Monte Carlo instead of Alpha-Beta?

by Stephan Oliver Platz
1/30/2019 – Until now, chess programs mostly worked according to the so-called alpha-beta pruning search algorithm. The spectacular Deep Mind project Alpha Zero is based on the Monte Carlo Tree Search (MCTS). The Komodo developers, too, are now increasingly relying on this method — with success. STEPHEN OLIVER PLATZ explains the difference and shows the progress. | Photo: (Komodo Dragon) Ryan Vaarsi, CC BY 2.0

In computer chess there is no getting past Komodo, a two-time ICGA Computer World Chess Champion. Find out how Komodo can take your game to the next level!

Lessons from Alpha Zero

Just over a year ago, the Alpha Zero chess program made headlines by sensationally defeating Stockfish 8. In the published games the playing style was really impressive, reminiscent of Anderssen and Morphy with positional sacrifices of pawns and pieces. Many chess aficionados would like to buy Alpha Zero if it were available and could be run on conventional computers. While that's not an option, fortunately, Komodo MCTS is a chess program available for an affordable price that uses similar techniques as Alpha Zero and can be installed on consumer PCs. This version has made great progress since its first release, as can be seen from two remarkable games played in December 2018.

A tactical masterpiece

In the first round of Division 2 of the current TCEC 14 tournament, which is regarded by many as the actual, albeit unofficial World Chess Computer Championship, Komodo MCTS defeated Nirvana 2.4 in a spectacular way. Playing with the white pieces against a Sicilian Defence, Komodo sacrificed a pawn, then the exchange and finally a bishop, but Black was already lost after grabbing the pawn:

 
New ...
Open...
Share...
Layout...
Flip Board
Settings
MoveNResultEloPlayers
Replay and check the LiveBook here
1.e4 c5 Playing the Sicilian Defence requires thorough preparation, but the electronic monsters should be able to handle it, right? 2.Nf3 e6 3.d4 cxd4 4.Nxd4 Nc6 5.Nc3 a6 6.Be2 Qc7 7.f4 Nf6!? 7...b5! 8.Nxc6 8.e5!? 8...dxc6 9.0-0 b5 10.Be3 Be7 11.e5 Nd5 12.Nxd5 cxd5 13.c3 Bd7 14.Qd2 a5 15.Bd4 0-0 Castling into the enemy's attack. Can this be good? 16.a3 Bc5 Strategically this move is problematic, because White might exchange his own bad bishop for Black's good one. But on the other hand a king attack is threatening, and it might be advisable to reduce the white attacking potential by exchanging pieces. By playing b5-b4 Black might establish some counter play and give his bad bishop a little more scope, but this is not so easy to accomplish, e.g. 16...Rfb8 17.Bd3 g6 18.g4 Bf8 19.f5 b4 20.axb4 20.fxg6 fxg6 21.Qg5 Qd8!= 20...axb4 21.Rxa8 Rxa8 22.Qf4! this is better than 22.cxb4 Qb8 23.Bc3 Qb6+ 24.Qf2 24.Rf2? exf5 25.gxf5? Ra1+ 26.Kg2 d4!-+ 24...Qxf2+ 25.Rxf2 Rb8 , for White cannot defend the extra pawn without admitting a draw by a threefold repetition: 26.Rf4 Bh6 27.Rf3 27.Rd4? Be3+ 27...Bf8 28.Rf4 Bh6 , etc. 22...Ba4 23.h4 bxc3 24.bxc3 Qe7 25.h5 g5 26.Qf3 h6 and White is better. He can open the f-line at a favourable moment, where the queen and rook are already doubled and regroup his forces beforehand without Black being able to do much himself. 17.Rfc1 After 17.Bxc5 Qxc5+ 18.Qd4 White has a little advantage in the endgame due to his good bishop. 17...Rfc8 18.g4 Qd8 A logical move is 18...Bxd4+ in order to exchange pieces 19.cxd4 or 19.Qxd4 Qc5 exchanging queens 19...Qb6 20.Kf2 if White wants to avoid the exchange of rooks he must leave the c-file to his adversary 20.Rf1 b4! 21.f5 bxa3 22.Rxa3 Qb4! 20...Rxc1 21.Rxc1 b4! 19.Kg2 Rab8 19...Qh4!? 20.Qe3! Qe7 20.Kg3 Bf8 20...b4!? 21.cxb4 Bxd4 22.Rxc8 Qxc8 23.Qxd4 axb4 24.a4! 21.b4! Thus Komodo prevents any possible black counter play by b5-b4, and in addition the bishop d7 remains blocked by his own pawn on b5. Be7 22.Bd3 g6? You shouldn't play a move like that unless you are forced to. Now White can easily open files on the kingside. 22...Rc6 was an option to control the sixth rank as a precaution, which might become important after a later f4-f5. 23.h4! Bxh4+? This looks very much like suicide. A human player might have tried 23...h6 with the idea 24.h5 g5 25.Rf1± 24.Kf3+- The open h-line is worth much more than the lost pawn. Furthermore, the black bishop on h4 is in danger of being cut off by g4-g5. Be7 24...Qf8? 25.g5! and Black has no defence against the threatening loss of a piece by Rh1, Qh2 or Kg4. 25.Rh1 Threatening 26.Qh2 attacking h7 or even 26.Rh3 followed by Rah1 and Qh2 placing both rooks and the queen on the open h-file. Bf8 25...Qf8 26.f5! 26.Rh3 is good enough, too, e.g. f5 27.gxf5 exf5 27...gxf5? 28.Qh2+- 28.e6! Bxe6 29.Qe2 Bd7 30.Qe5+- 26...f6 26...exf5 27.gxf5+- 27.Rxh7! Kxh7 28.Rh1+ Kg8! 29.Qh2 Kf7 29...Qg7? 30.g5!+- 30.fxe6+ Ke8! 30...Bxe6? 31.Bxg6+! Kxg6 32.Qh7+ Kg5 33.Qh5# 30...Kxe6? 31.Bf5+! gxf5 32.gxf5+ Kxf5 33.Qh3+ mate in 2 31.Bxg6+ Kd8 32.exd7 Kxd7 32...fxe5+ 33.Bf5+- 33.exf6 Bxf6 34.Bf5+ Kc6! 34...Kd8? 35.Bc5! Be7 36.Qe5+- 34...Ke8? 35.Re1+ Be7 36.Qd6 Qf7 37.Bf6+- 34...Ke7? 35.Qe2+ Kd8 36.Qe6+- 35.Qh6+- 26.Rh3 This is better now than 26.Qh2 h6 26...h6 27.Rah1 Bg7 27...axb4 28.axb4 Bg7 29.Rxh6!+- 27...Kh7? defending h6 is bad on account of 28.Qh2 f5 29.Rxh6+! Kg8 29...Bxh6? 30.Qxh6+ Kg8 31.Qh8+ Kf7 32.Rh7# 29...Kg7 30.Rh8+- 30.Rh8+ Kf7 31.Qh7+ Ke8 31...Bg7 32.Qxg7+! Kxg7 33.R1h7# 32.Rxf8+! Kxf8 33.Qh8+ Ke7 34.Rh7# 28.Rxh6! A beautiful exchange sacrifice in the style of Paul Morphy. Rxc3 This is pure desperation. 28...Bxh6 doesn't help either: 29.Rxh6 threatening Qh2 Kg7 29...Qf8 30.f5!+- 30.f5! Qg8 30...exf5 31.e6+ 31.fxg6 Kf8 32.Qg5 Rb7 33.Qf6+- 29.Qh2! A fine move sacrificing a bishop. This is even better than taking the black rook. Rxd3+ 30.Ke2 f5 30...Rxd4? 31.Rh8+ Bxh8 32.Qxh8# 30...Bxh6? 31.Qxh6 and Black will be mated 30...f6 31.Rxg6!+- or 31.Rh8+ Kf7! 31...Bxh8? 32.Qxh8+ Kf7 33.Rh7# 32.Rxd8 Rxd8 33.Kxd3+- 31...Qf8 32.Rxf6!+- 31.Rxg6! Qf8 31...Rxd4? 32.Qh8+! Kf7 33.Rxg7# 31...Kf8? 32.Rxg7! Kxg7 33.Qh7+ Kf8 34.Bc5+ 32.Rf6! Re3+ A check before resigning, but there were no more saving moves: 32...Bxf6? 33.Qh7# 32...Qf7 33.Qh7+ Kf8 34.Bc5+ Ke8 35.Qh8+! Bf8 35...Bxh8 36.Rxh8+ Qf8 37.Rhxf8# 36.Qxf8+! Qxf8 37.Rxf8# 32...Qxf6 33.Qh7+ Kf7 34.exf6 Ke8 34...Rg8 35.Qh5+ Kf8 36.Bc5# 35.Qxg7 Kd8 36.Rc1! this cuts off the black king's retreat, so that checkmate is unavoidable 32...Bh6 33.Rxf8+ Rxf8 or 33...Kxf8 34.Qxh6+ Ke8 35.Qh8+ Kf7 36.Rh7+ Kg6 37.Qg7# 34.Qxh6 Kf7 35.Bc5! now good advice is expensive, because apart from pointless rook sacrifices there remains only Rg8 35...Re8 36.Qh7# 36.Qf6+ Ke8 37.Qe7# 33.Kxe3 Qxf6 and Black resigned at the same time. This game proves that Komodo MCTS is able to play in the style of the famous romantic chess players like Morphy or Anderssen. The sacrifices made had a good positional basis. 1–0
  • Start an analysis engine:
  • Try maximizing the board:
  • Use the four cursor keys to replay the game. Make moves to analyse yourself.
  • Press Ctrl-B to rotate the board.
  • Drag the split bars between window panes.
  • Download&Clip PGN/GIF/FEN/QR Codes. Share the game.
  • Games viewed here will automatically be stored in your cloud clipboard (if you are logged in). Use the cloud clipboard also in ChessBase.
  • Create an account to access the games cloud.
WhiteEloWBlackEloBResYearECOEventRnd
Komodo MCTS 2217.003479Nirvana 2.431601–02018B49TCEC Season 14 - Division 21.1

In this game, the open-minded approach of the program is particularly impressive as a result of the Monte Carlo tree search used by Komodo MCTS.


The Sicilian has been known for decades as the most reliable way for Black to obtain an unbalanced but good position. Among the most popular Sicilians at the top level the two that certainly stand out are the Najdorf and the Paulsen.


Monte Carlo tree search (MCTS) — what is it?

In contrast to the conventional alpha-beta pruning search algorithm, no chess knowledge is required in this procedure, apart from the rules of the game and the possible outcomes of a chess game: win, defeat or draw. Roughly speaking, this is what happens: In a given position, a bunch of possible moves are tried out by the program playing against itself until a result is obtained: 1, 0 or 1/2. What the move that achieves the best result looks like does not matter. Whether the move abandons a pawn, a piece or the queen, allows a double or triple pawn, or violates any iron principle formulated by great chess teachers, is irrelevant as long as it is successful. You could call this an open-minded approach.

By playing out a large number of games to the end, conclusions are drawn about the best moves in a particular position. The program deals with the most successful moves in more detail than with the unsuccessful ones. The result is a "search tree", which at the end allows the choice of a move which is purely based on the best performance.

This method is particularly effective combined with "neural networks". It is well known that Alpha Zero played countless games against itself for nine hours and learned from the results. However, enormous computing capacities were made available for this, because otherwise, nine hours would hardly have been enough.  Afterwards Alpha Zero was able to beat Stockfish 8, one of the best chess programs in the world. In a nutshell, this means: AI (artificial intelligence) learns by trial and error purely from its experience.

Effective = aesthetically attractive?

In this context, I would like to mention the discussion between world champion Dr Emanuel Lasker and his archrival Dr Siegbert Tarrasch about the "aesthetic" value of a move.

Just before his World Chess Championship match against Tarrasch in 1908 (pictured via Wikimedia Commons), Lasker made these remarks:

Lasker and Tarrasch

"Dr Tarrasch is a thinker who loves deep and complex considerations. He is willing to acknowledge the effectiveness and usefulness of a move if he considers it both beautiful and theoretically correct. But I accept this kind of beauty only if and when it is appropriate. He admires an idea for its depth, I admire it for its effectiveness. My opponent believes in beauty, I believe in strength. I think that a move is also beautiful in that it is powerful." (a)

From this, it becomes clear that world champion Dr Emanuel Lasker's considerations were guided by the attempt to achieve the best results. This is exactly the approach Komodo MCTS also pursues.

Will alpha-beta pruning algorithms soon be exhausted?

In an interview published in December 2018, Komodo developer IGM Larry Kaufman stated that Komodo MCTS is making ten times faster progress than the "normal" Komodo version, which is known to rely on the proven alpha-beta pruning. (b) In a very simplified way, the following can be said about this:

AlphaThe alpha-beta pruning is about excluding as many moves and variations as possible because their results are obviously unfavourable. The alpha value indicates how much player "A" will reach at least if a certain move is played. The beta value, on the other hand, describes the maximum amount that player "B" will achieve by his best countermove. So if I check a move that wins a knight and find out that the opponent gets at most one pawn by playing the best continuation, I could assign a value of +2.00 to this move, if all other conditions are the same. Now I look at another move in the same starting position and notice that although I can capture an adverse rook, my opponent will get at least my queen and three pawns by proper counterplay so that the final result is -7.00.

BetaIn this way, all possible moves are assigned certain values. Those moves with the best values come right to the top and are immediately examined closer and deeper, while those with the worst values come last and usually don't give better results even at greater search depths. If nevertheless, this case should occur, then immediately a re-evaluation takes place, and a move evaluated so far with -2.55, for example, suddenly moves with a plus value to the first, second or third place.

In contrast to the Monte Carlo method, however, the positions are not played out (unless a mate is found), but are followed up to a search depth that increases over time and the resulting positions are evaluated according precisely to previously defined criteria. Here, not only the values of the pieces and pawns are important, but also positional factors such as king safety, pawn weaknesses, open lines, space advantage, control of the centre, etc.

This ingenious method has been refined more and more in recent years and has proven to be extremely successful. Nevertheless, it is possible that the scope for increasing playing strength may be exhausted at some point. Think of the latest Stockfish version 10, in which the programmers and their countless volunteer testers invested about ten months of development work, to gain just 15 ELO points compared to its predecessor version Stockfish 9 (64-bit, 1CPU, according to CCRL Elo list 40/40, dated January, 12th, 2019). 

Let's compare the three previous versions: Stockfish 7 was released on January, 2nd, 2016 and has an Elo rating of 3246. Stockfish 8 from November, 2nd, 2016, just 10 months later, jumped to 3301, a plus of 55 Elo points. This is equivalent to an increase of 5.5 points per month of development. Stockfish 9 from January, 31st, 2018, came out about 15 months later and reached a rating of 3367, 66 points more than the previous version, which means 4.4 Elo increase per month of development. Then ten months went by before the latest version Stockfish 10 was released on November, 30th, 2018. This new version has an Elo rating of 3382, only 15 points more than Stockfish 9. This shows that the curve suddenly flattens out with only a 1.5 point increase per month.

These observations might indicate that the development potential of the chess programs with the conventional alpha-beta pruning search algorithm has, in fact, already been largely exhausted. But of course one should not judge hastily. So let's wait for the next generation of popular programs. Only then we'll know whether this trend is consolidating or not. Anyway, from the above it becomes clear why the Komodo developers Larry Kaufman and Mark Lefler believe that their MCTS version will make the greatest progress. In the current Top Chess Engines Championship (TCEC 14) Komodo MCTS could already sensationally work its way into the "Premier League", and thus into the class of the best eight chess programs.

Komodo's positional masterpiece

It is remarkable that the new program with the Monte Carlo tree search was already able to win some games against a giant like Stockfish 10. The following game was played on December, 21st, 2018 with a time control of 40 minutes per 40 moves each. Both programs could use four processors. By a surprising pawn sacrifice, Komodo MCTS succeeded in positionally outmanoeuvring its formidable opponent. It's kind of weird that after White's 26th move, all three knights are placed on the edges of the board:

 
New ...
Open...
Share...
Layout...
Flip Board
Settings
MoveNResultEloPlayers
Replay and check the LiveBook here
The following game of the "giants" Komodo 12.3 MCTS and Stockfish 10 shows computer chess at the highest level: 1.e4 c5 2.Nf3 Nc6 3.d4 cxd4 4.Nxd4 e6 5.Nc3 Qc7 5...a6 6.Be2 Qc7 is also played and often means only a transposition of moves. 6.Be3 6.Ndb5 loses time after Qb8 , because the knight will soon be driven away by a7-a6. 6...a6 7.Qd2 Nf6 8.0-0-0 Bb4 Pinning Nc3 and threatening 9. ..Nxe4 8...Be7 9.f3 0-0 10.g4 b5 11.g5 Ne8 12.h4 Ne5 13.Kb1 Bb7 14.h5 Rc8 15.Qg2∞ Karjakin - Morozevich (Wijk aan Zee 2009). You'll find this interesting game in the Mega Database annotaded by Sergey Karjakin himself. 9.f3 Ne5 10.g4 0-0 10...h6!? 10...Bxc3!? 11.g5 Nh5 12.Kb1 Rb8 13.Qe1 Bxc3 Black waited with this exchange until White had moved his queen in order to gain a tempo. 14.Qxc3 14.bxc3? b5 is better for Black. 14...Qxc3 15.bxc3 b5 16.Bc1 Re8 17.Nb3 g6 17...Nxf3? 18.Be2!+- 18.Be2 Nc4 19.Rhe1 Rb6 20.Ka1 Rc6 21.h4 f6 22.Bf1! 22.gxf6 Nxf6 this only places the black knight on a better square. By playing 22.Bf1! Komodo is planning a positional pawn sacrifice, as we shall see soon. 22...Ne5 Now two white pawns are attacked at the same time: c3 and f3. 23.gxf6 23.Nd4!? Rxc3 24.gxf6 Nxf3 24...Nxf6? 25.Bb2 Rc4 26.Nxb5! Nxf3 27.Nd6 Nxe1 28.Bxc4 Nxc2+ 29.Kb1 Ne3 30.Rc1 Nxc4 31.Rxc4 Rf8 32.Bxf6 Rxf6 33.Rxc8++- 24...Rc7!? 25.Ba3!= Nxf6? 26.Bd6+- 25.Kb2! Nxe1 26.Kxc3 e5 27.Nxb5! axb5 28.Rxe1 Nxf6 29.Bxb5 Bb7 30.Kd3 Ra8 31.a4 23...Nxf3! 23...Nxf6?! 24.Na5 Rc7 25.Bf4 d6 26.Rxd6 Nxf3 27.Re3 Nxh4 28.Bg5! Nd5! 29.Rg3 this is better than 29.exd5 Nf5 29...Nxc3 30.Bd3! Nh4 has no safe retreat now, and one of the two black knights is lost. 24.f7+! A nice intermediate move, forcing the black king to take f7 which is an unfavourable square for him. The f6-pawn would have been lost anyway. 24.Re3 Ne5 25.Na5 Rb6 26.Ba3 Nxf6 27.Bc5 Rb8 28.Bd6 Neg4!∞ 24...Kxf7 25.Re3 Nh2 25...Ne5 26.Na5 Rc7 27.Be2 Nf6 28.Ba3 Neg4 29.Bxg4 Nxg4 30.Rf3+ Kg7 30...Kg8 31.Bd6 Ra7 32.Rdf1 Ra8 31.Bd6 Ra7 32.Bc5 Ra8 33.Nb7! Bxb7 34.Rxd7+ Kh6 35.Rff7 Rh8 36.Rxb7 looks unpleasent for Black. 26.Na5 All three knights are now on the edges of the board, and that under the responsibility of two Elo giants! Rc7 27.Bh3! Nh2 has no safe retreat now. White is threatening 28.Rh1 or Rd2. Nf4 27...Nf6? 28.Re2 Nhg4 28...Nf3? 29.Rf1 Rxc3 30.Bb2 Nd4 31.Bxc3 Nxe2 32.Rxf6++- this variation is another proof that the black king is poorly placed on f7 29.e5! Rxc3 30.Rd3 Rxd3 31.cxd3 Nxe5 32.Rxe5 Black has three pawns for his bishop but the white pieces are dominating. 28.Ba3! A strong move threatening 29.Bd6. It can easily be seen that there are three weak black pieces on the diagonal b8-h2. 28.Rh1 Nxh3 29.Rxh3 Ng4 30.h5 gxh5 31.Rxh5 h6∞ 28...e5 29.Bd6 Ra7 30.c4 Ra8 The white knight is well placed on a5 preventing the development of the c8-bishop even if the d7-pawn were not attacked by Bh3: 30...Bb7? 31.Bc5! attacking the bishop's defender what about luring the knight to c4? 30...bxc4? 31.Nxc4 unfortunately e5 is attacked twice now Kf6 or 31...Nxh3 32.Rxh3 Ng4 33.Bc5!+- 32.Rd2 Nxh3 avoiding the loss of a knight 33.Rxh2! Nf4 34.Rf3 Bb7 at last the black queen's bishop has been developed (move 34!) attacking e4, but now White wins the exchange thus: 35.Bxe5+! Rxe5 36.Rxf4+ Ke6 37.Nxe5 Kxe5 38.Rf7 Bxe4 39.Rxh7 31.Rde1 Black cannot really enjoy his extra pawn. Neither the rook on a8, nor the bishop on c8 can intervene in the game. Now Stockfish exchanges the only well-placed black piece in order to bring the knight stranded on h2 to safety. Nxh3 Once more there are three knights on the edges of the board. 32.Rxh3 Ng4 33.Rf3+ Kg8 34.Kb2 Ra7 34...Re6 35.c5! Re8 35...Ra7? 36.Rf8+ 36.h5! gxh5 37.Rh1 Kg7 38.Rxh5 Kg6 39.Rff5 h6! This move holds the black position together, just barely, but in any case White now regains his sacrificed pawn with a better position: 40.Rh4 Nf6 41.Rxe5 , etc. 35.Ref1 Ra8 36.a4! Another surpising pawn move by Komodo. bxa4 36...bxc4 37.Kc3! 37.c5 A positional masterpiece by Komodo MTCS. Black is completely outsmarted. The white king can be distracted temporarily, but that will only cause a slight delay: a3+ 38.Kc3 a2 39.Kb2 a1Q+ 40.Kxa1 h5 41.Ka2 Kh8 42.Rf7 Kg8 42...Ne3? 43.Bxe5+! Kg8 or 43...Rxe5 44.Rf8+ Kg7 45.R1f7+ Kh6 46.Rh8# 44.Rg7+ Kh8 45.Re7+ Kg8 46.Rxe8+ Kh7 47.Rh8# 43.Kb2 Ne3 44.Rf8+! Rxf8 45.Rxf8+ Kg7 46.Re8 Ng4 47.Re7+ Kf6 48.Nc4 Bb7 48...g5 is an attempt to get a passed pawn on the h-file: 49.hxg5+ Kxg5 50.c6! dxc6 or 50...h4 51.Nb6+- 51.Nb6! h4 52.Nxa8 h3 53.Bxe5 Nxe5 54.Rxe5+ Kf4 55.Re8 h2 56.Rh8 Kg3 57.Rxh2! Kxh2 58.e5 Kg3 59.Nc7 Kf4 60.e6+- 49.Nb6! Bxe4 49...Rd8 50.Nxd7+ Rxd7 51.Rxd7 Bxe4 52.Bc7 Ne3 53.Bd8+ Kf5 54.Ra7+- Nxc2? 55.c6! 49...g5 50.hxg5+ Kxg5 51.Rxd7 Bxe4 52.Be7+ Kf5 53.Nxa8 Bxa8 54.Ra7+- 50.Nxa8 Bxa8 51.Rxd7 a5 After 51...Bc6 52.Be7+ Ke6 53.Rd6+! Kxe7 54.Rxc6 Kf7 55.Rxa6+- White wins easily with his rook against the knight. 52.Be7+ Kf5 52...Ke6 53.Rd8! Kxe7 54.Rxa8 Nh6 55.Ra7+ Kd8 56.Rxa5 Nf5 57.Ra4+- 53.Rd6! 53.Ra7 is good enogh, too. 53...Ne3 54.Rf6+ Ke4 54...Kg4 55.Rxg6+ and Be7 is defending h4. 55.Rxg6 Kf5 56.Rh6 Bf3 57.c6! Nd5 58.Bd8 Ne3 59.Bxa5 Kg4 60.Bd8 Kf5 61.Kc1 61.c7!? Bb7 62.Kc3!+- 62.Rb6? Nc4+ 61...Bg4 62.c7 Ke4 63.Rxh5! Kf3 63...Bxh5? 64.c8Q , but there are no more saving moves. 64.Rxe5 and Black resigned. What an impressive game! 1–0
  • Start an analysis engine:
  • Try maximizing the board:
  • Use the four cursor keys to replay the game. Make moves to analyse yourself.
  • Press Ctrl-B to rotate the board.
  • Drag the split bars between window panes.
  • Download&Clip PGN/GIF/FEN/QR Codes. Share the game.
  • Games viewed here will automatically be stored in your cloud clipboard (if you are logged in). Use the cloud clipboard also in ChessBase.
  • Create an account to access the games cloud.
WhiteEloWBlackEloBResYearECOEventRnd
Komodo 12.3 MCTS 64-bit 4CPU3295Stockfish 10 64-bit 4CPU34641–02018B48CCRL 40/40134.3

In computer chess there is no getting past Komodo, a two-time ICGA Computer World Chess Champion. Find out how Komodo can take your game to the next level!


Komodo MCTS is available together with the conventional Komodo version in the ChessBase shop. You can advantageously use the Monte Carlo tree search without abandoning the time-proven alpha-beta pruning. By using the two search algorithms alternately, you will get deeper insights into complicated positions which is useful both for the analysis of games and for the computer-aided preparation of opening variations.

References

(a) Quoted according to Harold C. Schonberg, "Die Grossmeister des Schach", Frankfurt 1976, p. 119


Stephan is a passionate collector of chess books and for years he has been successfully playing as an amateur for his German club. The former musician and comedian works as a freelance journalist and author in Berlin and in the Franconian village Hiltpoltstein.

Discuss

Rules for reader comments

 
 

Not registered yet? Register

We use cookies and comparable technologies to provide certain functions, to improve the user experience and to offer interest-oriented content. Depending on their intended use, analysis cookies and marketing cookies may be used in addition to technically required cookies. Here you can make detailed settings or revoke your consent (if necessary partially) with effect for the future. Further information can be found in our data protection declaration.