Friday, November 27, 2015

Season 8 superfinal, games 61-70

After 70 games in the Komodo-Stockfish superfinal match, Komodo is leading 4-1 with 65 draws. The last 10 games have all been draws, again. An important change is that from game 67 the openings are supposed to be more unbalanced, as preselected by Cato to try to lower the draw rate. The evals out of the openings are higher than before, but the first 4 games 67-70 were still draws. Still 30 games to go so we will have to wait and see.

Game 61 was very short. Stockfish went for the b pawn with the queen, and Komodo almost trapped it, leading to a draw by repetition. Game 62 started with many captures, reaching an RN vs RN ending in move 28 with Stockfish a pawn up. Both engines continued with 0 evals and eventually the game was adjudicated even though both had dangerous passed pawns. There was a possible repetition in the end but it wasn't played out... a draw in any case.

In game 63 Komodo started with a small eval advantage, but it disappeared when almost all the queen side pawns were captured and the pieces left were QRN vs QRB. Then Komodo's eval went up again as it got into the black back rank, but once the queens were exchanged Stockfish secured a king side fortress. It took many shuffling moves for Komodo to realize there was nothing to do, finally it broke the pawn structure to get out of the 50 move draw. This only led to a repetition draw later. In game 64 Stockfish held on to an eval advantage with an advancing passed pawn. All pieces focused on attacking and defending this pawn until the position was stuck, the engines chose to end with a repetition of checks.

Game 65 reached a RB vs RB ending by move 22, and then continued for more than 100 shuffling moves with a few pawn moves to reset the 50 move rule. Komodo insisted it had a small advantage, and the game could not be adjudicated until it realized what everyone knew for so long - it was a draw... Game 66 was played in a different line but just as drawish. This time the game reached a B vs N at move 30 and both engines had a 0 eval, allowing the game to stop without a long wait.

Game 67 was the first of the 'Cato openings' of the superfinal. There was a lot of hope that the games would be more decisive the way these openings were chosen. The first 4 were draws but there were higher evals in these games, so there is still hope.

Komodo started with an eval advantage that got up to 0.63 in game 67. Stockfish defended well and after 26 moves the pieces left were QRB vs QRB, Komodo a pawn up, passed and isolated. After the queens were exchanged and the pawns were taken one by one the evals dropped to zero. The game reached a 6-man drawn position. In game 68 Stockfish held its advantage longer and gave an eval of 0.9 for the isolated passed pawn it had, also a pawn up. Komodo defended well and Stockfish couldn't convert a win, gradually the pieces were captured until there was a rook ending. Stockfish was still a pawn up but not enough for a win.

In game 69 Komodo again had an eval advantage that got up to 0.7. Material remained equal but Komodo created a passed pawn after the queens were exchanged. Pieces kept being captured until there were only B vs N. It then took a while for both engines to lower their evals to agree on a draw. Stockfish didn't get as much out of the opening in game 70. As the game progressed Komodo was attacking on the king side and Stockfish countering on the queen side. Both engines started to see perpetual check in the future but Stockfish went for a bishop and pawns for rook exchange, forcing the draw when Komodo was left with KR with no pawns.

Wednesday, November 25, 2015

Season 8 superfinal, games 51-60

After 60 games in the Komodo - Stockfish superfianl match, Komodo leads by 4-1 with 55 draws. The last 10 games have been draws, mostly uneventful.

In game 51 the Komodo's eval jumped a little when Stockfish chose not to exchange pieces in the center. It took only a few moves for the eval to get back down, not clear what Komodo saw there. In the reverse game Stockfish had increasing evals up to 0.7,  chance to get a game back. After move 30 only the queens and rooks were left with pawns, both engines had an advanced passed pawn. However, when queens were exchanged and the passed pawns eliminated the evals plummeted to 0 in a drawn rook endgame.

Komodo held a small eval advantage for about 25 moves in game 53. The advantage disappeared as the pieces that were left started to be exchanged, until a drawn queen ending was reached. In game 54 Stockfish held the eval advantage longer, each engine staying behind almost static lines. When files started to open to a QRB vs QRN game, Stockfish had a passed pawn but Komodo countered with control over the 2nd rank. The threats balanced out to another draw.

In game 55 Stockfish seemed very cramped on the king side with many pieces and no room to move. Nevertheless it was an effective fortress which Komodo could only penetrate by exchanging pieces. By the time the black king was exposed it turned out that the white king was vulnerable to Stockfish's perpetual check, and a draw. Game 56 started with a series of captures that resulted in a RRB vs RRB endgame on move 17. The game continued until move 40 but the evals were 0 long before that.

In game 57 the center and queen side were blocked by pawns, and for a while it looked as though Stockfish as black had some pressure on the white king. The pressure evaporated after a few captures, and the pieces left could not move the pawn blockade, draw. In game 58 Stockfish again got a little excited with evals up to 0.5. Komodo's defense showed that Stockfish had nothing,  pawns again blocking any progress in a RRN vs RRN endgame.

In game 59 the center and king side were stablilized early in the game, and the queen side was used to exchange pieces. The remaining pieces stated to shuffle until the draw was declared. Game 60 reached a BNN vs BNN endgame in move 24, and then became a long shuffle contest since Stockfish kept a small positve eval. The eval decreased to 0 when the 50 move rule got close and the draw followed.

 

Monday, November 23, 2015

Season 8 superfinal, games 41-50


After 50 games, at the halfway point of the Komodo-Stockfish superfinal match, Komodo is leading 4-1 with 45 draws. Komodo increased its lead with another win, the chances of a Stockfish comeback are getting really low.

Game 41 got into an interesting imbalance where Komodo was down B vs RP but Stockfish couldn't castle and had badly placed pieces. At one stage Stockfish turned down a repetition twice, but had to give the exchange back. It still had a passed pawn but with the king exposed to queen attacks the game still ended in a draw. There were a few moves in game 42 where Komodo had a small eval advantage as black, but it faded quickly back to 0 and another draw, nothing to get excited about.

In game 43 Komodo kept a small eval advantage but could not realize it on the board. Evals came back to zero as the pieces came off the board, yet another draw. Game 44 featured almost 90 moves of shuffling in a RRB vs RRN position blocked by pawns, close to eight hours of waiting for the final draw.

Game 45 started with a small eval advantage for Komodo, but the evals soon dropped to zero, and this time the game ended quickly in a repetition. In game 46 all the pawns stayed on the board with the queen side blocked. Komodo gradually increased its eval with a little more space on the king side.



After some piece shuffling Komodo was able to realize its advantage by taking a pawn after the queens were exchanged.

Stockfish tried to trade pieces, eventually reaching an opposite color bishop ending with a pawn down, usually a sure draw.

However, in this position Komodo could hold the white king to defend the h pawn, and its king could walk safely to the queen side and attack the pawns there. Stockfish sttacked the black pawns on the king side with its bishop, still only one pawn down, but Komodo's pawns were better placed in the end.



Another win for Komodo, again in black, with opposite color bishops. Inconceivable! You get the feeling that Komodo is just better than Stockfish in this match.

In game 47 Komodo built an eval advantage which turned into an ending of BB vs BN. Stockfish had a beautiful defense, look at the position:



It's a bit like checkers. The light square bishop cannot penetrate, only be exchanged, and the dark square bishop can't do anything to white pieces. After a lot of shuffling Komodo exchanged a bishop for the knight, and for a few moves its evals shot up. It was a false alarm since although Komodo got to queen a pawn first, the only thing the queen was good for was stopping Stockfish from queening as well.



In game 48 Stockfish had two bishops and Komodo two knights, but all the rooks stayed on the board. This made the game much more drawish and it ended faster.

Game 49 was a little wild. Stockfish went for a king side attack, sacrificing two pawns and neglecting to castle. Komodo calmly counterattacked on the queen side, exchanging pieces when possible, then went for a combination that left RBB and pawns vs QN. The kings were so exposed that no progress could be made, so again a draw but at least interesting. Game 50 was also imbalanced, this time Stockfish sacrificed a knight for 3 pawns. Komodo maintained an eval advantage as black and tried to make use of its extra piece. However, by the time it got a pawn back the game was in a 7-man tablebase draw of RN vs RPP, adjudicating a few moves later.

Friday, November 20, 2015

Season 8 superfinal, games 31-40


After 40 games in the Komodo-Stockfish superfinal Komodo leads by 3 wins to 1, with 36 draws. Stockfish has finally won a game, showing that it is possible to beat Komodo, just very very hard. On the other hand Komodo added another win too, this time playing black. My feeling - Komodo is going to win this match.

Game 31 started with both queens moving forward unsupported, followed by a queen exchange as well as many other pieces captured. The game reached a double rook endgame on move 25, with equal material and where Komodo had an isolated passed pawn. Stockfish had no problem defending until all pawns were gone but one that couldn't advance, another draw. Game 32 was only a little different, reaching a RN vs RN endgame on move 26. All pieces were eventually captured, the game ended in a drawn king and pawns position.

In game 33 Komodo had a small eval advantage that didn't get anywhere. Once all the heavy pieces were exchanged Komodo's knights were a little more effective than Stockfish's bishops, but it still ended a draw. In game 34 there were no castlings and the pawns blocked all files except one. Stockfish had more room and an eval advantage but there was no way to improve. Instead both engines shuffled a lot with one capture to avoid the 50 move draw once. The game was adjudicated before the 50 moves were reached the second time.

In game 35 Komodo's king became very exposed but Stockfish had a 0 eval. Komodo was able to stabilize its position, and after rooks were exchanged the white king was much safer and the game drifted into a draw. Game 36 took a different path, Stockfish held to a small eval advantage and the heavy pieces remained on the board. At move 37 the position was:



Here Komodo decided to sacrifice a pawn, perhaps to gain initiative or to get rid of the white bishop pair that was gaining strength. A few moves later all the light pieces were gone and Komodo had to defend carefully to keep the b pawn from falling as well.



The eval started to increase gradually, but with all the heavy pieces playing it seemed that Komodo had enough to defend as well as threat the white king. The engines maneuvered for a long time until the queens were exchanged.



Now Stockfish's plan was getting clear, exchange pawns and keep the rooks. Indeed this is what happened, with Stockfish holding on to the connected e+f pawns.



This was a decisive advantage that the engines could not ignore, and even the most skeptic of fans had to realize that Stockfish was really winning. Finally, a sigh of relief. Stockfish with a clear win, without a lot of help from the opening. A score of 2-1 is very close and there are still many games to be played. This was a happy Stockfish moment.

Game 37 was very quiet, nothing special from either side and evals close to zero throughout the game. A draw adjudication was reached at the first possible opportunity. Game 38 started in a similar way, but around move 20 Komodo took the initiative and the evals became slightly negative. At move 30 Komodo had more space but the position looked blocked and the PV showed only shuffling.


After 20 shuffling moves Komodo broke the pawn structure with f6 and the board cleared up a little. Stockfish was getting low on time, and after the queens were exchanged Komodo's eval increased a little, what was it planning?



The black b pawn and the white c pawn were eliminated with a bishop exchange as well, and Stockfish (and its fans) started to feel the pressure. Komodo's plan was still not clear and material was equal but something was happening.

 Komodo placed the knight on f5, effectively defending its pawns while attacking the white h pawn. Meanwhile the black rook could attack pawns, the bishop or the king. Eventually Stockfish lost the d pawn, creating dangerous black pawns in the center.


As they started marching together Stockfish had no choice but to sacrifice the bishop to stop them, and the evals jumped accordingly.

The 8-piece ending was still very difficult and the game was adjudicated long before the mate was in sight. Komodo had to hang on to its remaining pawn to avoid a theoretical draw, but both engines agreed that the end result would be a Komodo win. Back to 3-1, so close to Stockfish's first victory Komodo produces a win in black. And the reverse game was so boring, where did this come from?

Games 39 and 40 were two relatively short uneventful draws. In both games the two engines had their kings exposed to attack, and when a threat became serious enough the other engine closed the game by perpetual checks.

Approaching the halfway point, the match is still quite close. Stockfish has shown it can win, but can it close the gap to Komodo? My gut feeling: no it can't. We'll know in about two weeks at the most.

Wednesday, November 18, 2015

Season 8 superfinal, games 21-30

After 30 games in the Komodo-Stockfish superfinal match Komodo is leading by 2 wins with 28 draws. The last 10 games have all been draws, but one of them was the most dramatic game of the season that everybody is talking about  in the chat and Stockfish forums ever since.

Game 21 started with a small eval advantage for Komodo. After a series of 11 consecutive capures the remaining pieces were QRR vs QRR with many pawns. This is a sure sign of a draw. Once the queens were off the evals reached 0 and the pawns were gradually removed until a tablebase draw was reached.
The reverse game 22 was a totally different story. The pieces stayed on the board for a longer time in a complicated position. The evals started to jump after Komodo went for a skewer attack with a bishop.




Really difficult to understand what is going on with all the pins and threats, but white could be losing an exchange immediately. And yet Stockfish's eval went up as pieces started to fall. After a while things settled down a bit.


Stockfish was an exchange down, but it was quite optimistic according to the eval. Was it seeing something that Komodo missed? The queen side pawns had potential but was this enough? It should be noted that the website was down for over three hours, so nobody saw the game live at this stage. The white eval gradually increased and Komodo's eval started to agree with this as well. When the live broadcast returned the Stockfish fans started to get excited, it looked like the first Stockfish win is near. The evals started to jump when rooks were exchanged.


Stockfish was over the 6.5 eval threshold a few moves later, having established a pair of connected passed pawns.

Komodo was still behind in eval but was rising as well. It was just a matter of time... Stockfish just had to be careful of potential check threats but it had it covered in the PV at least. Then - the website went offline again. So close to the end, it was clear that we won't see it live but the win was sure. When the website returned though - A DRAW, WHAT??? Nobody could believe it. The evals were over 26 for Stockfish and almost 5 for Komodo, and then Stockfish threw it away in this position:

The move played was Kg4, allowing Komodo to escape with checks. The post game consensus was that Bc2, which appeared in the Stockfish PV a few moves back, was a sure win. Nobody could explain this choice by Stockfish and the shock was total. The Stockfish eval function became a joke, and the overall feeling was that if it can't convert an eval of 26 then how will it win games at all in this match?

In game 23 Komodo had an isolated passed pawn but it was not enough for a win in the RB vs RB ending that was reached. In game 24 again Komodo had an isolated passed pawn, this time as black, this time the ending was of opposite colored bishops. The draw took a surprisingly long time, Stockfish kept its eval at 0.06 not allowing the adjudication rule to be applied.

Game 25 reached an RB vs RB ending on move 20. I don't know why the evals stayed positive, perhaps because white won a pawn later, but before long the evals dropped to zero resulting in a draw. In game 26 Stockfish managed to create an eval advantage of 0.6, Komodo's pieces had very little room to move. However, there didn't seem to be a plan for Stockfish to improve and then pieces started to be taken off. The rook ending was again a dead draw.

In game 27 Komodo seemed to have a promising attack on the black king but Stockfish countered on the queen side, sacrificed a bishop for pawns and got a perpetual check draw. In game 28 the evals dropped to zero much faster and the game reached a repetition without a lot of action.

In game 29 the line the engines chose led to a material imbalance of BN vs RP. Komodo had a small eval advantage of 0.5 that made Stockfish fans nervous again.



Stockfish managed to capture a bishop, at the price of letting Komodo queen a pawn which cost a rook.



In the ending Komodo's extra bishop let it eliminate some of the black pawns, but when all the white pawns were off the board the game was drawn. In game 30 Stockfish had a small advantage but the game progressed without material imbalances. towards the end Komodo sacrificed two pawns to open the white king position, and the game was adjudicated just before the perpetual checks started.

Sunday, November 15, 2015

Season 8 superfinal, games 11-20


After 20 games in the Komodo-Stockfish superfinal match Komodo is ahead by 2 wins, with 18 draws. The games continue to be quite fast, about 5 hours per game, which is remarkable given that each engine has 3 hours for the game plus increments. The high draw rate was to be expected but it's still annoying. Stockfish hasn't been able to win a game yet, and the Stockfish fans (myself included) are getting nervous. Is Komodo too strong this season?

In game 11 Komodo held on to an eval advantage that reached almost 0.5 at its peak. The pieces left in the middlegame were RRBN vs RRBB with Komodo a pawn up. Stockfish defended well, at one point it initiated a combination that left it with a bishop for two pawns down, one of which was an advanced passed pawn.



In order to stop the pawn Stockfish lost a bishop, reaching a drawn endgame of opposite color bishops with Komodo a pawn up. In the reverse game the small advantage Stockfish had as white evaporated much faster. The game contained a series of 16 consecutive plys where 15/16 were captures, leading to a quick tablebase draw.

Game 13 started with Komodo maintaining an eval advantage of about 0.4. Starting at move 20 Stockfish started moving the bishop and knight forward and back again, allowing Komodo several tempi to arrange the pieces for an attack.



Komodo put pressure on the d file until it was ready for the c5 push. When it came there started a series of captures that resulted in a double rook endgame with Komodo ahead by a knight for three pawns.

Stockfish put up a fight, and it was a slow technical endgame. The black pawns were slowly taken as well as a pair of rooks. Komodo had to keep at least one pawn to avoid a RN vs R ending.



Stockfish also was getting low on time, down to the increments. The 1 second overhead worked well to keep from time failure, but Komodo converted the win on the board. 

In the reverse game 14 Komodo defended differently against Stockfish's advantage in the opening. The position simplified quickly as before but this time the heavy pieces were left as well as almost all pawns.Stockfish space advantage was meaningless in this position and the eval remained constant for a while and then began to decrease. Komodo held the draw, adding an official win from this opening.

There was one unexpected event in game 15, a negative eval !! This was very rare, a first for this superfinal if I am not mistaken. It was only for about 5 moves and only got as far as -0.09, but at least we know it is possible. In game 16 Stockfish's eval gradually increased to about 0.4, with a pawn advantage. Komodo blockaded well and then exchanged a rook for a bishop to take over the 2nd rank. The white and black threats canceled out, another draw.

The negative eval came back in game 17, this time up to -0.14. Other than that there's not a lot to say about this draw. Stockfish needs much better scores if it wants to win a game. In game 18 Stockfish built an eval advantage of 0.8 when Komodo sent a rook forward and had to play carefully to get it back. The rook was caught and exchanged for a bishop in the end, but by then the position was simplified and the evals closer to 0. Komodo sealed the draw with a rook sacrifice and perpetual check.

Game 19 featured many captures from the start, reaching a RB vs RB endgame by move 23 which continued to a tablebase draw. In game 20 the pieces lasted longer but the evals sank to 0 again and another draw.

As the match progresses I find it difficult to both write reports and watch the games. I need more time !! 

Wednesday, November 11, 2015

Season 8 superfinal, games 1-10


The Komodo-Stockfish season 8 superfinal has started !! After a week's holiday and a few delays it is finally here: 100 games at time control of 3 hours per engine, on a new fast 24-core machine with a huge RAM, and with access to 6-man tablebases. Both engines have new versions which should be even better than before. Hopefully Stockfish will not have time problems - as a precaution it is running with a full second overhead.

After 10 games Komodo is ahead by 1 win, with 9 draws.

The first 10 games were much quicker than anybody anticipated, I predicted many games would be more than 6 hours long but so far many draws were quite short. In the games so far it seems that both engines see a draw from a mile away and converge to it together with zero evals. The games stop around move 40 by adjudication, or earlier by repetition. 

The final started with a Komodo win in game 1 and a quick draw in game 2. More on this below. Also see this video analyzing game 1 by kingscrusher. However, the next 8 games were draws, mostly boring. This is going to be a long match and exciting games may be a rare bonus.

The opening position for games 1 and 2 had black with two pawns up and white with open files on the king side for immediate attack. Surprisingly for such an unbalanced position, both evals were 0. In fact Komodo was expecting a very short game ending in repetition and this was indeed the result of game 2. In game 1 Stockfish chose to play on. After move 20 Komodo was two pawns behind but it had a strong center and black's king was still uncastled and vulnerable. Komodo had a small eval advantage but Stockfish still felt safe.



The black king moved, giving up on the chance of castling. a few moves later the white center was even stronger, only one pawn down but with evals shooing upwards over 1 for both engines.

 Stockfish tried to defend, exchanging queens and giving a rook for a knight. The situation was becoming hopeless. With the king vulnerable to attack and white having pawns on the sixth row both engines saw the win coming.


It was an easy win for Komodo from here.

Games 3 and 4 were both quite boring, evals were never far from 0 and both engines played with a draw in mind.

Game 5 started with an advantage for white, but it quickly disappeared to zero again. Here Stockfish as black had an extra pawn, a passed pawn too, but did not have the strength to move it forward safely. The reverse game had even less action, but it was fast with a repetition end after 30 moves.

Game 7 reached a rook ending by move 25 and a drawn tablebase position on move 50. Game 8 was a different matter altogether. Komodo's queen took an unguarded g pawn just a few moves after the 8-move book and Stockfish's eval jumped. Was this a case of a poisoned pawn?




A few moves later black's queen escaped but Stockfish prepared an attack on the king side and had an eval of about 1. Would this be a win for Stockfish?


White sacrificed two rooks and a knight for black's queen and shattering the king's position, Komodo remained cool with a rather low eval.

The momentum of the attack was mostly over. Stockfish regained some material but it was left with only the queen and pawns against a RBN combination.

It took a while but eventually Stockfish lowered its eval and recognized it couldn't win. A good effort though.

In game 9 all the heavy pieces were off before move 20. The play continued in absolute equality ending in a tablebase position. There was a little more action in game 10, the pieces stayed on longer and the evals were non-zero for a bit, but the RB vs RB endgame was again a draw.

Thursday, November 5, 2015

Season 6 superfinal, games 1-8


I sometimes think about going over the TCEC archive to look back at the games, look for special gems, relive the excitement. Anyway, since we're waiting for the superfinal of season 8 and there is a break of over a week, I thought I'll give it a go. I chose to start with the season 6 superfinal between Stockfish and Komodo. This was the season where Stockfish was the most dominant, winning its first (and only so far) TCEC title.

The format was 32 openings, each played in reverse color pairs, time control was 2 hours per engine and the usual 30 second increment.

Game 1
Link to TCEC archive

The opening favored white,  black had open files to attack on the king side but the blocked center made it hard to maneuver. White had more space and could attack on the queen side since black chose to castle long.

Stockfish's eval advantage gradually increased after Komodo decided to give a rook for a bishop and started to play defensively.

The pressure on the queen side increased and after a series of pins and threats the knights were exchanged. Could Komodo hold the position?

At this point Komodo made what seems to me as the crucial mistake. It decided to go after the b5 pawn, ignoring the king side. Stockfish saw this and immediately sent the passed h pawn forward.

Komodo had to give up the pawns in the center to stop the h pawn, and after queens were exchanged Komodo couldn't stop the remaining white pawns without losing material.

In the reverse game Stockfish chose to exchange queens early and not to castle at all. Komodo managed to get two connected passed pawns on the king side but Stockfish built an effective blockade defense and the game ended in a draw. Stockfish in the lead after first opening !!

The next two games were quick draws, both ending in 3-fold repetition. Game 5 was also a draw, reaching a rook endgame by move 27 and a tablebase position on move 43. In the reverse game Komodo held on to an eval advantage for a bit longer but Stockfish defended well until almost all pieces were removed. In the ending of B vs N Komodo sacrificed the bishop for two pawns, then both sides queened, but it was still a draw.

In game 7 Stockfish gave the illusion it had an advantage, but after two long stretches of shuffling the position opened up and the advantage slowly grew smaller, until there were almost no pieces in a tablebase draw. 

Game 8
Link to TCEC archive

Komodo started with a small eval advantage in the reverse game, with Stockfish giving a rook for a knight early on. However once the queens were exchanged the evals became drawish.


As the game progressed Stockfish started to get better. The black knights were much more effective in a closed position than either rooks or the white bishop.



Komodo found it harder and harder to move while Stockfish arranged its pieces for an optimal pawn advance.



When the break finally came Stockfish had a huge eval advantage. Komodo gave its useless bishop for a pawn but could not untangle itself. Once a pair of rooks was exchanged the black knights became too strong and the game was over.

Stockfish up 2 (6 draws) after 8 games, still a long way to go.