Sunday, February 18, 2018

Season 11 premier division, after RR1

Crosstable after RR1

The draw rate is only 57.1% so far. This is much lower than expected for bookless games between strong engines. The difference in strength between engines at the top and those at the bottom is proving to be high.

The top 3 engines Stockfish, Houdini and Komodo are already leading the table. Komodo beat Stockfish in the last game of the RR, quite an upset - again, these are bookless games.
Fizbo is 8th so far, losing 4 games. Ginkgo and Andscacs lost 3 games each, Andscacs had an advantage because it beat both Fizbo and Ginkgo.

In some of the games the engines lower their evals when they realize the game is heading for a draw. As a result the draw rule is applied earlier making the games shorter. This effect appeared in previous seasons in the late stages when strong engines played each other. 

A few interesting games

Fizbo - Andscacs: Andscacs equalized and had a small eval advantage starting from move 20. Both engines had active pieces, Fizbo had a pawn majority on the queen side and a strong knight outpost, while Andscacs pushed pawns on the king side. On a probable Fizbo blunder Andscacs' eval jumped over 2, suddenly Andscacs had a dangerous attack on the white king. Fizbo gave a rook to stop the immediate threat, it still had a few assets but eventually Andscacs traded down to a Q vs QN position and a win. Andscacs is first to draw blood in the premier division, a win in black in a bookless game.

Fire - Fizbo: Evals were low with a small advantage for Fire, they remained stable for many moves. The game reached a same color bishops ending on move 62, Fizbo was a pawn up but had a vulnerable backward pawn on the flank that had to be watched by the king. Fire used this (together with another Fizbo blunder?)  to create a passer on the queen side, trading for the black bishop and winning the game with its final pawn.

Komodo - Ginkgo: Ginkgo sacrificed a pawn early, Komodo had a small eval advantage in a RRN vs RRB position on move 20. The evals slowly increased as Komodo outplayed Ginkgo in the endgame. The knight was stronger than the bishop, and the white king was more active, allowing Komodo to capture a second pawn and eventually win the game.

Houdini - Chiron: Houdini had an eval advantage close to 1, when the position opened up Chiron traded RR for Q which later became a Q vs RN imbalance. Evals peaked over 1.5, Houdini couldn't break Chiron's defense and the engines agreed on a draw after a long game.

Chiron - Fire: a miniature draw after 18 moves. The exact move sequence appeared in Fire's PV of move 6.

Komodo - Andscacs: Andscacs let its bishop be trapped on move 19 and was forced to trade it for two pawns. Komodo's eval jumped over 2 and continued to increase. The game was adjudicated when Andscacs lost another piece. A surprising blunder at this level of play.

Stockfish - Fizbo: Stockfish started with an eval advantage that kept growing. By move 29 it was a pawn up with a pair of connected passers in the center. Eventually Fizbo had to give a rook to get rid of the passers, an easy win for Stockfish.

Houdini - Ginkgo: The game was balanced until Ginkgo sacrificed a pawn on move 20, in an attempt to weaken the white center and open a file for its rook. Houdini's eval immediately started to increase, slowly at first. Houdini had a pawn majority on the king side. After a few exchanges Houdini attacked through the queen side with a rook that Ginkgo wouldn't take for fear of creating a passer on the 6th rank. Houdini's eval was winning and the game was over after several more moves.

Andscacs - Stockfish: Stockfish advanced pawns on the queen side, Andscacs ignored the danger until it was too late. Stockfish gave a pawn to create a queen side passer on the 2nd rank, and Andscacs eventually lost a piece after getting rid of it. The game ended in a R vs RB position, Stockfish winning in black.

Stockfish - Fire: All knights were off early, Stockfish with a small eval advantage. When Stockfish started to push pawns on the king side with many backing pieces the evals jumped over 1.5. Fire traded down to a RRB vs RRB position, the black pawns were weak while white had supported passers. Stockfish started to win material, with quick adjudication. An important win for Stockfish, Fire's first loss.

Houdini - Andscacs: Houdini had an eval advantage that went over 1 when Andscacs traded a rook for a bishop and a pawn. Andscacs thought it was relatively safe as it traded down to a R vs B ending, but this was only because it did not have tablebase support. Houdini knew it was a 6-man win.

Fizbo - Chiron: Fizbo chose to trade R for BN, after exchanging queens Chiron was two pawns up with eval over 2. Chiron's pawns started to advance and the black king came forward, an easy win.

Andscacs - Ginkgo: Andscacs played the opening better, its eval constantly increasing. Andscacs was a pawn up on move 28, and after several exchanges the game reached a RR vs RN position. Andscacs slowly improved its position in the endgame, it took many moves but the win was never in doubt. 

Featured game: Komodo - Stockfish
Premier division, round 7
Link to game on TCEC

The engines played an opening with an early R vs BN imbalance. Evals were low, Komodo had a small eval advantage with a potential king side passer. Stockfish had a bishop pair but the LSB was limited by the black pawns.

The black king was exposed and had to be defended by pieces. Komodo used the open file on the king side to bring the rooks forward. Stockfish managed to free its LSB but at the cost of a ruined pawn structrure. Stockfish's eval indicated it was in trouble.

Komodo created a connected pair of passers in the center and attacked with its rooks. After Komodo won a piece Stockfish made a weak attempt to counter with an attack on the white king.

Komodo defended easily, it traded its passers for the black passer and targeted the remaining black pawns to get the win. It was a surprise for me to see Komodo beating Stockfish in a bookless game, the Stockfish fans must be in shock.

No comments:

Post a Comment