Page 2 of 3
Re: Komodo benefits less from TBs?
Posted: Sun Oct 08, 2017 5:54 pm
by mbabigian
An alternative explanation is simply that Komodo has more endgame knowledge. This would give TB's less to improve on. The data already suggests this as komodo was best with no Tablebases.
My two cents.
Re: Komodo benefits less from TBs?
Posted: Sun Oct 08, 2017 5:58 pm
by Guenther
mbabigian wrote:An alternative explanation is simply that Komodo has more endgame knowledge. This would give TB's less to improve on. The data already suggests this as komodo was best with no Tablebases.
My two cents.
This explanation is of course a logical hypothesis and therefore already was mentioned in this thread by Evert.
Re: Komodo benefits less from TBs?
Posted: Sun Oct 08, 2017 7:22 pm
by Modern Times
Guenther wrote:mbabigian wrote:An alternative explanation is simply that Komodo has more endgame knowledge. This would give TB's less to improve on. The data already suggests this as komodo was best with no Tablebases.
My two cents.
This explanation is of course a logical hypothesis and therefore already was mentioned in this thread by Evert.
Yes - and it is the most likely explanation I think.
Re: Komodo benefits less from TBs?
Posted: Mon Oct 09, 2017 2:37 pm
by Laskos
Modern Times wrote:Guenther wrote:mbabigian wrote:An alternative explanation is simply that Komodo has more endgame knowledge. This would give TB's less to improve on. The data already suggests this as komodo was best with no Tablebases.
My two cents.
This explanation is of course a logical hypothesis and therefore already was mentioned in this thread by Evert.
Yes - and it is the most likely explanation I think.
It seems so to me too, but one result is not in line with this explanation. In a gauntlet from a suite of 6-men Wins, against the perfect player on 6-men, BrainFish enabled with 6-men Syzygy, Komodo is not that successful without TBs.
Code: Select all
# PLAYER : RATING ERROR POINTS PLAYED (%) CFS(next)
1 BrainFish Syzygy-6 : 172.6 12.9 1745.5 3000 58.2 100
2 Houdini_602 NO TB : -44.8 20.1 424.0 1000 42.4 84
3 BrainFish NO TB : -62.3 20.3 416.0 1000 41.6 57
4 Komodo 11.2 NO TB : -65.5 20.4 414.5 1000 41.5 ---
The correct pentanomial error margins are about 4 times smaller than shown in Ordo, or 5 ELO points, because the positions are very unbalanced, they are theoretical 6-men Wins.
So, Komodo without TBs doesn't convert more 6-men Wins than the other two engines. And the explanation would be more convoluted: Komodo plays better 7-8-9 men or 14-men endgames without TBs than other engines without TBs, but on 6-men it plays no better. A somehow contorted explanation.
Re: Komodo benefits less from TBs?
Posted: Mon Oct 09, 2017 3:35 pm
by Jouni
Houdini's 1. position is surprise. But obviously Houdart has improved syzygy implemention as my tests have also shown. Version 5 was buggy.
Re: Komodo benefits less from TBs?
Posted: Mon Oct 09, 2017 10:00 pm
by Damir
Kai,
Can you make a test of which engine benefits most whitout use of tbs ?
Re: Komodo benefits less from TBs?
Posted: Tue Oct 10, 2017 11:05 pm
by mjlef
Laskos wrote:Evert wrote:It is possible that Komodo has end-game knowledge that helps it in case there are no tablebases. If that is the case, then there is less to gain from tablebases.
This is just speculation, of course, but I wouldn't be surprised if Komodo has such knowledge.
I also thought of that possibility. Amongst these 3 top engines, at this time control, Komodo is significantly weaker than the other two in normal games with no TBs. But here, in endgames without TBs, it comes as the strongest, especially pronouncedly in the first endgame suite. In other words, without TBs it overperforms in endgames compared to general play.
So, yes, this speculation seems to me a possibility.
Komodo does seem to have some more general endgame knowledge than other programs I have looked at. And the endgame has some modifications to the search as well.
Re: Komodo benefits less from TBs?
Posted: Thu Oct 12, 2017 2:21 am
by Tdunbug
I am currently running a 10K head to head match between default Komodo verses a tweaked variation of Komodo( I made it so it does not hit TBS so quickly). I want to see if Komodo would benefit from relying on it's own calculations more during the middle stages of the games.
Though it is still early I have noticed a couple of things:
#1 Komodo 11.2.2 TB seems to be roughly 2% faster than the default
#2 While Komodo 11.2.2 TBS is faster during the beginning and middle phases of the game it is slightly slower during the endgame phase.
So there seems to be a symbiotic relationship here
Re: Komodo benefits less from TBs?
Posted: Thu Oct 12, 2017 4:35 am
by shrapnel
I want to see if Komodo would benefit from relying on it's own calculations more during the middle stages of the games.
Good ! It does, actually, but I have no Proof, as its only an Observation of mine while watching online Engine-Engine Matches. Unfortunately, its not consistent enough, but perhaps that's because current version of Komodo is inherently weak.
But why are you using
speed as a parameter/benchmark ? I don't think that's a good idea and has little to do with Komodo's unique approach to the Middle Game.
Re: Komodo benefits less from TBs?
Posted: Thu Oct 12, 2017 8:54 am
by tpoppins
mjlef wrote:Komodo does seem to have some more general endgame knowledge than other programs I have looked at. And the endgame has some modifications to the search as well.
How do you account for it not seeing a draw in the following position, then?
[d]5b1k/7P/6P1/1B6/8/5K2/8/8 b - - 0 58
Taken from
the LTC match in progress vs. Houdini 6.02, this position was reached after White's (Komodo) move 58 in game 48*. It's an obvious 6-men draw - confirmed with the Let's Check DB to be on the safe side.
Both engines utilized 5-men Syzygy with Probe Depth = 1 which, coupled with the generous increment (30s) would have one expect that the game would end in peace quickly. Houdini, in fact, reported 0.00 as early as move 50; Komodo's eval, in contrast, never dipped below +1.50 for another twenty moves and it happily kept shuffling the pieces this way and that way until I had to interfere to end the silliness.
There are plenty of draws in that match, and most of them are examples of Komodo lagging way behind (typically 10-12 moves) Houdini in acknowledging the draw. The few times Komodo does see 0.00 earlier than Houdini is, paradoxically, in early-to-mid middlegame, when you would think there's still everything to fight for.
Playing for a win in such obviously drawn positions may be a valid strategy against much weaker engines - those with inferior search and/or without TBs - that's what Contempt is for, isn't it? It's an odd behavior when Contempt is set to zero, however; or perhaps it points to a significant deficiency in Komodo's endgame knowledge, in particular in opposite-colored bishops endings (game 2 is another example).
* - complete game scores and detailed match conditions in the linked thread