Saturday, January 27, 2018

Season 11 division 1, after 2RR

Crosstable after 2RR



The current draw rate is 44/56 (78.6%). The high number of draws could be explained by the bookless openings. Another possibility is that the engines in division 1 are similar in strength. I hope that in the next RRs the book sequences will help reduce the draw rate.

The score differences between the engines are small so far, only 2.5 points separate first and last place after 14 rounds. Fizbo, Andscacs and Gull are leading, these are the engines with a positive score. The rest of the engines are more likely to struggle against relegation. The division 2 engines Laser and Jonny are in the bottom of the table together with Nirvana. Laser has 4 losses and is still in the race due to its wins against Jonny and Nirvana. I still think it is the most likely engine to finish last in the division.

There were two crashes so far, Nirvana crashed on move 1 against Fizbo, Hannibal crashed against Nirvana.

A few interesting games

Laser - Andscacs: Andscacs equalized after the opening, and following a series of exchanges only RRB vs RRN were left at move 22. Laser had an isolated passer that Andscacs blocked with the knight, Andscacs had a pawn majority on the king side and was ahead on eval. The pawns on the board limited the rooks' movement, Andscacs found a path for one rook and was able to capture a pawn. Evals were over 1 and rising, Andscacs brought its king to support the king side pawns. Laser exchanged a pair of rooks but could not avoid defeat. This was the first decisive game after 9 draws. It is interesting to note that Andscacs won the endgame without tablebase support, while Laser does have access to tablebases.

Hannibal - Jonny: The game reached a 6-man drawn position KQP vs KRB. Jonny knew this was a draw but Hannibal did not have tablebases support and gave a winning eval, even a ridiculous mate in 15 eval at one point.

Nirvana - Fizbo: Fizbo was sure it was ahead, with evals constantly increasing. After exchanging almost all pieces the game reached an opposite color bishops ending, and despite being 2 pawns up and evals over 3 Fizbo could only get a draw.

Jonny - Gull: There were almost no exchanges for the first 25 moves and evals were close to 0. On move 34 Gull created a pawn majority on the queen side, Jonny chose to trade a knight for a pawn and eliminate the queen side pawns. The extra piece was enough for Gull to win the game. Why did Jonny give a piece for a pawn, was it losing in any case?

Laser - Nirvana: Most pawns remained on the board in the beginning of the game, all knights were off on move 24. Laser's pawn structure looked a mess, but it was ahead on eval. It managed to capture a pawn and then stabilize the king side with its bishop. Evals were over 1.5 after several exchanges that left QRB vs QRB, Nirvana with a queen side passer and Laser 2 pawns ahead. Laser chose to exchange queens, and since the bishops were of opposite color there was a potential for a draw despite the eval and pawn advantage. However, after a long period of shuffling Laser found the winning line, its king joining the king side pawns as they marched forward.

Andscacs - Fizbo: Fizbo came out of the opening with a better pawn structure. On move 34 it was a pawn up, with a passer on the queen side an eval over 1.5. Andscacs defended well, it captured the passer and exchanged down to a RN vs RN position. Fizbo continued to fight but Andscacs secured a draw. It was surprising for me that Andscacs could get in trouble with white, and with no book. Perhaps Fizbo is stronger than I think.

Booot - Gull: Booot sacrificed a bishop for 2 pawns and exposed the black king, with a small eval advantage for white. Gull gave the material back and the engines exchanged pieces until reaching a rook ending with Booot a pawn up. Gull was sure it was safe, its eval remained constant. Booot, though, found a way to improve and realized it was winning. The game ended in a tablebase RP vs R win (usually these are draws).

Laser - Hannibal: Only RB vs RB were left by move 27. Laser had tablebase support and gave a rook for a bishop to get a 6-man draw. Hannibal had no tablebase support and it thought it was winning for 70 more moves.

Gull - Laser: Gull outplayed Laser after the opening, with the black queen stuck in the queen side Gull took over the center with a strong bishop. Laser exchanged knights for the two strong bishops, but then Gull was a pawn up with central passers. Evals continued to increase as Laser was forced to lose a knight to stop the passers. The game reached a RN vs R position, a win for Gull. 

Fizbo - Booot: Booot played the opening badly and Fizbo's eval was over 1 after 10 moves. Fizbo's king remained in the center while it opened up the king side and threatened the black king. Booot exchanged queens and a pair of rooks, in a RBN vs RNN its king was out of immediate danger but evals were climbing. Fizbo's pieces were much better placed, Booot's knights could hardly move and its king was still under threat. Fizbo won a pawn and the game was adjudicated before we could see the actual win.

After 1RR Fizbo, Andscacs and Gull are leading at +1; Jonny, Nirvana and Laser are behind at -1; Booot and Hannibal are at 0. 

Fizbo - Laser: Fizbo attacked on the king side, Laser realized it was in danger when it was too late. The defending black pieces could hardly move and Fizbo began capturing pawns and advancing a passer. The game was quickly over.

Gull - Nirvana: The game reached a BN vs NN position on move 34 with evals close to 0. Nirvana chose to sacrifice a knight for pawns, but this turned out to be a mistake. Gull had enough time to stop two queen side passers and another on the king side, and it kept a single pawn that was enough to secure a win.

Laser - Jonny: Jonny sacrificed a bishop for pawns to expose the white king. Laser secured its king and there was no immediate threat. The engines exchanged pieces until only QRB vs QR were left. Jonny had 3 extra pawns but it was Laser's central passer than made the difference. Laser advanced the passer to the 7th rank, tying down the black rook. The white bishop was the same color as the queening square, and eventually Laser used it to capture the black rook, winning the game.

Featured game: Andscacs - Laser
Division 1, round 14
Link to game on TCEC

Andscacs had a small eval advantage from the opening, it increased to about 0.6 and stayed there for a long time. Laser seemed to be holding in a RRN vs RRB position with a slightly worse pawn structure.



Andscacs' king came forward with evals jumping over 2. The king was the extra piece that was needed to break Laser's defense. The black pawns were all isolated, the pieces guarding them needed to be protected as well.


After exchaging a rook pair Andscacs focused on the black f pawn. It was protected by the bishop and the other black pieces could not help and had other targets to defend.



Laser had to let go of the pawn eventually. This created a passer for white, enough to win the game.

Season 11 division 1, bookless openings summary

The games of first two RRs of division one were played without book. Here are a few statistics of the openings:

First move
33/56 (58.9%) start with d4
23/56 (41.1%) start with e4

By engine:

Andscacs: 3 d4, 4 e4
Booot: 6 d4 1 e4
Fizbo: 4 d4 3 e4
Gull: 3 d4 4 e4
Hannibal: 4 d4, 3 e4
Jonny: 7 d4
Laser: 5 d4 2 e4
Nirvana: 1 d4 6 e4

Jonny, Booot and Laser prefer d4, Nirvana prefers e4.

Opening by first ECO letter

A - 2 (3.6%), both Queen Pawn Games similar to D, no English, Dutch or Reti
B - 5 (9.1%), 4 of which are Caro-Kann, one Sicilian
C - 18 (32.7%), of which 15 are French Exchange, 2 are other French variants, and one Giuoco Piano, no Ruy Lopez
D - 25 (45.5%), 18 of which are QGD variants.
E - 5 (9.1%), 3 of which are Nimzo-Indian

Decisive games openings

D47 Semi-Slav: Meran, Capablanca (black win)
D35 QGD: Exchange (one black win, two white wins)
D37 QGD: Westphalia (white win)
D38 QGD: Ragozin (white win)
D00 Queen's Pawn: Mason Variation (white win)
A46 Indian: London (white win)
C01 French: Exchange (two white wins)

The crashes are not included.

Thursday, January 25, 2018

Season 11 division 1, openings after 1RR


The games of first two RRs of division one are played without book. Here are a few statistics after 1RR is over:

First move
18/28 (64%) start with d4
10/28 (36%) start with e4

By engine:

Andscacs: 2 d4, 2 e4
Booot: 3 d4 1 e4
Fizbo: 2 d4 1 e4
Gull: 1 d4 2 e4
Hannibal: 3 d4, 1 e4
Jonny: 3 d4
Laser: 4 d4
Nirvana: 3 e4

Jonny and Laser start with d4, Nirvana with e4. The other engines tried both options.

Opening by first ECO letter

A - 1 (3.6%), no English, Dutch or Reti
B - 3 (10.7%), all Caro-Kann, no Sicilian
C - 7 (25%), of which 6 are French Exchange, and one Giuoco Piano, no Ruy Lopez
D - 13 (46.4%), 9 of which are QGD variants.
E - 4 (14.3%), 3 of which are Nimzo-Indian

Decisive games openings

D47 Semi-Slav: Meran, Capablanca (black win)
D35 QGD: Exchange (one black win, one white win)
D37 QGD: Westphalia (white win)
D38 QGD: Ragozin (white win)
D00 Queen's Pawn: Mason Variation (white win)

Sunday, January 21, 2018

Season 11 division 2 statistics

Draw rate, wins


Final draw rate was 50.0%.


Game termination

The three most common game termination causes were:
46.4% - TCEC win rule
19.6% - TCEC draw rule
18.8% - TB position

There was one crash in the division.

Moves per game



Median= 62.3
Average= 67.6

The game lengths were similar to those in division 3.

Time per game (hours)



Median= 1:39
Average= 1:38

Openings

There were 2-move book openings in this stage chosen by Cato. The first letter of the ECO codes was distributed as follows:


For each round all four games had the same book sequence, and again in the reverse round. There were 14 book sequences, each repeated 8 times.  For some sequences all the openings had similar codes (e.g. Sicillian or French). In others the engines had more freedom to choose (in rounds 17 and 24 there were 4 Grunfelds, 3 King Indians and one Maroczy Bind Sicilian).

If we use the opening 'family name' (using format FAMILY_NAME: VARIANT....) the top 3 are:
Sicillian - 25 times
English - 9 times
QGA - 9 times
Queen's Pawn - 9 times

The engines had a lot of freedom to choose the opening variant. 39.3% of the game pairs repeated the same ECO code twice, and 16.1% repeated the same opening variant. In 12.5% of the game pairs the ECO first letter was not repeated twice.

Reverse pairs, wins



There were 6 game pairs with a 1-1 biased result. It is likely that the engines playing have more influence on the outcome than the 2-move opening, especially when other games with the same opening have different results.


Reverse pairs, same moves 

Pairs of reverse games diverged very quickly, 46.4% of the pairs diverged immediately after book, 87.5% repeated at most 2 plys. The maximal repeat length was 9 plys (rounds 19 and 26, Arasan - Jonny).




 

Season 11 division 2 final results

Division 2 is finished and the qualifiers for division 1 are Jonny and Laser.

The final crosstable is



The final draw rate is 56/112 (50%). There were more draws in the second half of the division, the bottom two engines had much better results and this is reflected in their score and in the draw rate.

Laser and Texel were very close from the start to the very end. Laser lost its lead after RR 3 with a loss to Bobcat, its only loss in the division. In RR 4 Texel lost twice, against Bobcat and Arasan, Laser regained the lead and held it to the end. This is Laser's second promotion of the season, after starting in division 3.

Bobcat and Wasp are the bottom 2 engines and will be relegated to division 3 next season. Bobcat scored 6 points in the second half of the division, with 4 wins, but could not close the gap after its poor score in the first half.

There were no crashes in the second half of the division, Fritz' crash in the first round remained the only crash of the division and had no impact on the final results.

The qualifiers will now join Andscacs, Booot, Gull, Fizbo, Hannibal and Nirvana in division 1. Based on the results of season 10 I would assume that Andscacs, Booot and Gull will fight for promotion and the rest will fight against relegation. There is a good chance that Laser will be relegated back to division 2, hopefully it will have a better score than Wasp had in division 2.

The time control in division 1 will be 60+10, slowing the pace a little more. There will also be 4 double RRs, twice as many games as in divisions 2 and 3. I hope to have time for more live blogs between reports. The openings in division 1 will be interesting. This season TCEC invited a fan to select the openings for 3 DRRs, while the first DRR will have no book at all. Cato can still be blamed, naturally.

A few interesting games
Laser - Arasan: Laser had an eval advantage after the opening, Arasan took over after 20 moves with a threat on the white king. After most pieces were exchanged Arasan's eval was over 1.5, the game reached a queen ending with Arasan a pawn up. Laser managed to hold, still in second place. The leading 3 engines drew all their matches in rounds 15-17, no changes at the top of the table.

Bobcat - Laser: Bobcat captured a pawn with its queen early in the game. The queen looked trapped and didn't move for most of the game, but instead of a poisoned pawn situation Bobcat's eval constantly increased. It had a pawn advantage on the queen side, and Laser had no counterplay and only defended. After 30 moves the white queen became active again in a QRN vs QRN position. For a while it seemed Laser found a drawing line, after exchanging queens and trading pawns for a knight the evals dropped below 1. However the white pawns were too strong and Bobcat won the game eventually. A big surprise and Laser's first loss in the division. Bobcat is fighting against relegation, Laser is now 3rd after Texel.

Bobcat - Fritz: Both engines thought they had an advantage in the middlegame. Fritz had a bishop pair while Bobcat had a knight pair. Fritz opened the position, trading a rook for a knight and a pawn. Despite Fritz's passer and Bobcat's exposed king, Bobact's rooks and queen took over the open files on the queen side. After exchanging pieces the game reached a RR vs RB position, won for Bobcat. 

Texel - Laser: Texel had an eval advantage after the opening, its pieces with a lot of space. After several exchanges the position opened up and Laser traded its queen for RN. Evals were around 1.5 in a QB vs RBN position, and they increased over 2 at some point. However Texel could not break Laser's fortress and the game ended in a draw. Laser keeps its chances of promotion.

After 3RRs Bobcat is 1.5 points below Vajolet2 still with a small chance of avoiding relegation, Jonny is leading Texel by 2 points, Laser is 0.5 points behind.

Bobcat - Texel: Texel was a pawn up from the start but Bobcat had the eval advantage. Texel had passers on the queen side and Bobcat had a strong center and a king side attack. Texel gave material in an attempt to stop the attack, after several exchanges Bobcat was up a rook for pawns. The queen side passers were not strong enough and Bobcat converted the win. Bobcat is getting better results in the second half, trying to close the gap to Vajolet2. A setback for Texel in the promotion race.

Vajolet2 - Bobcat: Most pawns remained on the board with one open file in the center. The engines blocked the queen side and exchanged pieces through the open file, Vajolet2 had a small eval advantage but the game was heading to a draw. Then Bobcat blundered in the bishop ending and lost the game. This head to head loss against its main rival means that Bobcat has almost no chance to catch up with Vajolet2.

With 5 rounds remaining Bobcat is 2.5 points behind Vajolet2, Laser and Texel are equal.

Laser - Bobcat: After 20 moves the pawns became static and the engines slowly exchanged pieces on the queen side. On move 40 only RRN vs RRB remained and the engines were mostly shuffling with an occasional pawn move. Laser's eval started to increase when all rooks were exchanged, the knight was much stronger than the bishop and after capturing two pawns Laser was winning.

Arasan - Texel: For 30 moves the game was balanced with most pieces on the board. Then Texel trapped its own bishop and Arasan's eval started to climb. Texel was forced to give the bishop for pawns and the evals were above 1.5. Texel tried to counter with its king side pawns but could not avoid losing more material and the game. Texel drops to 3rd below Laser.

With 3 rounds left Bobcat is still 2.5 points behind Vajolet2, relegation for Bobcat is almost sure. At the top of the table Laser is second 1 point ahead of Texel. Laser still has to face Jonny with the black pieces, as well as a head to head match against Texel. This race is not finished yet.

Laser - Texel: see my live blog of this game. The game had a slow start, almost all pieces stayed on the board for a long time. Laser pushed pawns on the king side and had a space advantage. A series of exchanges opened the center and Laser was a pawn up with evals around 1 after move 40. After a second series of exchanges only RBN vs RBN were left and evals started to drop. The game ended in a tablebase draw. Texel needed a win, but it was Laser that was ahead for most of the game. The draw almost guarantees a second place finish for Laser.

Arasan - Wasp: Arasan did not play the opening well, this was the only game of 8 Philidor Defenses played where black came out ahead in eval after the opening. Wasp started to press on the king side, advancing pawns and opening a file. Evals kept increasing and reached over 3 when Arasan gave a queen for BN to avoid mate. This only delayed the inevitable loss. Wasp's first and only win in the division.



Saturday, January 20, 2018

Season 11 live blogging, Saturday 20th January

Today I'm watching the round 27 game Laser - Texel. Laser is in second place, 1 point ahead of Texel with two rounds left. Texel has to win to have a chance of promotion. In the reverse game in round 20 Laser barely managed to draw, and there have been no black wins for this opening so far. Texel needs a miracle.

move 4: following the reverse game, small eval advantage for Laser.

move 6: Texel diverges with c6, looks like a rare move.

move 8: white is ahead in development, evals without change. Laser's a3 seems a little slow. Is it preparing b4, blocking something?

move 9: now Laser goes h3, I don't get it. Texel's eval drops, it doesn't get it either.

move 11: Laser has more space, Texel's pieces can hardly move.

move 14: almost all pieces are still on the board. Texel is seeing exchanges in the near future according to the PVs, Laser not so much.

move 17: Laser moved its LSB from the a2-g8 diagonal to the f1-a6 one, is that better? Meanwhile Texel's eval dropped to 0, and still no exchanges. Laser offered the b2 pawn, Texel isn't biting.

move 21: The b pawn is still hanging and Texel continued to not take it. Still no exchanges, Laser pushed g4, weakening its king slightly.

move 25: Laser pushed the f and g pawns exposing its king, though there does not seem to be a threat from Texel. Evals not moving.

move 28: Laser's eval jumps above 0.5, it has a central passer in the PV. Texel remains calm at 0.3.

move 31: Laser's eval still climbing, now above 0.7, Texel still calm. The black queen looks misplaced, like a huge pawn.

move 32: Texel's eval jumped above 0.9! I'm not sure what's changed, Laser was expecting this move.

move 37: Finally a few exchanges, opening up the center. Laser is a pawn up but both evals are coming down and the PVs show more exchanges coming. Texel may survive this game but this is still a good result for Laser.

move 41: a pair of rooks is gone, Texel is defending, can Laser find an effective attack?

move 46: Texel got the pawn back but the evals are over 1 now. Moves are played faster at this stage of the game.

move 54: after a series of exchanges we are down to RBN vs RBN. The PVs look drawish.

move 61: Laser is again a pawn up, but evals are dropping fast. Only RN vs RB now, looks more and more like a draw.

move 76: position is a 6-man tablebase draw, and we just have to wait for TCEC adjudication. Laser seemed to be ahead in the game and Texel defended well. Only one round left and Laser is still 1 point ahead of Texel, Laser will probably promote to division 1.

Wednesday, January 17, 2018

Season 11 division 2, after 2RR

Crosstable after 2RR
 


The current draw rate is 23/56 (41.1%). This is lower than the division 3 draw rate, and the reason is that the two engines at the bottom of the table are losing most of their games. Restricting to the top 6 engines the draw rate is 16/30 (53.3%)


Jonny is leading the table, confirming its result in season 10 stage 1 which was closer to the engines in division 1 than those in division 2. Laser in 2nd place is a bit of a surprise since it started in division 3 and has a good chance of getting promoted again. Laser's results are also surprising when compared to Fritz, the engine that won division 3. Texel in 3rd is also still in the promotion race only 0.5 points behind Laser.

The candidates for relegation to division 3 are Wasp and Bobcat. They are literally out of their league, scoring 1.5 and 2 points in 12 games against the other engines.

In the second game of the division Fritz crashed against Jonny, not a good start for the division or for Fritz. There was no time trouble and the position was balanced with a small eval advantage for Jonny. This was the only crash in division 2 so far, hopefully the second half will be crash-free.

A few interesting games
 
Texel - Jonny: Texel was ahead on eval after the opening, in a position with no knights and with all pawns still on the board. After several exchanges Jonny equalized and moved in front, with a connected pair of passers in the center. Texel had two passers as well but they were harder to support. Eventually Texel had to give a rook for a bishop to stop the black pawns, leading to a won RB vs BB ending.

Arasan - Fritz: For 30 moves the evals were close to 0 with almost no exchanges. Fritz opened a hole in the enter allowing Arasan to take the initiative. Arasan put pressure on the king side and Fritz gave a rook for a knight to try to get free. This led to a RR vs RN ending and a win for Arasan.

Fritz - Texel: Fritz had an eval advantage that increased after the opening and reached 1.3 when Fritz traded its queen for two rooks. Texel's eval was more conservative, and after the trade dropped to 0. Texel's queen was very active while Fritz' pieces were passive, Texel's eval turned negative after it captured a pawn and was over 1.5 after exchanging bishops in a QN vs RRB position. Fritz was outplayed in the endgame, after losing two more pawns the position was winning for Texel.

Arasan - Texel: The position was closed with most pawns on the board, evals favoring Texel. At move 31 both engines thought they were doing well, but Texel allowed its bishop to get trapped on the queen side and this turned out to be key. When the king side opened several moves later it was Arasan that had a strong king side attack with evals approaching 2. Pressing on the king side let Arasan clear the black pawns on the queen side. Just before adjudication Arasan finally captured the trapped bishop. 

Saturday, January 13, 2018

Season 11 division 3 statistics

Draw rate, wins




Final draw rate was 54.5%.


Game termination

The three most common game termination causes were:
40.2% - TCEC win rule
28.6% - TCEC draw rule
17.0% - TB position

The draw rule is back, much fewer 3-folds compared to division 4.
There were 3 crashes, both Pedone (one crash and one update strike) and ChessBrain (2 time outs) came close to being disqualified with two strikes.

Moves per game



Median= 63
Average= 69.6

The games were generally longer than in division 4, though a smaller fraction of the games 
were very long (only 4.4% were over 130 moves, compared to 10.7%).

Time per game (hours)



Median= 1:15
Average= 1:16

Openings

There were 2-move book openings in this stage chosen by Cato. The first letter of the ECO codes was distributed as follows:



For each round all four games had the same book sequence, and again in the reverse round. There were 14 book sequences, each repeated 8 times.  For some sequences all the openings had similar codes (e.g. Sicillian or Caro-Cann). In others the engines had more freedom to choose (in rounds 16 and 23 there were 5 English variants, one Sicilian, one QGD and one Nimzo-Indian defence).

If we use the opening 'family name' (using format FAMILY_NAME: VARIANT....) the top 3 are:
Sicillian - 25 times
English - 13 times
QGD - 11 times

The engines had a lot of freedom to choose the opening variant. 41.1% of the game pairs repeated the same ECO code twice, and 14.3% repeated the same opening variant. In 26.8% of the game pairs the ECO first letter was not repeated twice.

Reverse pairs, wins


There were 8 game pairs with a 1-1 biased result. However since every opening was repeated in 4 game pairs and only one pair is biased, we conclude that this is not an opening bias but depends on the engines playing.


Reverse pairs, same moves 



Pairs of reverse games diverged very quickly, 28.6% of the pairs diverged immediately after book, 85.7% repeated at most 2 plys. One pair repeated 15 plys after book, Fritz - Ethereal rounds 19 and 26, Russian Game, 2 draws.

Friday, January 12, 2018

Season 11 division 3 final results

Division 3 is finished and the qualifiers for division 2 are Fritz and Laser.

The final crosstable is



According to the rules published on the TCEC main site the last two engines, Ethereal and ChessBrain, will be relegated to division 4. However it is not clear whether this is the final decision.

There were no crashes in the last two RRs. Pedone was a strike away from being disqualified, but it managed to finish its 14 games without crashing. ChessBrain also had two time losses in the first half, but since it was ranked 7th another crash would not have made a difference.

Of the 7 openings of RRs 3-4 two had drawish results. Rounds 19 and 26 were 8 Russian Games with 6/8 draws, rounds 21 and 28 were mostly Pirc and Old Benoni with all 8 games drawn.

Ethereal had a rare win against Senpai in round 22. Senpai got into a worse position after the opening, and had to give a rook for pawns. The game was over quickly, with many pieces on the board and Ethereal with a significant material advantage.

Nemorino managed to beat Fritz in round 23, Fritz's only loss in the division. Nemorino had a strong king side attack that got it a two pawn advantage, and then it traded down to a winning queen ending.

Laser lost to Senpai in round 24, the only minimatch Laser lost in the division. Laser started the game with an eval advantage, Senpai equalized and took the initiative. Senpai exchanged pieces until only RB vs RB were left. Laser's king was under a mating threat, and Senpai's king was able to march forward and join the attack. Eventually Senpai captured the white bishop, winning the game.

Reducing the crosstable to the top 5 engines the ranks are as follows:
Pedone +2, Fritz and Laser +1, Nemorino =, Defenchess -4. This is a bit surprising to me. Fritz, Laser and Defenchess are ranked high mostly because of their positive results against the bottom 3 engines.   

The qualifiers will now join Jonny, Bobcat, Texel, Vajolet, Wasp and Arasan in division 2. I think Jonny has a good chance of qualifying since it had better results than the other 5 engines back in season 10. It's hard to say how Fritz and Laser will do in this division. Division 2 will have 2 double RRs, 28 games per engine, same as division 3. The time control will be 45+10, slightly longer games but still too fast for me.

Wednesday, January 10, 2018

Season 11 division 3 after 2RR

Crosstable after 2RR



Laser and Fritz lead the rest of the engines by 2.5 points and are the obvious candidates for promotion to division 2. Fritz has not lost a game yet, Laser lost two but won the reverse games so it is at least drawn against all the engines.
At the bottom of the table Ethereal and ChessBrain are 1.5 points below their closest rival and are facing relegation. ChessBrain has lost two games on time, I'm not sure whether these losses count as crashes. It was losing these games in any case. On the other hand there is no doubt about Pedone crashing once, and since it took a strike for a bug fix between divisions it is one crash away from being disqualified and automatically relegated.

There have been 7 2-move start positions so far, 8 games for each book sequence. In rounds 1 and 8 there were 8 Nimzowitsch-Larsen openings, 5 of which were black wins which is quite unusual. The games in rounds 7 and 14 were all Sicilian Alapin openings, all 8 were draws.

After a quick glance at the games there is one unusual event that is worth mentioning. In the round 13 game Ethereal vs Nemorino the engines got into a long shuffle in a RB vs Q position. Nemorino had a solid eval advantage close to 3 but Ethereal had a clear fortress and the game was heading for a draw. Nemorino avoided a 50-move draw and the shuffle continued, until on move 145 Nemorino simply sacrificed its queen, leading to a quick loss of the game. Nemorino's PV showed it immediately loses the queen, yet its eval was 1.34 in its favor. What a strange bug, similar to the c7f7 game of last season.

Monday, January 8, 2018

Season 11 division 4 statistics

Draw rate, wins

 

Final draw rate was 46.4%.


Game termination

The three most common game termination causes were:
48.2% - TCEC win rule
19.6% - 3-fold repetition
16.1% - TB position

The number of 3-fold repetitions was unusually high, the engines preferred repeating over lowering the eval to 0 and adjudication. There were 3 crashes, Pedone came close to being disqualified with two crashes.

Moves per game



Median= 55
Average= 65

The distribution has unusual peaks, in particular there were 6 games (10.7%) with more than 130 moves. This was mainly because some engines had problems evaluating drawn positions.

Time per game (hours)
  
Median= 1:13
Average= 1:13

Openings

There were 2-move book openings in this stage chosen by Cato. The first letter of the ECO codes was distributed as follows:


For each round all four games had the same book sequence, and again in the reverse round. There were 7 book sequences, each repeated 8 times. For some sequences all the openings had similar codes (e.g. Sicillian or French). In others the engines had more freedom to choose (rounds 3 and 10 were either QGD or Indian defences, and one Benoni).

If we use the opening 'family name' (using format FAMILY_NAME: VARIANT....) the top 3 are:
Sicillian - 16 times
Caro-Kann - 8 times
French - 8 times

The engines had a lot of freedom to choose the opening variant. Only 18% of the game pairs repeated the same ECO code twice, and none repeated the same opening variant. In 7.1% of the game pairs the ECO first letter was not repeated twice.

Reverse pairs, wins

 

The 3 "biased" results all involve Toga, this says more about Toga's ability than the openings which were all very popular, short and unbiased. 


Reverse pairs, same moves 


Pairs of reverse games diverged very quickly, 60.7% of the pairs diverged immediately after book.


Saturday, January 6, 2018

Season 11 division 4 results

Division 4 is finished and the qualifiers for division 3 are:
Defenchess, Senpai, Pedone, Ethereal, and ChessBrainVB

The final crosstable is



The qualifiers were determined in the last round, ChessBrain needed a draw against Senpai to finish ahead of Toga. Senpai was ahead but ChessBrain managed to hold.

There were 5 qualifiers and not 4 because the division 3 engine Fruit was pulled out by its author for unknown reasons.

Pedone qualified despite crashing twice, close to being eliminated by the 3 strikes rule. There was some discussion about whether to allow it to continue crashing in division 3, the ruling was to allow engines to be updated (without testing), at the cost of one strike in the next division. Pedone and Ethereal chose to update, hopefully they will not crash in division 3.

The qualifiers will now be joined with Fritz, Laser and Nemorino in division 3. Fritz and Laser are the favorites to qualify, but nothing is certain until the games are played. Division 3 will have 2 double RRs, 28 games per engine, so less luck is involved than in division 4. The time control will still be 30+10, too fast for me to go over all games and write reports. I'll try to write one at the half point.

Friday, January 5, 2018

Season 11 live blogging, Friday Jan 5th

Season 11 has started with division 4, a quick 2RR with eight engines trying to finish in the top 4 to qualify for division 3. The games are about an hour long each and we are about half way through the stage now playing the reverse games.

Now playing are Scorpio vs The Baron, currently in 7th and 8th place.

move 27: on the board black is a pawn up, with QRN vs QRN on the board. White has a passer on the queen side for compensation.

move 30: The queens are off, evals are negative in favor of The Baron, and increasing. Can Scorpio hold?

move 32: looks like The Baron is going to exchange rooks and take white's passer. My bet is that black will win the knight ending.

move 35: and there we go, Scorpio got a pawn back so The Baron is a pawn up, all pawns on the king side. Evals are at 2, not sure if this is a sure win. The engines could also blunder, especially Scorpio which is in some time trouble.

move 40: the kings and knights are dancing. The PV from The Baron doesn't show a lot of progress, Scorpio is not giving us any PV....

move 45: Scorpio's eval was at 0 for a move, it jumps like that sometimes. Still no pawn moves or captures, evals stable.

move 50: The Baron is expecting a pawn move but Scorpio is giving checks with the knight.

move 58: Eval jump for the Baron, does is see an improvement? The king tries to advance on the queen side.

move 62: There's the white pawn move, a zugzwang? Scorpio's eval jumps to almost 3 and the Baron's PV shows movement on the king side.

move 65: now the Baron's eval jumps to almost 4. This looks like a win for black. Now back to 3, we'll see in a moment.

move 69: pawn exchange on the king side, The Baron's eval over 5 now, probably a win. Scorpio not seeing it yet.

move 75: all over, the white pawns will soon fall.

Scorpio is probably out of the race, the Baron stilll has a small chance, it is 1.5 points behind 4th place with 6 games to go.

My time is up for today, tomorrow the stage will be over and the qualifiers will be decided.

Opening analysis, all seasons, late stages

Earlier in the season I accidentally came across a game whose opening moves were identical to a game in a previous TCEC season. This led me to think about whether TCEC openings have special properties that increase the probability of such repeats. I have been reporting on the openings as part of my statistics reports for each stage since season 8, but I have never looked at all the seasons and stages together.

Designing the opening book for TCEC is an important and difficult task, especially in the later stages. The opening book has conflicting goals. On the one hand the draw rate should not be too high in order to make the games exciting to watch. On the other hand openings should not be too biased as games are played twice with engines swapping colors, if each engine wins once the result gives no indication which engine is stronger. So the openings chosen should be biased in just the right amount, where a stronger engine has a better chance of finding a win when attacking, and finding a draw when defending.

Cato has been responsible for the openings for the last 5 seasons, and lately Jeroen has assisted with the superfinal openings. Cato imposes other restrictions on the openings using his extensive game database, such as minimal number of games played from the position, more than one reasonable continuation, low draw rate, etc. It is reasonable to say that the 'TCEC opening space' is a 'small' subset of the possible starting positions, but how small is it? Has TCEC already covered a substantial part of it and we are likely to see many repeats in the next seasons?

After analyzing the data, my short answers for these questions are 'not so small', and 'no'.

For the analysis I used openings from the following stages:
  • Season 6 stage 3 - 112 games
  • Season 6 stage 4 - 96 games
  • Season 6 superfinal - 64 games
  • Season 7 stage 3 - 112 games
  • Season 7 stage4 - 72 games
  • Season 7 superfinal - 64 games
  • Season 8 stage 3 - 90 games
  • Season 8 superfinal - 100 games
  • Season 9 stage 3 - 224 games
  • Season 9 superfinal - 100 games
  • Season 10 stage2 - 112 games
  • Season 10 superfinal - 100 games
Total = 1246 games in 623 reverse pairs

In the analysis I use the term entropy of a distribution, though the number I calculate is exp(entropy) as it is defined normally. Without going into detail I use this as the effective size of a set of elements with a finite probability distribution. A set of size n with a uniform distribution has an entropy of n, if most of the distribution is concentrated on one element the entropy is close to 1.

When considering ECO codes and opening names you should always keep in mind that the engines playing can influence these, especially when the book sequence is relatively short. For Cato's favorite length of 16 plys the opening is almost always determined by the book. So, mostly Cato's fault as it should be.

ECO, first letter

 

Entropy =  4.92
Close to a uniform distribution, some preference to Sicilians and flank openings, closed games a little under-represented.


ECO, full code

There are 500 different ECO codes. The entropy of the ECO codes in the list is 246.9, covering 312 different codes out of the possible 500. The most frequent ECO codes are
  • B90 Sicilian Najdorf, 20 times
  • B12 Caro-Kann defense, 14 times
  • B80 Sicilian Scheveningen, 14 times
  • C02 French advance, 14 times
  • C45 Scotch, 14 times
ECO codes are usually defined by short opening sequences, much shorter than most TCEC book sequences. The B90 code for example is defined by 10 plys, quite long for an ECO code. In TCEC the 10 game pairs with code B90 had book lengths of 11-21 plys. There were a total of 7 different named variants, one repeated twice and one three times. Only one book sequence was a full book repeat, Season 8 stage 3 games 14 and 29 Gull-Protector, Season 9 stage 3 games 2 and 30 Andscacs-Rybka.

In the under-represented openings we find:
  • Gruenfeld defense (D80-D99), only 7/20 codes represented. 
  • King's Gambit (C30-C39), 4/10 represented.
  • Ruy Lopez (C60-C99), 19/40 represented.

Opening, full name 

Analyzing the full opening names including opening variants, we find there is a much more uniform distribution. The entropy is 573.3, with 618 distinct openings. The most repeated opening variants are:
  • Dutch: 2.c4 Nf6 3.g3 e6 5.Nf3 d5 6.O-O Bd6, 8 times
  • Trompowsky: 2...Ne4 3.Bf4 c5 4.f3 Qa5+ 5.c3 Nf6 6.d5, 8 times
  • Neo-Grֳuenfeld: Alekhine's, 7.Be3 O-O, 6 times
  • Pirc: 4.Be3, 150 Attack, 6 times
  • Sicilian: Kan, Polugaevsky, 6.Nb3 Ba7, 6 times
  • Sicilian: Najdorf, 6.f3, 6 times

Book sequences

If we look at the full book sequences, there are 618 distinct sequences out of 623 game pairs, with entropy 616.3. Only 5 book sequences are repeated in two game pairs.

I truncated the book sequences at fixed lengths to measure the expansion as the length increases. For short lengths I also list the most frequent sequences.

After 2 plys the entropy is only 12.74, a total of 34 sequences for 622 game pairs. The leading sequences are:
  • 1. d4 Nf6, 24.4%
  • 1. e4 c5, 18.2%
  • 1. e4 e5, 10.3%
  • 1. d4 d5, 10.1%
  • 1. e4 e6 , 6.9%
After 4 plys the entropy is 50.0, a total of 133 sequences for 622 game pairs. The leading sequences are:
  • 1. d4 Nf6 2. c4 g6, 8.2%
  • 1. e4 c5 2. Nf3 d6, 8.0%
  • 1. e4 e5 2. Nf3 Nc6, 7.6%
  • 1. d4 Nf6 2. c4 e6 , 7.1%
  • 1. e4 e6 2. d4 d5, 6.1%
After 6 plys the entropy is 132.9, a total of 260 sequences for 620 game pairs. The leading sequences are:
  • 1. e4 c5 2. Nf3 d6 3. d4 cxd4, 6.8%
  • 1. d4 Nf6 2. c4 g6 3. Nc3 Bg7, 4.8% 
  • 1. d4 Nf6 2. c4 e6 3. Nc3 Bb4, 3.4%
  • 1. e4 e5 2. Nf3 Nc6 3. Bb5 a6, 3.4%
  • 1. e4 c5 2. Nf3 e6 3. d4 cxd4, 3.2%
  • 1. e4 c5 2. Nf3 Nc6 3. d4 cxd4, 3.2%
After 8 plys the entropy is 230.2, a total of 360 sequences for 604 game pairs.
After 10 plys the entropy is 328.0, a total of 432 sequences for 587 game pairs.
After 12 plys the entropy is 442.5, a total of 487 sequences for 575 game pairs.
After 16 plys the entropy is 523.4, a total of 530 sequences for 543 game pairs.


Monday, January 1, 2018

Season 11 starts this week

Season 11 is about to start, the target date is Jan 3rd 2018.

You can see the announcement on Chessdom, including the participats list, as well as a more details on the new league format.

The divisions will be played from the lowest to the highest, gradually increasing the time control and the number of games per engine. The premier division will be a huge 12RR event with time control of 90+10 (336 games) and the superfinal with the usual 100 games and time control of 120+10. Each division will have promotions and relegations so I expect a lot of excitement.

This season the engine Fritz will make a first appearance, replacing its predecessor Rybka. Rybka has been on TCEC from season 1, it was very strong to begin with but has not been developed for a long time and is not considered very strong now.

The twitch channel is going to feature GM commentary for the first time, something the viewers always wanted. It should be interesting to see how that works out. Another first for the season is a new '3 crashes and out' rule, engines crashing three times in a division will be automatically relegated.

Here's to a happy new year and a great new TCEC season.

Go TCEC!

Season 10, blitz tournament statistics

Draw rate, wins



Final draw rate was 31%.

Game termination

The three most common game termination causes were:
38.4% - White mates
29.0% - Black mates
18.8% - 3-fold repetition

These are very unusual, due to the fact that there was no adjudication by tablebases or by score in the blitz tournament.

There were only 9 crashes.

Moves per game



Median= 75.5
Average= 80.5

No adjudication caused the games to be longer than usual. There were 55 games (10%) with more than 120 moves.

Time per game (hours)

  

Median= 0:10:04
Average= 0:10:14

Openings

There were 2-move book openings in this stage chosen by Cato. The first letter of the ECO codes was distributed as follows:



Open games (C) and indian openings (E) were under-represented, very similar to the rapid stage.

If we use the opening 'family name' (using format FAMILY_NAME: VARIANT....) the top 3 are:
Sicillian - 105 times
English - 46 times
QGD - 38 times

The engines had a lot of freedom to choose the opening variant. Only 37.3% of the game pairs repeated the same ECO code twice, and only 18.1% repeated the same opening variant. In 18.1% of the game pairs the ECO first letter was not repeated twice.

Reverse pairs, wins



Reverse pairs, same moves  



The reverse games diverged very quickly, 53% diverged in the first move and 87% diverged at or before the third ply. Only 7 games repeated more than 6 plys. One game pair (Gull-Arasan) repeated 26 plys after book before diverging.