This season Leela has shown that it can compete with the best engines, better than I anticipated after the previous season. Stockfish was clearly better in the premier division, but Leela gets better from event to event. This superfinal should be exciting, the engines have different playing styles and different approaches to evaluation. They may make moves that surprise their opponent as well as the spectators. I hope we see exciting draws, wins for both engines and a close result.
After 10 games Stockfish leads 2-0 with 8 draws.
Games 1-2 started with white a pawn up and the center pawns cleared. Game 1 reached a R vs BB imbalance on move 26 after a series of exchanges. Leela was two pawns up with a queen side passer, its eval was over 1 for a long while. Stockfish's eval dropped to 0, the engines mostly shuffled in a RN vs BBN position. Eventually after more exchanges Leela's eval dropped to 0 as well, the game ended in a tablebase draw on move 131. In the reverse Stockfish had an eval over 1 for 40 moves, the game reached an BN vs R imbalance with more pieces on the board, but reduced to a BN vs R position later. Leela had a queen side passer and evals dropped to 0 as it pushed it forward. After a long shuffle Stockfish countered with a passer on the king side, the game reached a tablebase draw on move 116.
In game 3 queens were off early, the engines gradually exchanged pieces until reaching a rook ending on move 36. Leela had a pair of connected passers on the queen side and it remained hopeful for a while. Stockfish countered with its pawns on the king side, and this was enough to hold a draw. In game 4 Stockfish had a small eval advantage, there were several early exchanges but the queens stayed on the board until move 30. The engines targeted the pawns, and by move 37 only two pawns were left for each engine. Evals dropped to 0, Stockfish could close the game with a perpetual check but changed its mind at the last move. In the end the draw rule adjudicated the game.
Leela's eval jumped over 1 after a few moves in game 5. It exposed the black king and pushed pawns on the king side. Stockfish exchanged pieces to stop the attack, then gave a knight for two pawns and reduced to a BN v B ending. Leela needed 30 more moves to lower its eval, the game ended in a tablebase draw. Stockfish started game 6 with an eval advantage,but its eval dropped to 0 by move 22. Leela's eval became negative, it threatened the white king on the queen side and Stockfish's pieces had less space to move. Stockfish reduced to a rook ending, and 40 moves later the game ended in a tablebase draw.
Games 7-8 were a King's Gambit variant, both started with black a pawn up and a significant eval advantage. In game 7 Leela castled short and Stockfish pushed its king side pawns forward. Leela pushed pawns on the queen side and captured a free pawn, but it delayed the development of the queen side pieces. Stockfish thought Leela's 15th move was a blunder, its eval jumped over 1.5 as it used its king side pawns to expose the white king.
Leela gave a rook for a bishop and Stockfish's eval was already over 4. Stockfish gave a knight and pawn, leaving the white king almost defenseless. The black major pieces joined the attack while most of the white pieces were still stuck on the wrong side of the board.
The white king was under mate threats and fighting for its life. Leela barely managed to defend, eventually exchanging pieces to save its king. The game reached a R vs RB position, black with two passers on the king side. The attack on the king was over but evals were close to 10.
There was no doubt Stockfish was winning. Due to a configuration bug the game continued beyond the normal win rule, the rooks were exchanged and white couldn't stop all the marching pawns. First decisive game of the match, with some help from the biased opening.
In the reverse game Stockfish castled long and kept its king safe. Leela kept the pawn advantage and had an eval of over 1 that constantly increased. The engines exchanges all major pieces by move 25 and Leela's eval was around 4. Stockfish's eval slowly increased, it seemed that Leela was slowly improving. The game reached a BBN vs BBN position with stable pawns, Leela just needed to find a way to capture a pawn and the white defense would collapse. Then suddenly Stockfish's eval dropped, it exchanged a bishop for a knight to get this position:
Leela's LS bishop had no targets and its DS bishop was blocked, Stockfish could keep the black king away, a perfect fortress. Leela conceded the draw 70 moves later. A lucky escape for Stockfish, disappointment for Leela.
Leela started game 9 with a small eval advantage. Stockfish's eval was close to 0 after a few moves and stayed there throughout the game. The engines exchanged pieces and pawns gradually, by move 38 only QRB vs QRB remained. Leela's eval came down, but when it captured a pawn and reduced to a bishop ending its eval jumped over 2. Stockfish had an advanced passer, Leela gave its bishop to take it. That left two white passers, Stockfish's bishop and king were on time to stop them, the game ended in a draw.
The start of game 10 was similar, Stockfish started with an eval advantage and the engines opened the center. Stockfish's eval jumped over 1 as it captured a pawn on the king side and exposed the black king. Leela exposed the white king in return, and then Stockfish allowed the black king to capture its bishop.
Evals started to jump as Stockfish attacked the king side with all available pieces. The black king was in the open and there were not enough black defenders. In a few moves the black king was trapped and several black pieces were pinned down, evals were over 4.
Stockfish got the piece back for a pawn. Leela managed to get its king to safety on the queen side but it lost another piece in the process. The game reduced to a QRB vs QR position, black two pawns up.
Leela tried to counterattack the white king, this delayed the end a little. When the checks stopped the black king was in trouble again, the game was adjudicated before Leela lost more material.
No comments:
Post a Comment