Sunday, August 28, 2016

Season 9, stage 3, RR 6

Table after 6 RR

Stockfish has extended its lead to 2.5 points after 4 wins in this RR including a second win against Fire, one of the toughest engines in this stage. It is very hard to imagine that Stockfish will not be first at the end of this stage, though there are 14 games left for each engine.
The race for second place is still very close between Houdini and Komodo. This may be decided only at the last game, as we saw in this RR. The final games for both the engines are playing white against weaker opponents, and unless one of them has a spectacular result we may not know which qualifies until the end of RR 8.

Some of the game highlights

Komodo - Houdini, again a draw. It seems the race will be decided by results against other engines.

Jonny couldn't get anything going against Stockfish, and Stockfish started to attack the white king with a queen and knights combination. Evals went Stockfish's way for a while but Jonny defended well and was only a rook for a bishop behind, evals becoming static in a QRB vs QRR position. After a long shuffle with a rook exchange and a few pawn moves Stockfish found a crack in Jonny's defense. Once the queens were gone the black king went after the white pawns and Stockfish had a won B vs R position. Stockfish's lead is 2 points again.

Fire reached a QN vs QB endgame with a pawn up after 21 moves against Gull. Gull held on for a while but after the queens were exchanged Fire was able to move all its pawns to dark squares, invisible to the black bishop, and Gull's position became totally lost. Fire's first win against Gull this stage.

Gull - Jonny entered a closed position early in the game and both engines shuffled for a long time. Gull was ahead on eval so it was unexpected when Jonny moved pawns and reset the 50-move counter. Evals started to jump and Gull won a pawn, but the engines started to shuffle again. This time it was Gull who moved a pawn, causing black's position to open and eventually winning a knight and the game. This was Gull's second win of the stage, and first time it won a pair of reverse matches. Both wins were against Jonny.

Stockfish had an easy game against Rybka, with a big eval advantage and a rook for a bishop already at move 20. Evals continued to increase while Rybka lost a piece, Stockfish with a rook advantage is too strong.

Komodo - Andscacs was the reverse game of Komodo's loss in the previous RR. Andscacs held on for about 30 moves with a small eval advantage for Komodo. With a pawn sacrifice Komodo initiated a series of exchanges, after which Komodo's pieces were much stronger than Andscacs' and the evals were over 2. The RN vs RN ending started with equal material but after a while Komodo was a pawn up. The white king was very active in the center as a third piece. The game ended with Andscacs facing material loss and queening by white. Komodo got its revenge.

Houdini didn't get its revenge on Stockfish. Stockfish equalized quickly and exchanged pieces until a draw was declared. Houdini and Komodo are tied again after 3/7 games of this RR.

Rybka - Gull was quite dull for about 50 moves, reaching a RB vs RB ending which seemed like a dead draw. Then Gull allowed Rybka to exchange rooks, and suddenly Rybka had a big advantage since Gull's pawns were targets for the white bishop while Rybka's pawns were safe. It took a while but the black pawns started to fall and Gull realized it was lost. This was Rybka's first win of a reverse pair of games.

Fire - Rybka set a new length record for the stage: 205 moves, 6:40 hours of waiting for a draw.

Komodo couldn't penetrate Stockfish's defense, Houdini had hope for about 25 moves against Fire. Only 2 games per engines left in this RR, Komodo and Houdini still tied in second place.

Jonny started with an eval advantage against Houdini, but Houdini equalized quickly and took the initiative. Jonny had a bishop pair that could hardly move, while Houdini's knights were very active on the king side. Houdini exchanged a knight for a bishop to clear the white king side pawns and gain a passer on the 3rd rank. Jonny had to give a piece to take out this threat, leading to a RB vs RBN position which Houdini won. Once again Houdini leads Komodo by 0.5 points, one game left in the RR.

Komodo - Gull (game missing from archive, used chessbomb to review game but it's hardly the same): Komodo sacrificed a pawn and a rook for a knight but got a strong attack against an uncastled king. Gull gave the material back but its king was trapped and Komodo had two passers. Komodo had no problems converting its advantage to a win.

Stockfish - Fire started with an eval advantage for white with evals close to 1. For a long time evals remained almost constant while the engines exchanged pieces. The games reached a RRN vs RRB position with Stockfish a pawn up, all the pawns on the king side. Stockfish's eval started to move after move 50, and one pair of rooks was traded. The endgame was very hard for me to understand. Stockfish was able to put pressure on the black king, and the evals really jumped after Fire moved a pawn to avoid mate. This created a passer for Stockfish and the game was over.

Houdini - Rybka: Rybka held for about 20 moves before Houdini's eval started to increase. After the center was cleared Houdini had a passer. Most pieces were exchanged and in a Q vs RR position Rybka had to give a rook to stop the pawn, leading to a won position for Houdini. The last game of the RR and Houdini is back in second place, 0.5 points ahead of Komodo.

Featured game: Stockfish - Andscacs
Stage 3, round 39
Link to game on TCEC

The pieces to watch in this game were the white knights. Right from the start they made a stand on the queen side, a bit off center but still interfering with black's development.

Andscacs' queen tried to do something on the king side, Stockfish drove it to a weak square on the side. Meanwhile the white knights moved to the center, one following the other mutually guarding each other.

Stockfish now went after the queen side, opened it up and won a pawn. Andscacs could not get enough defenders around and constantly had to watch out for forks and discovered checks. 

Stockfish tried to exchange pieces, and with less pieces on the board the knights could seperate and be more active. One knight was involved in threatening the black king while the other put pressure and later won a second pawn.

After the queens were exchanged the white knights made sure the black king would remain in its corner and it was just a matter of time before Stockfish's extra pawns became dangerous, winning the game.

Monday, August 22, 2016

Season 9, stage 3, RR 5


It was almost ready and it got deleted and now I can't find it anymore. I apologize to you the readers, but I won't write it from scratch. TCEC does not wait and 28 games are a lot to go over again. It's a shame since RR 5 was very exciting. What I will do is give a short overview of what happened.


Here we go:

Table after 5 RR

Stockfish is still firmly in the lead, 1.5 points ahead of the field. RR 5 has not been great for Stockfish with only one win (and no loss), but it was a quality win against Houdini. Stockfish also dodged a few difficult positions as black, perhaps an indication it will do better in the reverse games.
The race for second place is fought hard between Houdini and Komodo. Komodo's loss to Andscacs was a shocker, and Houdini lost its first game this stage. Komodo is not as dominant as I thought it would be, perhaps it's just bad luck. With 3 RRs left this race can go either way. Doubly important are the head-to-head matches, so far 5 draws.

Some of the game highlights

Houdini - Komodo was another draw for these engines, still tied for second place.

Houdini beat Andscacs to get a 0.5 points lead in second place. The game reached a queen ending with Houdini 5 (!!) pawns up, after a nice king side attack.

In the next game Andscacs shocked everyone and beat Komodo. The opening seemed quiet enough, and it seems that Komodo blundered, not taking the king side attack possibility seriously enough. Andscacs executed the attack well, leading in the end to material advantage and a pawn on the 7th rank.

Komodo's unexpected loss was a great opportunity for Houdini to extend its lead in the race for second place. However, in the next game Stockfish beat Houdini. Three consecutive decisive games, and Houdini's first loss of the stage. Again the opening seemed innocent, and we got to see Stockfish in one of its finest moments of the season with a well coordinated attack on an uncastled king. Just before the final mass exchange the position was:

Just incredible !! As a result of all this excitement Stockfish extended its lead at the top and Houdini is still 0.5 points ahead of Komodo.

Houdini used its superior endgame skills to beat Gull in a rook endgame. Gull's lack of tablebases may also have contributed to its loss. Houdini now 1 point ahead of Komodo.

Andscacs put Stockfish under some pressure but Stockfish managed to exchange most pieces and get out safely with a draw.

Stockfish - Komodo was a relatively quiet draw.

Komodo beat Rybka, and immediately Houdini beat Jonny, no change in the race between these two.

Gull - Komodo was a long Q vs RR endgame that was well on its way to a draw. however Gull blundered yet again - not sure whether tablebases would have helped here, causing Komodo's eval to jump to 6, winning several moves later. A free 0.5 point for Komodo.

Fire pressed Stockfish hard, with a rook to bishop advantage. Stockfish exchanged down to a RN vs BN ending and with almost no pawns left Fire didn't have enough for a win.

Rybka held Houdini to a draw after Houdini had the advantage in black. At the end of the RR Houdini is 0.5 points ahead of Komodo.

Tuesday, August 16, 2016

Season 9, stage 3, RR 4

Table after 4 RR

(I will stop anouncing the draw rate, it is clearly around 65%. The final rate will appear in the statistcal report of the stage)

Stockfish extended its lead to 2 points, drawing the rematch against Komodo and beating Fire (only the second loss for Fire this stage). Could this be Stockfish's season?
Komodo and Houdini are tied in second place in a very close race. With half of the stage over, only 1/12 games within the top 3 engines has been decisive so far. The race may be determined by the results of the games against less strong opponents.

Some of the game highlights

Komodo had an eval advantage against Houdini but did not have enough for a win. Houdini still second by half a point.

Jonny had a significant eval advantage against Stockfish (Jonny 2.5 vs Stockfish 1), but when Stockfish exchanged pieces the evals came down. Jonny could only get a draw in a QBB vs QNN position.

Gull had a significant eval advantage from the start against Jonny. There were many early exchanges and the game reached a RN vs RB position, with Gull's rook controling the 7th rank. Gull had a queen side passer that marched quickly, and through a knight sacrifice it was able to queen and force Jonny to lose a rook for it. This was Gull's first win of the stage, with a lot of help from Cato (Jonny won the reverse game in RR 3).

Stockfish - Rybka reached a RB vs RN endgame, Stockfish had an eval advantage but the evals were stable for a long while even when Stockfish won a pawn. Evals started to rise rapidly around move 65, Stockfish outplayed Rybka in the endgame forcing a bishop and knight exchange leading to a won rook ending. Stockfish extended its lead by another 0.5 point.

Komodo had a pawn advantage and eval over 1.5 early in the game against Andscacs. For many moves Komodo didn't find a way to improve its position, even after trapping a black bishop. The breakthrough came when Komodo moved its king to a safer location and forced Andscacs to give a rook for a bishop to avoid a mating net. The QR vs QB endgame was far from easy, but Komodo had no problem getting the win. A double win for Komodo against Andscacs in this opening.

Houdini maintained pressure for 50 moves against Stockfish, but it was all over after most pieces were exchanged resulting in a drawn RB vs RB ending. Komodo is tied with Houdini, 1.5 points behind Stockfish.

Fire - Rybka got into a static pawn line standoff, Fire's eval a bit over the draw rule threshold. As the game approached the 50-move draw Rybka saw an opportunity to win a pawn, leading to a higher eval and more moves. Then with time running out Fire found an opening and created a passer on the 7th rank while its king hid behind a black pawn. Rybka had to use a bishop to block the pawn, then it lost a rook for a bishop and lost the game very quickly. 

Komodo - Stockfish: the rematch. This was the game everyone was waiting for, following the 1-0 result of the previous RR. Komodo fans wanted Komodo to win, so they could blame Cato's opening for the results and get Komodo back to a superfinal position. Stockfish fans wished for a draw so they can prove Stockfish is the best engine. Both claims are ridiculous based on one game, but we're talking about fans, not statistics experts. There was an interesting post by LPZ, a frequent chat contributor, showing the results of a 5-game simulation he ran of this game using similar hardware and conditions. With 4 draws and one Komodo win there was hope for everyone. The actual game followed the simulations for about 20 moves, Komodo with an eval of 0.8 and Stockfish with an awkward king side. Komodo held the tension for another 20 moves but then pieces started to be exchanged and Stockfish untangled itself. Komodo couldn't penetrate the white fortress, draw.

Andscacs was a pawn up early in the game against Gull, and then it exchanged pieces until only QNN vs QBN were left. With a passer on the 7th row, a pawn advantage and an active king after further exchanges, Andscacs had no problem to get the win.

Gull had an eval advantage against Stockfish, but Stockfish defended well and nothing came out of Gull's attempts. Stockfish's eval started to increase after 40 moves, but it only went up to 0.6 in a QRN vs QRB position. Then after what looked like shuffling moves Stockfish found a way to exchange pieces and reach a winning 6-man rook ending. Gull's lack of tablebase support was probably a factor in its loss. A double win for Stockfish in this opening, extending its lead in the crosstable.

Komodo started with an eval advantage against Gull and soon found a way to improve. It sacrificed two pawns to destroy the black pawn structure, then it won a rook for a knight. In a RR vs RN position the engines exchanged pawns until Komodo found a way to exchange pieces and reach a winning R vs N ending. First win for Komodo against Gull this stage. Second loss in a row for Gull, again tablebases could surely help it.

Houdini crushed Rybka in one of the shortest wins of the stage. Rybka's king side was filled with pieces that could hardly move, and when trying to defend against Houdini's passers it lost a rook for a knight and then another bishop. The material advantage was enough for an easy win for Houdini.

Featured game: Stockfish - Fire
Stage 3, round 28
Link to game on TCEC

The game started with evals close to zero, a sure sign of a draw. Fire was very hard to beat in the stage, another sign. Then Stockfish's eval started to increase following an inaccuracy by Fire.

Stockfish applied pressure on the black king with its rooks and through the long diagonal. Fire tried to defend but then its knight got trapped.

After taking the knight the g file opened and the white king was exposed. Stockfish had to be careful not to allow Fire counterplay. After the king side attacks were over Fire created a dangerous passer on the queen side.

Stockfish calmly let the pawn reach the second rank before taking it. Fire had no more offensive options, and with pieces being exchanged the white bishop advantage became bigger. The game ended in a winning RB vs R position. This was only the second loss of the stage for Fire, Stockfish firmly leading the table.

Friday, August 12, 2016

Season 9, stage 3, RR 3

Another busy week for me, so the post is a bit late. I hope to catch up with the games this week... 

Table after 3 RR

The draw rate is 56/84 (67%). 
Stockfish had a wonderful round, 4 wins, first win with black and it now leads the table with a 1 point margin. The most notable win was against Komodo, first decisive game between these legendary rivals this season. Komodo's round was below average for it with one win and one loss, it is now only third behind Houdini.
Can Komodo make a comeback? Is Houdini strong enough to qualify? Are these results just the result of an opening bias that will correct itself in the next RR? Exciting stuff...

Some of the game highlights

Stockfish had an eval advantage from the opening against Jonny, and it gradually increased as the game continued. First Stockfish created a passer, this was then converted into a pawn advantage, and finally two pawns advantage in a QB vs QB ending.

Komodo had a knight for pawns imbalance against Fire but there were not enough pieces on the board to get anything more than a draw.

Jonny had a pawn advantage in a RRB vs RRB position against Gull. It took a while for Jonny to find the best way to continue. It gave the pawn back to gain a passer, Gull then gave a rook for a bishop to get rid of the passer. The RR vs RB ending was slow, with all pawns on the king side it looked difficult, but Jonny managed to convert a win.

Houdini had an eval advantage from the opening against Andscacs, but evals did not change a lot for 30 moves. Houdini then improved its position, winning a pawn in a RN vs RB ending. Andscacs could not hold and Houdini won another pawn for the win. Houdini still only 0.5 behind the leaders.

Komodo equalized very quickly playing black against Andscacs, and then Komodo's evals started to increase. Material was equal and the most visible weakness for white was a more exposed king, but still evals were at 1 and increasing. A few moves later Komodo won a pawn, and after several exchanges Komodo was two pawns up in a QB vs QN ending, enough for a comfortable win.

Gull was close to a win against Rybka, the game reached a BB vs BN ending in a closed position with evals over 2 for Gull. Apparently Gull blundered in the end and allowed Rybka to close the pawn line, block the white king from approaching and force the draw. Gull is still the only engine with no wins in the stage.

Fire - Jonny: Jonny sacrifices a knight for three pawns, and when the center opened up it turned out that the extra piece was more important. Fire traded a bishop for a rook and then excanged pieces to get a winning RB vs R ending.

In a rare situation Komodo was a pawn down against Jonny and evals favoring black. However after exchanging pieces the game reached a drawn rook ending.

Andscacs started with an eval advantage against Stockfish, but it was not careful enough allowing Stockfish to have connected passers in the center. Andscacs was not able to stop the advancing pawns, supported by all the white pieces. When Stockfish started to win material the game was over. The first win in black for Stockfish.

Houdini had an eval advantage against Jonny, and after several pawns were exchanged evals climbed above 1. Houdini's bishop pair was strong in the open position, together with the queen they effectively targeted the black pieces. Houdini was able to capture a rook for a bishop and that was enough for a win.

Stockfish - Gull was in a closed position for a long time. Stockfish won a pawn in a series of exchanges and the game reached a rook ending. The white king was able to penetrate and assist the white passer, causing black's position to collapse. Three wins in a row for Stockfish.

Rybka was a pawn up early in the game against Houdini, but Houdini compensated with a strong king side attack. Rybka sacrificed a knight to try to relieve the pressure but its king was too exposed and its pieces vulnerable to attacks. Houdini won the game easily.

Featured game: Stockfish - Komodo
Stage 3, round 19
Link to game on TCEC

What a battle this was! Stockfish was a pawn up very early and had an eval of around 0.8. Komodo's eval was close to 0 and for a long time there was no apparent purpose to Stockfish's moves and the evals remained steady.

Stockfish gave up a pawn and opened up the center, the evals started to slowly increase. The advantage was starting to show as Stockfish had two connected passers on the queen side. Can Komodo hold?

As the endgame approached Stockfish regained the pawn and won another. With passers on both sides Komodo's defense became impossible.

When the pawns started marching the game was over. A huge result for Stockfish.

Friday, August 5, 2016

Season 9, stage 3, RR 2

I was away for a few days and had to race through playbacks of the games, unfortunately TCEC waits for no man. At least the draw rate is still high, most of the draws I just glance at to see if anything exciting happened. I hope to get back to live viewing by next week.

Table after 2 RR

The draw rate is 38/56 (68%).
With the first two rounds over the table is split into a top half and a bottom half. There has been only one decisive game so far within the top 4 engines (Komodo - Fire), so anything can still happen. Houdini is within half a point of a superfinal spot, will it cause an upset this season?
In only 1/28 openings both engines won, Cato's opening choices continue to be successful, being just biased enough to show differences in strength of engines.

Some of the game highlights

Komodo - Houdini: There was a lot of interest in this game after the controversy of the first game of the stage. Komodo took the game to a different direction right from the first move. The eval advantage of the opening came down temporarily, and then went back up when Komodo won a rook for a bishop. The game reached a RN vs BB ending, and the evals around 1. Houdini defended very well, the bishops had pawn support and Komodo had no easy target. It took a very long time for the evals to get down but the game was a draw.

Rybka - Andscacs: Andscacs somehow got itself into giving a rook for a strong pawn position with three connected passers. Both engines saw that Rybka had the better deal, and though black got the material back white could sacrifice pieces to assure queening and win the game. Andscacs' loss was a bit of a surprise for me, perhaps the opening was too biased.

Stockfish was up a rook for a knight against Rybka, but the RRB vs RBN endgame was static and evals were only at 0.4. Both engines were playing on increment and the 50-move draw was getting close, but Rybka blundered the endgame allowing Stockfish to win a piece and the game.

Komodo was two pawns down against Andscacs but had a strong knight on the 6th rank and prevented black from castling. Andscacs' position collapsed when it allowed Komodo to attack the back ranks with rooks. After a series of exchanges the pieces left were QNN vs RRN, and with an exposed king and too many unguarded pieces black was lost.

Jonny was quite happy with the Q vs RRB, 4 pawns up position against Komodo. Komodo thought the best Jonny could do was get a draw, but then Jonny lost two pawns and Komodo had the initiative. With less pawns the black piece advantage became too strong, and the only defense was giving checks. It was slow and tricky, but Komodo got the win, double win against Jonny with this opening.

Fire - Rybka reached a QR vs QR endgame that looked drawn. Rybka's king was exposed to checks, Fire used this to gain a pawn. The position was probably still drawn but Rybka blundered the endgame and let Fire win another pawn and reach a won 6-man queen ending.

Gull - Houdini started with an eval advantage for Gull but Houdini exchanged pieces and equalized. Then, in a QRR vs QRR position Houdini started to dominate. The white king was vulnerable and had to be defended, and Houdini had a dangerous passer. The evals rose higher and higher, and after queens were exchanged the passer was worth a rook, game over.

Stockfish - Andscacs was more or less balanced for about 20 moves. Andscacs did not take Stockfish's king side attack seriously enough, and was very sorry a few moves later. Stockfish took a bishop for a rook and left Andscacs' king with very little defense, the game was stopped before its position collapsed. Four decisive games in a row, a full round.

Komodo played most of the game with a Q vs RN imbalance against Stockfish, but could not get past the Stockfish defense.

Andscacs - Gull was headed for a 50-move draw, but with both engines playing on increments Andscacs' eval started to rise. It started with a small crack in the pawn line, and after slowly exchanging pieces Gull's whole pawn line crumbled, Andscacs won a piece for a passer and reached a winning 6-man RNP vs R position.

Andscacs came very close to beating Jonny. It had a few choices of endgames with a pawn advantage, but the choice it made led to an opposite color bishop ending with two (doubled) pawns up, and it wasn't enough for a win. 

Featured game: Jonny - Houdini
Stage 3, round 13
Link to game on TCEC

Jonny had space advantage against Houdini, with the queen side pawns locked. The game started with an eval advantage for Jonny but Houdini equalized without too much trouble. 

Houdini then started to use the semi open king side to create pressure on the white king. Jonny couldn't get enough defenders and got into a lot of trouble very quickly, the evals now in Houdini's favor and increasing.

The king side opened up and after a few exchanges the white king had to run as the pawns around it were gone. The material was still equal but Jonny was under a lot of pressure.

Jonny exchanged pieces and was only a pawn down in a RB vs RN ending, but there was no way it could defend its pawns. A nice win for Houdini in black, still with a good chance to make it to the superfinal.