TCEC - Division 3 - Season 1 - Live! Discuss.

[Martin Thoresen]

Standings after 6 of 14 rounds:

Note I: After the final standings, the top two finishers will be
promoted to Division II for the next Season, while the bottom
two finishers will the out of TCEC for the next Season.

Note II: TCEC is using the 3-1-0 scoring system.

Note III: All games can be downloaded from the official site at http://www.tcec-chess.org/division_3.php .

Code: Select all

   Nr  Engine                             Pts       SB     J H T A G D U S
   1 Jonny 4.0                           12.0     31.5000  - = - = 1 1 1 =
   2 Hannibal 1.0a                       10.0     31.0000  = - 1 - 1 = = =
   3 Tornado 4.4                          9.0     21.0000  - 0 - 1 0 1 1 0
   4 Arasan 12.2                          8.0     21.0000  = - 0 - 1 0 = 1
   5 Gaviota 0.8                          7.0     19.0000  0 0 1 0 - 1 = -
   6 Daydreamer 1.75                      7.0     18.0000  0 = 0 1 0 - - 1
   7 Umko 1.1                             6.0     17.5000  0 = 0 = = - - 1
   8 Scorpio 2.6                          5.0     20.0000  = = 1 0 - 0 0 -

[Martin Thoresen]

Final standings of Division III:

Code: Select all

Nr Engine                               Pts      SB      Ha Jo To Sc Da Ga Ar Um
   1 Hannibal 1.0a                       30.0    181.0000  -- =1 1= == =1 11 11 =1
   2 Jonny 4.0                           28.0    154.0000  =0 -- 1= =1 10 11 =1 11
   3 Tornado 4.4                         21.0    112.0000  0= 0= -- 0= 10 01 11 11
   4 Scorpio 2.6                         18.0    125.0000  == =0 1= -- 0= 11 0= 01
   5 Daydreamer 1.75                     18.0    124.5000  =0 01 01 1= -- 0= 1= ==
   6 Gaviota 0.8                         12.0     70.0000  00 00 10 00 1= -- 0= =1
   7 Arasan 12.2                         11.0     73.5000  00 =0 00 1= 0= 1= -- =0
   8 Umko 1.1                            11.0     73.5000  =0 00 00 10 == =0 =1 --

Congratulations to Hannibal & Jonny for the promotion into next Season Division II.

Arasan and Umko are out of the TCEC system for the next Season.

Division II will start in two days.

Thank you everyone for following!

[BB+]

I prefer matches with ponder off, it's on the same system, because I'm always thinking that one engine could "cheat", and jump itself up to higher priority, or use more cpu time that it's opponent, by "working around" the ponder off settings.

I recall the old days, when engines used to do such dirty tricks as continually query the interface, to help prevent the opponent's engine, from getting it's full share of cpu cycles.

Another reason to use "ponder off" when on one computer is that most likely you only have one hard drive (or at least only one OS copying to/from it), and so the use of tablebases when not on-move could become an issue.

[BB+]

Another thing: most of the engines use all the time available in a "movestogo" format, so under the current scheme it's really 28 moves in 100 minutes (~3.5 min/move), followed by 20 moves in 50 minutes (2.5 min/move), and then "increment chess" (20 mins + 10 sec/move). I haven't seen too many interesting endgames past the move-60 mark so far, but it occasionally might be a little rushed compared to the earlier time controls (I know that FIDE or whoever does similar, though part of that is due to the here-irrelevant fatigue factor -- the other part is that watching a 150+ move drawn endgame drag on for some extra hours is a bit tedisome).

[BB+]

Here is a list of Division 3 games and what they looked like at the "move 60" mile-marker.

Arasan-Gaviota (Round 1, Division 3) might be an example of a game "degraded" by increment chess (though I'm not sure either side would have played this phase very well in any event). OTOH, Umko-Gaviota in Round 2 is an example of a ~200-move marathon that was essentially drawn for at least half of those. Round 3 had an semi-interesting imbalance in Arasan-Jonny, though I think it was clear with so few pans that neither side had much of a chance. Round 5 with Umko-Jonny is one that went from drawish at move 60 to a win about 15 moves later, though as Round 6 Tornado-Umko shows, a blunder can happen right at the time control too (59... g5? loses almost immediately). Scorpio-Hannibal in the same round also had at least some play after move 60, as did Scorpio-Gaviota in Round 7. Onto Round 8, where Tornado-Daydreamer was decided after move 60. Gaviota-Arasan had White a clear piece up at move 60, but with an airy King was unable to avoid a draw over the next 50 moves. Umko-Hannibal had a large Black advantage at move 60, and the win was rather easy. Similarly with Jonny-Daydreamer in the next round. Scorpio-Arasan was a unbalanced draw (N+P vs 3P) at move 60, and neither side erred for 80 moves after that. Tornado-Hannibal was pretty clearly drawn at move 60, and 30 moves reduced it to being a bit more obvious. In Round 10, Daydreamer-Hannibal still had some play at move 60, though Black's advantage is likely enough. Scorpio-Umko still had most of the pieces left at move 60, though Black was in dire straits and quickly succumbed. Round 11, Jonny-Hannibal had a blocked board with a slight White advantage at move 60, and Black won a rook endgame 70 moves later. In Arasan-Umko, White was probably simply lost at move 60, but managed to wriggle for another 15 moves. Daydreamer-Gaviota was quite interesting at move 60, and I'm kinda surprised White did not win (I was getting +8 and more for 69. Bh5!). Scorpio-Tornado again had play left at move 60, with Black having two pawns on the seventh rank in return for an exchange (with Queens on the board, a perpetual ensued as expected). In Round 13, Umko-Tornado had a large Black advantage at move 70, and I'm not quite sure why it took so long to convert. Hannibal-Scorpio had a blocked up position with White a pawn-up, and all to play for (it ended a draw in a Queen endgame). Jonny-Tornado in Round 14 was almost completely blocked at move 60, and after 70+ pointless moves it was a draw. Umko-Daydreamer had a RBN+4 vs RR+6 imbalance at move 60, and after various vicissitudes it drawn 70 moves later.

Again, I'm just enumerating evidence to try to help you tweak the system one way or the other.

[BB+]

Here is a list of Division 2 games and what they looked like at the "move 60" mile-marker.

Onto Division 2. In Round 1, Junior-Sjeng had White an insignificant pawn-up (RB+2 vs RB+1), and 60 moves later the result was unchanged. Gull-Crafty had a blocked board with no real advantage, and the 50-move rule hit about 55 moves later. In Onno-Crafty, Black was down a pawn at move 60, and failed in the struggle to hang on. Round 3 had Gulf-Protector with a double rook endgame where White had a nominal edge, and it ended in a draw rather quickly. Round 4 saw Sjeng-Komodo with a dead-drawn rook endgame at move 60, and after 60 moves of nothing it was agreed. Round 5 had Sjeng-Gull in a drawn RP vs BPP endgame that took 50+ more moves to ensure this (much of it in 5-piece bases). Crafty-Protector saw Black a pawn-up at move 60 in Round 6, but there was no real advantage, and it was drawn in another 20 moves. Sjeng-Onno had QP vs QP (both pawns on the seventh rank) on move 60. Thirty moves later the draw was obvious. Komodo-Spark had White a useless pawn up in an opposite Bishops ending - drawn at move 114. Junior-Gull had a blocked pawn structure and Black with a Knight for a (good) Bishop, and 100 moves later the draw was agreed. Round 7 saw Sjeng-Protector with a Knight for a Bishop, with a pawn on a7. Black had the play though (the a7 pawn dropped), but then after 10 moves in a drawn 5-piece NP vs B, Black simply blundered (to a one-move Knight fork) and lost. Komodo-Crafty had a slight White advantage (Knight for bad Bishop), but this was not enough to win (drawn at move 129, after 50 moves of the 5-piece NP vs B). Gull-Spark had 3 pawns per side in a Rook endgame at move 60, and it was resolved as a draw in about 25 moves.

In Round 8, Black was two pawns up at move 60 in Protector-Komodo, but it was possible for it to become delicate.(there were opposite bishops, and both sides had a queen). White made matters easier by pushing his pawn too much, and lost rather quickly. Sjeng-Junior was a double bishop endgame with a blocked pawn structure. It looked drawish at move 60 and indeed at move 81 the 50-move rule ended it. In Round 9, Crafty-Onno saw each side with a Rook and 3 pawns at move 60, but it was pretty clearly drawn (at move 73). In Round 10, Sjeng-Crafty had White with QPPP vs QPP, but it was drawn, and a perpetual concluded on move 77. In Round 11, Komodo-Sjeng had a symmetrical queen and rook endgames (both sides with 5 pawns), and a draw was agreed 20 moves later. In Round 12, Spark-Crafty saw White with a minor edge in RRN+6 vs RRB+6. By move 120, both sides were down to 4 pawns apiece, and still unclear concerning White's chances of winning. However, Spark gained a winning advantage over the next 10 moves and won on move 149. In Round 13, Onno-Sjeng saw White with a pawn (and the bishop pair) for the exchange (RBB+3 vs RRB+2), with the position essentially equal. The draw was agreed on move 109. Finally, in Round 14, Protector-Sjeng had White with two Knights and two pawns (doubled passed pawns on a6 and a7!) for a Rook and Bishop (both sides having a Queen), and +1.5 or more in evaluation. The game ended in a perpetual on move 97. Crafty-Komodo had Black with QBP vs QPP (and theoretically won according to bitbases) at move 60, and the conversion took 15 more moves.

[Martin Thoresen]

BB+,

Thanks for your little analysis of the endgames there.

In the "old" TCEC, the format was at the 60th move, 15 minutes + 30 seconds per move. But I decided that some of those games were, as you said, a bit tedious to follow when they dragged on forever. So I tried to compensate somehow by increasing the base number of minutes from 15 to 20.

If you would draw a conclusion of the current time control after the 60th move, what would it be? In regards to what you wrote above from Division 2 & 3.

Best,
Martin

[Robert Houdart]

From the engine's point of view, the abrupt change of pace from 2 min/move to 20 sec/move at move 61 is completely unexpected. The engine has no idea that this is coming, and it will often have used up most of its time at move 60.

I would suggest just getting rid of the TC at 60 moves, give the engines 60 minutes + 20 sec/move from move 41 on. This will allow for a more intelligent use of the time.

Robert

[Martin Thoresen]

From the engine's point of view, the abrupt change of pace from 2 min/move to 20 sec/move at move 61 is completely unexpected. The engine has no idea that this is coming, and it will often have used up most of its time at move 60.

I would suggest just getting rid of the TC at 60 moves, give the engines 60 minutes + 20 sec/move from move 41 on. This will allow for a more intelligent use of the time.

Robert

Robert, this is interesting. Are you saying that most engines don't understand 3 levels of TC?

Best,
Martin

[Robert Houdart]

Robert, this is interesting. Are you saying that most engines don't understand 3 levels of TC?

An engine only gets information about the current TC, and has no idea whatsoever about what is coming next. Every change of pace is completely unexpected.
For the TCEC, it's only at move 61 that the engine finds out that it now has 20 min + 10 sec/move left.