8 0 <= >=

14 Properties of α-β pruning pruning does not affect final result good move ordering improves effectiveness of pruning "perfect ordering," time complexity O(b m/2 ) example of reasoning about which computations are relevant: metareasoning

15 Evaluation functions Heuristic for game state: –order the terminal states correctly –fast to compute –good estimate of “chance of winning” Examples: –chess: weighted linear combination of features –e.g. pawn=1, bishop=3, etc. –based on judgements of chess experts

16 Applying evaluation functions perform minimax and use the evaluation function at the maximum depth –e.g. fixed depth limit, iterative deepening secondary search: –address nonquiescent states –address the horizon effect –select clearly superior moves singular extensions (b=1)

17 Deterministic games in practice Checkers: –Chinook ended 40-year-reign champion (1994) –Perfect play for games of 8 pieces or less (database) Othello: –champions won’t play computers (which are too good) Go: –champions won’t play computers (as not good enough) –branching factor can be as high as > 300 Chess: –Deep Blue defeats Kasparov in 6-game match (1997)

18 Deep blue 30 node IBM supercomputer (& 480 single-chip chess search engines). 3 level architecture mixing software and hardware searching million evolutions of board configurations per sec (reached 330 at one point). Searching with alpha-beta search with complex evaluation function In 3 minutes for each move can search full width 12 deep and some paths up to 40 deep. Optimised with knowledge of opening and end games.

19 Games with an element of chance expectiminimax algorithm chance nodes: use weighted sum of probabilities

20 Samuel’s checker-playing program Arthur Samuel (IBM) Learnt own evaluation function –tuned the weights of a weighted linear function (up to 16 terms) –used comparison with full search Remembered evaluation function values –extends the effective depth of the search


Download ppt "Artificial Intelligence for Games Game playing Patrick Olivier"

Similar presentations


Ads by Google