Season 12 division 4 final results

Division 4 final crosstable



Ethereal and Xiphos are promoted to division 3. Ethereal did not lose a game in the division and led from the start. In the early rounds Rodent was close to the top of the table, but with 4 losses to Xiphos it finished 3rd.

Scorpio crashed 7 times, mostly on the very first move after book. This is disappointing since its performance was better than it was in season 11 in the games it did finish. According to the 3-strike rule all its games are removed and it is automatically relegated (though there is no relegation in division 4). The formal crosstable taking this into account does not change the standings significantly.

Leela's games (LCZero) attracted a lot of attention, as it is a community effort and a different approach to the search and score algorithms in other engines. The project is trying to use the same learning approach of Google's Alpha zero, using normal hardware and public resources for training the neural nets. In division 4 it had several interesting games but it lost almost all of them (its single win was a Scorpio crash...), usually blundering in the middle and endgame even if it had an advantage. The project is rapidly evolving and there is a good chance it will perform much better in the next season.

Division 3 is about to start, with the engines Bobcat, Wasp, Nemorino, Pedone, ChessBrain and Senpai joining Ethereal and Xiphos. The format will be identical to division 4, 4RR with 30+10 time control. I expect division 3 to be a more exciting event than division 4, with a higher draw rate and closer scores. According to the CCRL rating list the engines are quite close. Ethereal's author appears regularly on the chat and thinks its engine should promote again, in season 11 it finished last in division 3 so that would be a significant improvement.

Season 12 division 4 statistics

Draw rate, wins



Final draw rate was 25.9%. This is unusually low, and in part due to the fact that Leela lost almost all of its games


Game termination

The three most common game termination causes were:
63.4% - TCEC win rule
11.6% - TCEC draw rule
9.8% - TB position

There were 2 games that ended in mate on the board, Leela games in both cases, extreme examples of Leela blunders. There were 7 crashes, all by Scorpio. 

Moves per game

 

Median= 59.8
Average= 66.3

Includes 5 games where Scorpio crashed almost immedately out of book, and one of the longest games in TCEC history with 353 moves (Fruit - Scorpio, round 6, draw).

Time per game (hours)
  

Median= 1:11
Average= 1:12

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.

If we use the opening 'family name' (using format FAMILY_NAME: VARIANT....) the top 3 are:
Sicillian - 24 times
QGD - 20 times
English - 13 times

The engines had a lot of freedom to choose the opening variant. Only 25% of the game pairs repeated the same ECO code twice, and only 5.4% repeated the same opening variant. In 19.6% of the game pairs the ECO first letter was not repeated twice.

Reverse pairs, wins

 

Reverse pairs, same moves 

 

Pairs of reverse games diverged very quickly, 82.1% of the pairs diverged at most after 1 move. One game pair repeated 7 moves after book (Tucano - Xiphos, rounds 25 and 22, Grünfeld: Modern Exchange).

 

Season 12 live blogging, Friday Apr 20th

Season 12 has started with division 4, a 4RR with eight engines, the top two will qualify for division 3. The games are about an hour long each and we are about half way through the second RR. Ethereal is leading, with Rodent and Xiphos close behind, it is likely that 2 of these three engines will promote.

A lot of interest in this division is the new NN engine Leela chess zero (LCZero). It entrered the competition at the last moment and has a lot of fans, many people are donating their computing resources to help train Leela. So far it is not doing well, usually it blunders in teh middle or end game. However it is always interesting to watch. There is a good chance that next TCEC season it will do much better.

The Leela - Tucano game is just starting.

move 6 - evals are close to 0, Tocano's just turned slightly negative. This is not a very popular line for QGD, but there are some GM games with this line, so it is playable.

move 10 - Leela is pushing pawns on the queen side, its king is still in the center. The bishop on b2 has control of the long diagonal. Tucano is ahead on developement.

move 14 - Leela opened the queen side, it should get its king to safety soon.Tucano's eval is increasing gradually.

move 21 - Tucano is a pawn up after a series of exchanges, eval over 1. Leela is finally ready to castle, but it will have to defend to hold the game. Will the black pawn majority on the queen side be enough for a win? Will Leela blunder?

move 24 - The engines exchanged a pair of rooks and only QRB vs QRN remain. Leela usually blunders the endgame, and this game is heading to a B vs N endgame with black a pawn up.

move 27 - Tucano has a rook on the 2nd rank, its eval is over 1.5. It's too bad it has no PV and we can't see what it is planning.

move 31 - Tucano's eval continues to increase, over 2 now. The remaining rooks are off the board and Leela may lose another pawn soon.

move 35 - I don't see how Leela can stop the queen side pawns in the long run. It's trying to counter on the king side but the queen alone cannot do a lot.

move 38 - there is a black pawn on d3, it will cost Leela the bishop. Can it do something with its queen and pawns on the king side? Leela's eval is still below 1.5.

move 44 - Leela is still optimistic but Tucano's eval is over 6 and it is winning the Q vs QN ending.

move 55 - soon Leela will run out of checks, it is a knight and pawn down and it's just a matter of time now.

I don't think Leela blundered badly in the game, perhaps it gave away a pawn too easily. Once it lost the second pawn and Tucano started to march on the d file Leela lost a piece and the game.