chess-algorithms-theory-and-practice_ver2017.pdf
- 1. Chess Algorithms Theory and Practice Rune Djurhuus Chess Grandmaster runed@ifi.uio.no / runedj@microsoft.com September 20, 2017 1
- 2. Content • Complexity of a chess game • Solving chess, is it a myth? • History of computer chess • Chess compared to Go • Search trees and position evaluation • Minimax: The basic search algorithm • Negamax: «Simplified» minimax • Node explosion • Pruning techniques: – Alpha-Beta pruning – Analyze the best move first – Killer-move heuristics – Zero-move heuristics • Iterative deeper depth-first search (IDDFS) • Search tree extensions • Transposition tables (position cache) • Other challenges • Endgame tablebases • Demo 2
- 3. Complexity of a Chess Game • 20 possible start moves, 20 possible replies, etc. • 400 possible positions after 2 ply (half moves) • 197 281 positions after 4 ply • 713 positions after 10 ply (5 White moves and 5 Black moves) • Exponential explosion! • Approximately 40 legal moves in a typical position • There exists about 10120 possible chess games 3
- 4. Solving Chess, is it a myth? Chess Complexity Space • The estimated number of possible chess games is 10120 – Claude E. Shannon – 1 followed by 120 zeroes!!! • The estimated number of reachable chess positions is 1047 – Shirish Chinchalkar, 1996 • Modern GPU’s performs 1013 flops • If we assume one million GPUs with 10 flops per position we can calculate 1018 positions per second • It will take us 1 600 000 000 000 000 000 000 years to solve chess Assuming Moore’s law works in the future • Todays top supercomputers delivers 1016 flops • Assuming 100 operations per position yields 1014 positions per second • Doing retrograde analysis on supercomputers for 4 months we can calculate 1021 positions. • When will Moore’s law allow us to reach 1047 positions? • Answer: in 128 years, or around year 2142! 4 http://chessgpgpu.blogspot.no/2013/06/solving-chess- facts-and-fiction.html
- 5. History of Computer Chess • Chess was a good fit for computers: – Clearly defined rules – Game of complete information – Easy to evaluate (judge) positions – Search tree is not too small or too big • 1950: Programming a Computer for Playing Chess (Claude Shannon) • 1951: First chess playing program (on paper) (Alan Turing) • 1958: First computer program that can play a complete chess game • 1981: Cray Blitz wins a tournament in Mississippi and achieves master rating • 1989: Deep Thought loses 0-2 against World Champion Garry Kasparov • 1996: Deep Blue wins a game against Kasparov, but loses match 2-4 • 1997: Upgraded Dee Blue wins 3.5-2.5 against Kasparov • 2005: Hydra destroys GM Michael Adams 5.5-0.5 • 2006: World Champion Vladimir Kramnik looses 2-4 against Deep Fritz (PC chess engine) • 2014: Magnus Carlsen launches “Play Magnus “ app on iOS where anyone can play against a chess engine that emulates the World Champion’s play at 21 different ages (5 to 25 years). 5
- 6. Chess Compared to Go • Go is played on a 19x19 square board where a new stone is placed on any free square each move (and never moved around) • Go has a much higher branching factor (starting with 361 and slowly descending) and much more complicated leaf node evaluation • For many years the best Go programs had amateur rating only • In 2016 Alpha Go surprisingly beat Lee Sedol (9-dan profession) 4-1 using a combination of machine learning (deep neural network) and Monte Carlo tree search algorithm. • Alpha Go beat Ke Jie (ranked no. 1 in the world) 3-0 in 2017 and retired afterwards. 6
- 7. Search Trees and Position Evaluation • Search trees (nodes are positions, edges are legal chess moves) • Leaf nodes are end positions which needs to be evaluated (judged) • A simple judger: Check mate? If not, count material • Nodes are marked with a numeric evaluation value 7
- 8. Minimax: The Basic Search Algorithm • Minimax: Assume that both White and Black plays the best moves. We maximizes White’s score • Perform a depth-first search and evaluate the leaf nodes • Choose child node with highest value if it is White to move • Choose child node with lowest value if it is Black to move • Branching factor is 40 in a typical chess position White Black White Black White ply = 0 ply = 1 ply = 2 ply = 3 ply = 4 8
- 9. NegaMax – “Simplified” Minimax Minimax int maxi( int depth ) { if ( depth == 0 ) return evaluate(); int max = -∞; for ( all moves) { score = mini( depth - 1 ); if( score > max ) max = score; } return max; } NegaMax int negaMax( int depth ) { if ( depth == 0 ) return evaluate(); int max = - ∞; for ( all moves) { score = -negaMax( depth - 1 ); if( score > max ) max = score; } return max; } 9 int mini( int depth ) { if ( depth == 0 ) return -evaluate(); int min = + ∞; for ( all moves) { score = maxi( depth - 1 ); if( score < min ) min = score; } return min; } max(a, b) == -min(-a, -b)
- 10. Node explosion ➢ 10 M nodes per second (nps) is realistic for modern chess engines ➢ Modern engines routinely reach depths 25-35 ply at tournament play ➢ But they only have a few minutes per move, so they should only be able to go 5-6 ply deep ➢ How do they then get to depth 25 so easily? 10 Depth Node count Time at 10M nodes/sec 1 40 0.000004 s 2 1 600 0.00016 s 3 64 000 0.0064 s 4 2 560 000 0.256 s 5 102 400 000 10.24 s 6 4 096 000 000 6 min 49,6 s 7 163 840 000 000 4 h 33 min 4 s 8 6 553 600 000 000 7 d 14 h 2 min 40 s A typical middle-game position has 40 legal moves.
- 11. Pruning Techniques • The complexity of searching d ply ahead is O(b*b*…*b) = O(bd) • With a branching factor (b) of 40 it is crucial to be able to prune the search tree 11
- 12. Alpha-Beta Pruning “Position is so good for White (or Black) that the opponent with best play will not enter the variation that gives the position.” • Use previous known max and min values to limit the search tree • Alpha value: White is guaranteed this score or better (start value: -∞) • Beta value: Black is guaranteed this score or less (start value: +∞) • If Alpha is higher than Beta, then the position will never occur assuming best play • If search tree below is evaluated left to right, then we can skip the greyed- out sub trees • Regardless of what values we get for the grey nodes, they will not influence the root node score White Black White Black White ply = 0 ply = 1 ply = 2 ply = 3 ply = 4 12
- 13. Analyze the Best Move First • Even with alpha-beta pruning, if we always start with the worst move, we still get O(b*b*..*b) = O(bd) • If we always start with the best move (also recursive) it can be shown that complexity is O(b*1*b*1*b*1…) = O(bd/2) = O(√bd) • We can double the search depth without using more resources • Conclusion: It is very important to try to start with the strongest moves first 13
- 14. Killer-Move Heuristics • Killer-move heuristics is based on the assumption that a strong move which gave a large pruning of a sub tree, might also be a strong move in other nodes in the search tree • Therefore we start with the killer moves in order to maximize search tree pruning 14
- 15. Zero-Move Heuristics • Alpha-Beta cutoff: “The position is so good for White (or Black) that the opponent with best play will avoid the variation resulting in that position” • Zero-Move heuristics is based on the fact that in most positions it is an advantage to be the first player to move • Let the player (e.g. White) who has just made a move, play another move (two moves in a row), and perform a shallower (2-3 ply less) and therefore cheaper search from that position • If the shallower search gives a cutoff value (e.g. bad score for White), it means that most likely the search tree can be pruned at this position without performing a deeper search, since two moves in a row did not help • Very effective pruning technique! • Cavecats: Check and endgames (where a player can be in “trekktvang” – every move worsens the position) 15
- 16. Iterative Deeper Depth-First Search (IDDFS) • Since it is so important to evaluate the best move first, it might be worthwhile to execute a shallower search first and then use the resulting alpha/beta cutoff values as start values for a deeper search • Since the majority of search nodes are on the lowest level in a balanced search tree, it is relatively cheap to do an extra shallower search 16
- 17. Search Tree Extensions • PC programs today can compute 25-35 ply ahead (Deep Blue computed 12 ply against Kasparov in 1997, Hydra (64 nodes with FPGAs) computed at least 18 ply) • It is important to extend the search in leaf nodes that are “unstable” • Good search extensions includes all moves that gives check or captures a piece • The longest search extensions are typically double the average length of the search tree! 17
- 18. Transposition Table • Same position will commonly occur from different move orders • All chess engines therefore has a transposition table (position cache) • Implemented using a hash table with chess position as key • Doesn’t have to evaluate large sub trees over and over again • Chess engines typically uses half of available memory to hash table – proves how important it is 18
- 19. Other challenges • Move generator (hardware / software) – Hydra (64 nodes Xeon cluster, FPGA chips) computed 200 millions positions per second, approximately the same as Deep Blue (on older ASIC chip sets) – Hydra computed 18+ ply ahead while Deep Blue only managed 12 (Hydra prunes search tree better) – Komodo 10 chess engine calculates 3-4 mill moves/second on my Surface Book (Intel i7 @ 2.6 GHz with 3 cores) and computes 20+ ply in less than 5 seconds and 25+ ply in less than 30 seconds • Efficient data structure for a chess board (0x88, bitboards) • Opening library suited for a chess computer • Position evaluation: • Traditionally chess computers has done deep searches with a simple evaluation function • But one of the best PC chess engines today, Rybka, sacrifices search depth for a complex position evaluation and better search heuristics 19
- 20. Endgame Tablebases • Chess engines plays endgames with 3-7 pieces left on the board perfectly by looking up best move in huge tables • These endgame databases are called Tablebases • Retrograde analyses: Tablebases are generated by starting with final positions (check mate, steal mate or insufficient mating material (e.g. king vs. king)) and then compute backwards until all nodes in search tree are marked as win, draw or loose • Using complex compression algorithms (Nalimov, Syzygy) • The newer Syzygy compression format uses less than 200 GB for all endgames with up to 6 piezes (compared to over 1 TB for Nalimov tablebases) 20
- 21. Lomonosov Tablebases • All 7 piece endgames (except 6 pieces vs a lone king) calculated for the first time in 2013 on the Lomonosov supercomputer in Moscow State University. • Took 6 months to generate • Needed 140 TB of storage • Longest forced mate: White to mate in 545 moves! • See http://chessok.com/?page_id=27966, http://tb7.chessok.com/ 21
- 22. Demo • Demo: ChessBase with chess engine Komodo 10 and Stockfish 7 • Best open source UCI chess engine (and may be best overall): – Stockfish (stockfishchess.org) 22
- 23. Thank you Presenter: Rune Djurhuus Contact: runed@ifi.uio.no runedj@microsoft.com Version: Autumn 2017 23