Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / Modern Chess Engines not Optimized for Correspondence
1 2 3 Previous Next  
- - By Bob Durrett (***) Date 2012-08-07 17:32
Chess engines not optimized for correspondence chess (?!)

The testing results I've noticed here seems for much faster games, so there may not be a whole lot of test data for correspondence time limits.

Based SOLELY on intuition, it seems to me that a chess engine optimized for correspondence chess would have a different search strategy. For example, I suspect that the tree would be much "fuller."

If this is true, then there is a lot of room for improvement if one wishes to create an engine optimized for correspondence chess applications.

Does anybody else see this the same way I do?

Bob Durrett
Parent - - By Uly (Gold) Date 2012-08-08 08:27
Yes, agreed.

I think the main difference between corr and faster time controls, is that when the game is quick, it suffices to play moves that don't lose, and to just pick a move and make sure it's "fine". Actually, engines may not consider another move unless this main move has a fail low. This causes very heavy pruning on alternative moves, and the exploding on branching factor may means that the move picked by the engine after 2 hours doesn't change after another 22 hours, which is wasted time! So indeed its strategy isn't good for corr time controls.

In corr games the move you make is critical, you have to have a reason for playing it, and there's no such thing as two moves tied at 0.22 score and choosing one of them by luck: in corr chess one of the moves will be clearly better because it was the plan to go with it since this position was just a possibility in the distance.

However, this is a plus, because if engines didn't have these flaws, had a better strategy for corr time controls, and found the best move without the user having to guide the engines, what would be the point of correspondence chess? It may as well kill the hobby if unattended engines were as strong as centaurs there, at least, if a strategy was implemented where the engines were able to recognize deep losing moves and avoid them, winning games would be much harder than it is now, and it's pretty hard already!

So from the perspective of the hobby player, I'm happy with the status quo, and can only assume top corr players wouldn't want "lazy players" (those that just leave the engine to analyze the position for a long time and play their move) or "engine monkeys" (those that have automatic interaction of positions and don't really have input of their own, so it's mainly the engine playing) to be as good as them just because the engines were better.
Parent - - By Bob Durrett (***) Date 2012-08-08 09:17

> However, this is a plus, because if engines didn't have these flaws, had a better strategy for corr time controls, and found the best move without the user having to guide the engines, what would be the point of correspondence chess? It may as well kill the hobby if unattended engines were as strong as centaurs there, at least, if a strategy was implemented where the engines were able to recognize deep losing moves and avoid them, winning games would be much harder than it is now, and it's pretty hard already!


Again, relying on intuition alone, I suspect that any game in which chess engines are involved would be played "at a higher level" with a better chess engine as compared to doing the same with a lesser engine.

I have not see people on this forum mindlessly playing the moves that the engines told them to play and, IMHO, this should not happen in the future just because the engines get better.

Chess engines are like matches. Matches can be used to start fires which burn buildings down. Knives can be used to kill people. Tools of all kinds can be misused. But the carpenter will not throw away his hammer because hammers can be used to commit crimes.

In the same way, a chess engine can be a useful tool for good or can be used for bad. It is the choice of the user as to which.

I see no moral problem with using tools intelligently. Especially desireable is finding new innovative ways to use the new and improved tools. Innovation and creativity used for "good" gets five stars in my book.

Bob Durrett
Parent - - By Uly (Gold) Date 2012-08-08 10:11

> I have not see people on this forum mindlessly playing the moves that the engines told them to play


Well, I've had games in where it seemed my opponents were just following Rybka "blindly" and I could get advantage of it by playing into positions Rybka didn't understand, and winning the games. Imagine if my opponents have interacted instead, got their homework done, and player much stronger: I couldn't have beaten them, or at least, I wouldn't have done it in the way I did.

Now, imagine if they played as strong as the latter case, by just following the engine "blindly", because its search and evaluation strategy were optimized for correspondence chess. Where would that leave me? Uncertain of if I'm helping the engine or not or if it would play just as strong I I just sit doing nothing and followed it blindly.

This already happens at fast time controls, say, I think that at 30 minutes/move I may play weaker than an unassisted engine, because I don't have enough time to think up my plans, but at some point the flaws of the engines become apparent and I go beyond their level.

Corr players that just follows the engine already happen and it's the only explanation for a player with lesser hardware to win games even when the person in faster hardware is taking longer to move. I'm also going to be venturous and say the main difference between top corr players and average corr players is that the former are better at punishing "engine slaves" and beating them, achieving a higher rating and win percentage even though their results are more or less equal against average corr players.

Fixing the engines' flaws only favor the weaker players.

>Knives can be used to kill people. Tools of all kinds can be misused. But the carpenter will not throw away his hammer because hammers can be used to commit crimes.


There have been machines constructed that have left humans without a job, because instead of having to hire 6 people to do some job, they build one machine that does the same which is way cheaper. Getting engines to decent corr chess status unattended (which I think is possible) could potentially leave the person without a reason to keep playing corr chess, because after so much effort not being able to defeat an "engine slave" would be disheartening (as the "engine slave" isn't putting much effort into the game in first place).

> Innovation and creativity used for "good" gets five stars in my book.


Getting the engines to play as good as very creative corr players by just leaving them analyze the same position for several hours would just lead to the creative players to quit the game and use their creativity somewhere else.
Parent - By Bob Durrett (***) Date 2012-08-08 14:09

> Fixing the engines' flaws only favor the weaker players.


Perhaps.

But, could it be that "fixing the engines' flaws" would force you to raise the level of your play? Instead of playing at the IM level in "the Rybka Forum variant of Correspondence Chess" you might have to play at the GM level. That would be good!

You would then be facing a greater challenge, but surely you could rise to the occasion.

> There have been machines constructed that have left humans without a job, because instead of having to hire 6 people to do some job, they build one machine that does the same which is way cheaper.


From a societal point of view, the real issue is whether or not technology is raising the quality of life for the entire society, or not. It is true that there will always be short-term negative consequences. Consider an example: "Awhile back," Thomas Edison introduced the electric light bulb. Soon, towns and cities (where I live) started appearing with electric street lights. This put some people out of work. Previously, street lights had been gas and had to be lighted manually.  The manual gas street light lighters were put out of work. I'm sure they cussed Thomas Edison. But who would say, in all seriousness, that the electric light bulb technology was bad?
Parent - - By Uri Blass (*****) Date 2012-08-09 04:09
I disagree with you.

I am a corr player who mainly follow the engines and I believe that the main advantage of the strong corr players is simply that they use better hardware or better software and not playing moves not suggested by the engine.

I replace micheal sorin in a tournament

http://ratings.fide.com/card.phtml?event=2805030

Micheal sorin is a player with fide rating of 2259 and he is better player than me but unfortunately it seems that he did not use a computer and after getting bad positions in half of his games decided to quit the tournament

Here is the cross table and you can see my finished games.
I hope to get more than 50% inspite of the fact that I started with bad positions in half of my games

http://www.iccf-webchess.com/EventCrossTable.aspx?id=26398

Here is one of my games that I started after 26.Kh2 with black and saved a draw.
It is clear that my opponent did not use a special sofware in KRPP vs KRP and missed a win but even without a special software
I find that Stockfish can find the right tablebases moves after a deep search.

Note that it is possible that my 33...Ka6 is a mistake but I did not analyze it to be sure if I can or cannot save the game with a different move.

[Event "EU/TC9/sf3"]
[Site "ICCF"]
[Date "2011.7.15"]
[Round "-"]
[White "Mourato, Mבrio"]
[Black "Blass, Uri"]
[Result "1/2-1/2"]
[WhiteElo "2322"]
[BlackElo "2603"]
[Board "6"]
[WhiteTeam "Portugal"]
[BlackTeam "Israel"]

1.d4 {[%ccsnt 2011.07.11,01:42]} d5 {[%ccsnt 2011.07.11,08:10]}
2.Nf3 {[%ccsnt 2011.07.13,20:42]} Nc6 {[%ccsnt 2011.07.13,21:18]}
3.g3 {[%ccsnt 2011.07.13,23:14]} Bg4 {[%ccsnt 2011.07.14,05:53]}
4.Bg2 {[%ccsnt 2011.07.15,01:26]} Qd7 {[%ccsnt 2011.07.15,10:28]}
5.O-O {[%ccsnt 2011.07.16,21:05]} O-O-O {[%ccsnt 2011.07.16,21:55]}
6.Ne5 {[%ccsnt 2011.07.17,17:04]} Nxe5 {[%ccsnt 2011.07.17,22:29]}
7.dxe5 {[%ccsnt 2011.07.18,22:24]} Bh3 {[%ccsnt 2011.07.18,23:05]}
8.Qd4 {[%ccsnt 2011.07.19,23:01]} Bxg2 {[%ccsnt 2011.07.19,23:14]}
9.Kxg2 {[%ccsnt 2011.07.20,23:05]} Kb8 {[%ccsnt 2011.07.21,06:06]}
10.Be3 {[%ccsnt 2011.07.24,03:23]} b6 {[%ccsnt 2011.07.24,07:57]}
11.a4 {[%ccsnt 2011.07.25,02:12]} e6 {[%ccsnt 2011.07.25,06:04]}
12.a5 {[%ccsnt 2011.07.26,02:06]} Bc5 {[%ccsnt 2011.07.26,06:04]}
13.Qd3 {[%ccsnt 2011.07.27,02:22]} Bxe3 {[%ccsnt 2011.07.27,19:15]}
14.Qxe3 {[%ccsnt 2011.07.28,10:12]} d4 {[%ccsnt 2011.07.29,09:15]}
15.Qa3 {[%ccsnt 2011.08.06,23:39]} Ne7 {[%ccsnt 2011.08.07,20:54]}
16.Nd2 {[%ccsnt 2011.08.09,20:44]} h5 {[%ccsnt 2011.08.09,21:59]}
17.Nf3 {[%ccsnt 2011.08.11,21:31]} Nc6 {[%ccsnt 2011.08.11,23:21]}
18.Rfd1 {[%ccsnt 2011.08.14,19:29]} Qd5 {[%ccsnt 2011.08.16,07:21]}
19.axb6 {[%ccsnt 2011.08.18,02:27]} cxb6 {[%ccsnt 2011.08.18,07:22]}
20.c3 {[%ccsnt 2011.08.19,02:35]} Rd7 {[%ccsnt 2011.08.19,16:45]}
21.h4 {[%ccsnt 2011.08.20,13:38]} Qe4 {[%ccsnt 2011.08.22,22:53]}
22.Rd2 {[%ccsnt 2011.08.24,22:18]} Rhd8 {[%ccsnt 2011.08.25,23:11]}
23.cxd4 {[%ccsnt 2011.08.26,23:02]} a5 {[%ccsnt 2011.08.28,19:16]}
24.Qb3 {[%ccsnt 2011.09.04,15:34]} Kb7 {[%ccsnt 2011.09.07,17:07]}
25.Ra4 {[%ccsnt 2011.09.08,12:34]} Nb4 {[%ccsnt 2011.09.12,22:14]}
26.Kh2 {[%ccsnt 2011.09.14,18:59]} Qf5 {[%ccsnt 2011.10.12,15:33]}
27.Qd1 {[%ccsnt 2011.10.16,14:46]} Rc8 {[%ccsnt 2011.11.02,10:50]}
28.e3 {[%ccsnt 2011.11.06,03:25]} Rc2 {[%ccsnt 2011.11.06,15:25]}
29.Rxc2 {[%ccsnt 2011.11.07,15:20]} Qxc2 {[%ccsnt 2011.11.07,15:27]}
30.Qxc2 {[%ccsnt 2011.11.09,11:14]} Nxc2 {[%ccsnt 2011.11.09,11:52]}
31.Ng5 {[%ccsnt 2011.11.19,01:40]} Nb4 {[%ccsnt 2011.12.08,09:54]}
32.Ra3 {[%ccsnt 2011.12.11,03:30]} Rc7 {[%ccsnt 2011.12.22,07:36]}
33.Kh3 {[%ccsnt 2011.12.24,01:30]} Ka6 {[%ccsnt 2012.01.07,20:48]}
34.g4 {[%ccsnt 2012.01.09,16:22]} hxg4+ {[%ccsnt 2012.02.16,04:44]}
35.Kxg4 {[%ccsnt 2012.02.17,01:42]} b5 {[%ccsnt 2012.02.21,10:59]}
36.f4 {[%ccsnt 2012.02.25,03:30]} a4 {[%ccsnt 2012.03.13,11:57]}
37.f5 {[%ccsnt 2012.03.15,11:08]} exf5+ {[%ccsnt 2012.03.15,19:12]}
38.Kxf5 {[%ccsnt 2012.03.16,16:27]} Nc2 {[%ccsnt 2012.04.15,10:23]}
39.Rd3 {[%ccsnt 2012.04.17,02:46]} b4 {[%ccsnt 2012.04.17,03:55]}
40.d5 {[%ccsnt 2012.04.18,03:00]} a3 {[%ccsnt 2012.04.18,16:04]}
41.d6 {[%ccsnt 2012.04.19,15:27]} Rb7 {[%ccsnt 2012.04.20,04:01]}
42.Ke4 {[%ccsnt 2012.04.21,00:49]} b3 {[%ccsnt 2012.04.22,13:53]}
43.bxa3 {[%ccsnt 2012.04.23,11:08]} b2 {[%ccsnt 2012.04.23,23:25]}
44.Rd1 {[%ccsnt 2012.04.24,22:52]} Nxa3 {[%ccsnt 2012.04.24,23:17]}
45.Nxf7 {[%ccsnt 2012.04.25,22:40]} Nb5 {[%ccsnt 2012.04.26,15:06]}
46.Kd3 {[%ccsnt 2012.04.27,10:57]} Rxf7 {[%ccsnt 2012.04.28,11:51]}
47.Rb1 {[%ccsnt 2012.04.29,03:24]} Nxd6 {[%ccsnt 2012.05.06,12:00]}
48.exd6 {[%ccsnt 2012.05.07,11:19]} Kb6 {[%ccsnt 2012.05.08,10:00]}
49.Rxb2+ {[%ccsnt 2012.05.12,02:28]} Kc5 {[%ccsnt 2012.05.12,07:32]}
50.Ke4 {[%ccsnt 2012.05.17,02:09]} Kxd6 {[%ccsnt 2012.05.17,04:11]}
51.Rb6+ {[%ccsnt 2012.05.19,03:05]} Kc7 {[%ccsnt 2012.05.19,17:17]}
52.Ra6 {[%ccsnt 2012.05.22,15:19]} Kb7 {[%ccsnt 2012.05.22,15:29]}
53.Rg6 {[%ccsnt 2012.05.24,15:06]} Kc8 {[%ccsnt 2012.05.25,12:21]}
54.h5 {[%ccsnt 2012.05.30,02:26]} Kd8 {[%ccsnt 2012.05.30,09:29]}
55.Re6 {[%ccsnt 2012.06.04,12:03]} Rf1 {[%ccsnt 2012.06.04,15:33]}
56.Re5 {[%ccsnt 2012.06.14,15:24]} Rf6 {[%ccsnt 2012.06.17,11:31]}
57.Rg5 {[%ccsnt 2012.06.22,00:46]} Rh6 {[%ccsnt 2012.06.22,01:44]}
58.Kf5 {[%ccsnt 2012.06.29,01:06]} Ke8 {[%ccsnt 2012.06.29,08:13]}
59.e4 {[%ccsnt 2012.07.02,02:28]} Kf7 {[%ccsnt 2012.07.02,17:09]}
60.e5 {[%ccsnt 2012.07.17,16:26]} Ra6 {[%ccsnt 2012.07.17,16:54]}
61.Rg4 {[%ccsnt 2012.07.18,02:16]} Ra1 {[%ccsnt 2012.07.18,10:41]}
62.Rb4 {[%ccsnt 2012.07.19,02:35]} Rf1+ {[%ccsnt 2012.07.19,07:39]}
1/2-1/2
Parent - - By Uri Blass (*****) Date 2012-08-09 04:23
Note that based on my memory houdini after a long search suggested 33...Ka6 and did not expect my opponent move 34.g4
and if it was a mistake it seems that the mistake was using the wrong software for the relevant position
and I found later that stockfish is clearly faster in failing low after g4 relative to houdini.
Parent - By Uly (Gold) Date 2012-08-09 04:38

> and if it was a mistake it seems that the mistake was using the wrong software for the relevant position
> and I found later that stockfish is clearly faster in failing low after g4 relative to houdini.


Knowing what software to use is part of the interaction I'm talking about, see below.
Parent - - By Uly (Gold) Date 2012-08-09 04:33

> it seems that he did not use a computer


> my opponent did not use a special sofware in KRPP vs KRP and missed a win


You examples only show poor opposition. I can agree "engine monkeys" may be as effective as "engine interactors" against opposition like the one you've shown, and that may be the reason "engine monkeying" continues, but success of "engine monkeys" against other "engine monkeys" on lesser hardware or worse software, or against people that don't use engines, doesn't say anything against performance of people that put more effort in their games.

Also, define "mainly following the engines", is it just putting the engine to analyze the root position for several hours and then playing the move it suggests? Or you're doing something else like investigating the variations (forcing the main move and seeing if the eval changes) or checking with a second engine, etc.? Because even minimal interaction like putting the engine in MultiPV and judging if playing the top move is better or not, or deciding at what point the top move is excluded to see if there's a better one that needs to be extended can make a difference.
Parent - - By Uri Blass (*****) Date 2012-08-09 18:02
I do something else
in part of the cases I use multi-pv and in part of the cases I choose the engine to use but basically I use infinite analysis
and cases when I play a move that is not the best move based on evaluation of some top engine after a deep search are rare and I think that I do not expect them to happen in a big majority of my games.
Parent - - By Banned for Life (Gold) Date 2012-08-09 18:25
And what do you do when you have multiple moves with the same evaluation?
Parent - - By Lukas Cimiotti (Bronze) Date 2012-08-09 19:22
Flip a coin - if it's 2 moves ;)
Parent - By Banned for Life (Gold) Date 2012-08-09 23:08
I don't think Uri is the coin-flipping type! :lol:
Parent - - By Uri Blass (*****) Date 2012-08-10 05:35
I tend to prefer the move when the evaluation does not go down if you search deeper but I have no clear rules always what to choose.
Parent - By Uly (Gold) Date 2012-08-11 05:58
You're interacting Uri! That's why your results are better against those players what will not do as much work as you do, even if your work is minimal, it can be critical. You're actually supporting my point.

Most of the human moves that I play that are better than what the engine suggests don't happen at the root anyway, they only help to refute lines faster and happen at leaf nodes (because I can cut the variation if I play better than the engine and don't like one side), so my claims aren't counted by how many root moves you play that the engines couldn't have found themselves, but about the move choice that you make in general that put together are better than what the engine sees alone.

If the lines of 3 unassisted engines look like this:

1.a ... 2.a ... 3.a ... 4.a ... 5.a ... 6.a ... <- line by engine a
1.a ... 2.b ... 3.b ... 4.b ... 5.b ... 6.b ... <- line by engine b
1.a ... 2.b ... 3.b ... 4.c ... 5.c ... 6.c <- line by engine c

And you play into

1.a ... 2.a ... 3.a ... 4.b ... 5.c ... 6.c ... <- Line that no engine would have played by itself (b and c preferred 2.b)

You're producing something better even if the first move was found by all engines and the others would have been played by different engines in different branches (branches to where those engines wouldn't have gotten themselves into), though I find myself often playing moves that no engine suggests as their preference and that only turn better after deeper examination.
Parent - - By Fellowship (**) Date 2012-08-12 06:01 Edited 2012-08-12 06:03

>


>


>


>


>


>


>


hi Friend

In response of your Ka6 ( your weak move ) , g4 is also weak move which  engine quickly finds .
Note :- e4 is good move  .

Friend
Neel
Parent - By Uri Blass (*****) Date 2012-08-18 20:28
I disagree

I remember that houdini prefered 34.e4 and not 34.g4 and that the pv was Ka6 g4 after a long search.
I believe that g4 is winning for white and it is the reason that Ka6 is bad.

I expected 34.e4 in the game.

After 34.g4 houdini evaluates the position as 0.00 if you do not give it to search for a very long time but houdini is wrong.
If 34.g4 is a weak move then what was my mistake after 34.g4?
Parent - - By leavenfish (***) Date 2012-08-08 12:20
How would a "fuller" tree be of use?

I'm also not sure how you 'optimize' an engine for this. A good human player, it seems to me, would just use the tactical gifts of an engine to help them navigate the minfields of play. An engine is great at calculation but its evaluation function isn't going to be as good as a good human. Maybe though you are thinking of how an engine would interact in the hands of a relatively poor correspondence player?

Let me rephrase that. I think it all depends upon the true chess abilities (understanding) of the human player. A good one, can use an engine as is if he needs one, a poor one...well, honestly, if you 'optimize' an engine to allow that person to play closer to the level of a 'good human', I am not sure how it would be done and even if it could, it's kind of a perverse sense of worth or false equilivancy if one is simply trying to 'guide' said engine with his lesser knowledge of chess, but it is up to each how he gets his jollies, IMHO.
Parent - By Bob Durrett (***) Date 2012-08-08 13:55
You have a number of distinct ideas here, each of which needs some deep thought.

I am just up out of bed, so I will tackle the job later today when I wake up.  :-)

Bob Durrett
Parent - - By Bob Durrett (***) Date 2012-08-08 17:48
Before we get too far into this, let me bring in another consideration:

Again, the following comments are based on intuition and not on expert knowledge of the state-of-art in design of chess engines.

My impression is that there are many ways, at least theoretically, to use "knowledge" in a chess engine. Perhaps most of these may not be practical.

My feeling is that a correspondence chess engine, designed to be used just for correspondence chess applications, would benefit from far more knowledge than would be the case for current engines.

Using knowledge can slow down an engine. Engines which must come up with a good move quickly simply cannot take the time to use a lot of knowledge. There is a trade-off between speed and amount of knowledge used. In our correspondence chess engine, time used for computation would not a major factor.

Now, please let be bring in yet another consideration:

My impression is that what the correspondence players here at Rybka Forum are doing, whether they realize it or not, is actually conducting RESEARCH! for example, in an earlier thread, Uly talked about how different engines are good or bad in different kinds of positions. In these games, the humans are researching the strengths and weaknesses of the various engines they experiment with.  This is actually experimental research.

IMHO, the outcomes of the games are not so important whereas the RESEARCH findings are. Engine developers would do well to take note of the research findings of these users!

> How would a "fuller" tree be of use?


I do not have a good feel for how much or whether the "fullness" of the tree would be influenced by use of more knowledge.

> A good human player, it seems to me, would just use the tactical gifts of an engine to help them navigate the minfields of play.


I do this during my post-mortem analyses. My tactical abilities are certainly not at the GM level, and the engines help me to navigate through the lines I investigate. I doubt that there are any strong GMs playing chess at Rybka Forum so they also probably need some help with tactics.

> but it is up to each how he gets his jollies, IMHO.


True.  Most of us find good ways to do that. Only a small minority grossly violate the norms of the societies we live in.  I guess I am going to keep on "doing my thing" without being overly fearful of the bad guys. I really do not like viruses and malicious software but I do what I can to protect myself and then just go on living the way I want to. There will always be people who "get their jollies" by violating agreements but I am not going to change the way I live because of it.

Bob Durrett
Parent - By Regularuser (***) Date 2012-08-12 07:59

>My impression is that there are many ways, at least theoretically, to use "knowledge" in a chess engine. Perhaps most of these may not be practical.


I think this point hits hardest in the endgame.   Through a combination of search and a static evaluation function that adds up points for features in a position in a fairly simple way engines have got stependiously good at middlegames.   Overall engines are better than human than humans at endgames but a human can still add a lot of value to an endgame because of the knowledge/understanding that human and a computer does not.   The nature of this knowledge/understanding is much much harder to program in than the adding up the points of individual features of the position - the position must be looked at as a whole.   A couple of examples that I have used previously on this forum:
- defending a same colour bishop ending, should your pawns go on the same or opposite coloured squares the bishops?   There is no general rule for this.   Good human players can look at the overall characterstics of the position and determine this pretty well.   It is really hard to program this into a comp, and would require a very diffreent desgin to current programs, and the resulting program would search much more slowly.
- a king cut off in a rook ending may or may not be important.  A human can judge this reasonably well, but it would be hard to score this in an evaluation algortihm in software.

A personal view:  we have a few more years of meaningful improvements with current methods, but eventually someone will need to brave and take the diffrent approach to make another leep forward.  I thinkt that new approach will probably emerge from academia.

>My impression is that what the correspondence players here at Rybka Forum are doing, whether they realize it or not, is actually conducting RESEARCH! for example, in an earlier thread, Uly talked about how different engines are good or bad in different kinds of positions. In these games, the humans are researching the strengths and weaknesses of the various engines they experiment with.  This is actually experimental research.


True.  And there are some surprising nuances here.   Uly has contact with the developers but I think most of us don't, so the feedback loop is not being very effective (and I guess I could do my bit to fix this).   On the other hand, it is kind of fun having engines behaving differently, and (perhaps sadly) it is part of the skill of playing correspondence knowing which engines play what types of positions well.
Parent - - By Permanent Brain (*****) Date 2012-08-08 14:34

> I suspect that the tree would be much "fuller."


The tree being fuller means the search being slower. So, it comes down to the question which engine does the better forward pruning, to reduce the width of the tree. I guess you won't find a non-forward-pruning engine within the top-100...
Parent - - By Uly (Gold) Date 2012-08-08 14:48

> The tree being fuller means the search being slower.


It can be as fast if the right variations are fuller. The problem with today's engines is that they're reducing variations that are good which they should be extending, and only find them at much higher depth, while they're extending and using too much time in variations that are best at shallow depth, but that eventually fall or are never better than the alternatives the engine is missing.

A fuller tree could balance this, just because X move was best up to this point doesn't mean you have to spend 80% of your resources on it. That works for short time control games when the opponent engine has worse eval and will use less resources in the variations you're "specializing" on the fly, but in corr chess, resources could be much better relocated without slowing down the engine, actually, the engine would be slower on irrelevant variations and the best moves would be found faster.
Parent - By Permanent Brain (*****) Date 2012-08-08 15:07
Generally, I think that the superior strength of modern engines over old (or beginner's) engines is, that they can choose with very good probability what's important to search deep, and what so stupid (on a first glance) to just require shallow search.

In other words, generally, if you match a strong engine which does much, but state-of-the-art pruning, against an engine which needs to stupidly investigate everything, the first one will always win.

Of course I am considering top qualitiy engines for such a comparison.
Parent - - By Bob Durrett (***) Date 2012-08-08 15:16

> The tree being fuller means the search being slower.


True enough.

However, . . .

For the Rybka Forum variant of correspondence chess, the search being slower should not be a problem since correspondence chess players "have all the time in the World," assuming that they are not simultaneously playing too many games or are working on two jobs just to make ends meet.  The guy with two jobs would probably have his computer working unaided while he/she were at work.

Again, based SOLELY on intuition, I would expect the correspondence player, using a currently available modern chess engine and modern GUI, would want at least ten alternatives evaluated thoroughly.

But the real problem, it seems, is that modern chess engines must prune extensively due to the need to obtain a good move in a short time. The typical engine vs. engine tournament does not allow engines to take as much time as they need to do a thorough job. From the perspective of a correspondence chess player, it seems that there is altogether too much pruning going on much too quickly.

Bob Durrett
Parent - - By Permanent Brain (*****) Date 2012-08-08 15:24
I am not a corr. player but I think that Multi-PV mode could be useful for you, because that forces engines to investigate 2nd, 3rd (etc.) rated continuations further.

It has been reported that when engines were tested in typical test positions, switching to multi-pv made them to solve quicker, sometimes. Maybe this is as useful hint for you! :lol: I wish you optimal success.
Parent - - By Bob Durrett (***) Date 2012-08-08 16:26

> I am not a corr. player but I think that Multi-PV mode could be useful for you, because that forces engines to investigate 2nd, 3rd (etc.) rated continuations further.


Like you, I am currently not a correspondence player, but I did that a lot before personal computers and the internet were widely available here in the USA. When I started, there was no chess software. When I quit, the software was pathetic by today's standards.

Nowadays, I sometimes use the mode where the GUI is showing multiple lines and I have had times when I felt that the program was doing a better job of finding the best move in that case. Of course, it takes longer to reach the desired search depth. Also, I sometimes use a feature of my GUI which forces the engine to examine the next best move.  [This is Ctrl-y in ChessBase GUIs.]

Truthfully, I am still a bit confused as to how the labor is divided up (allocated or shared) between the engine and the GUI. For example, when the user (human) tells the chess computer (PC plus chess software) to display multiple lines, does the engine "know" that the user has done that? Or, is the GUI the only one who "knows"? Does the GUI control the engine and tell it when to look at the next line?  Incidentally the display typically updates itself to show the "best" line on top and remaining lines in order of strength (as determined by the engine). Presumably, this re-ordering is a GUI function.

Incidentally, I am NOT trying to re-start a debate as to whether or not engines and GUIs actually think the way humans do. Obviously, they do not.

Bob Durrett
Parent - - By Vempele (Silver) Date 2012-08-08 17:48

> For example, when the user (human) tells the chess computer (PC plus chess software) to display multiple lines, does the engine "know" that the user has done that?


Yes. The GUI sets the MultiPV option to a value greater than 1 before telling the engine to analyze.

> Incidentally the display typically updates itself to show the "best" line on top and remaining lines in order of strength (as determined by the engine). Presumably, this re-ordering is a GUI function.


Not necessarily.
Parent - - By h.g.muller (****) Date 2012-08-08 20:29
WinBoard orders the lines (of same depth) by score, even if the engine finds them in a different order.

The GUI does not tell the engine to find the next-best move in normal multi-PV mode. It tells the engine before the search starts how many lines it has to come up with. The engine then determines in which order it finds them.

This is a bit different when you use the exclude-moves feature. (E.g. by asking for 'next best', or explicitly excluding a certain move.) Then the engine is told which moves it can search (if it is UCI) before the search command. That way you can prevent it searching the best move at all. But the moves that can be searched will still be searched in an order determined by the engine. Furthermore, I think that all UCI engines start a new search whenever you change the set of moves to be searched. This because UCI requires a stop and new search command to be given to alter the move list. So it is not like with WinBoard engines, where you can simply exclude moves one by one during an existing search, without losing any search results.
Parent - - By Bob Durrett (***) Date 2012-08-08 22:54

> The GUI does not tell the engine to find the next-best move in normal multi-PV mode.


I am terribly embarrased to admit it, but I do not know anything about multi-PV mode. I probably am using it and don't know it. Note that I used Rybka for a long time and now also Houdini, both as UCI engines only.

Thanks for the clarifications regarding engine vs GUI!

Incidentally, to go back to the initial bulletin in this thread, I would like to offer the following general principle I learned while an engineer (before I retired):

Principle:  If a design is optimal for a certain application, it is not necessarily optimal for a different application.

For example:  If chess engines are designed to win engine vs engine tournaments (at relatively fast time limits), they may have been optimized for that application. However, if the same engine is used in a different application, it may be far from optimal.

Corrolary: A chess engine optimized for fast chess and a chess engine optimized for correspondence chess may be two different animals! (Plants? Machines? Whatever?)

Bob Durrett
Parent - - By Barnard (Bronze) Date 2012-08-09 01:37

>Principle:  If a design is optimal for a certain application, it is not necessarily optimal for a different application.


wow,i dont need to be an enginer to know that...and i will tell you more:i never studied more than the school,not high school,and even i know that a medicine that is good for an illness,maybe isnt as good for another illness

so,with that assertion,you simply left me spechless,and with that face :roll:
Parent - - By Bob Durrett (***) Date 2012-08-09 02:03
It is old age. Like I said before, as I get older I tend more and more to think in terms of general principles instead of specifics.

It is like the story of a chess player consumed with analyzing many lines during a game versus someone who plays like a GM without all that analysis.

Of course, I knew that the principle would leave people speechless!

Nevertheless, I am glad you agree that it is true.

What is really at issue is whether or not modern engines, optimized for winning fast engine vs engine tournaments, are optimal for correspondence chess applications.

General principles can be deceptive! Their simplicity can mask their profound applications.

Bob Durrett
Parent - - By Barnard (Bronze) Date 2012-08-09 02:37

>Nevertheless, I am glad you agree that it is true.


i cant tell that the principle is wrong,sine it is a basic premisse for every situation at the life

>What is really at issue is whether or not modern engines, optimized for winning fast engine vs engine tournaments, are optimal for correspondence chess applications.


well,lets say a very drastical scenario:

a engine 'x' that is able to choose 'perfect moves' if it reach depth 40 based on his perfect knowledge embedded to it,but of course,that engine will work at very very very low kn/s and will reach a very very slower depths compared to another top engine that hasnt that knowledge embedded on it

the question is very clear for me:if you give enough time to reach the depth at it plays ''perfect chess'',the engine with the ''perfect knowledge'' will beat every single other engine

but...how is the time that he needs to reach that level that allow him to play perfect chess?maybe years for each move,or decades,or infinite time,i dont know

but a top engine that will reach,for example,the same depth in 1 hour,even without that knowledge,and the engine 'x' reach in that hour only depth 3,im sure that the engine that isnt able to play perfect chess if is given enough time,will defeat it game after game

now that aplied to the present engines:

if an engine with a vey high level of knowledge embeded is given enough time,will win more games than the faster engine but with less knowledge...

now the second part:what is enough time?if you play a correspondence chess,and you have an effective 8 hours of analisis each day (assuming you play 1 move day),and with that 8 hours that you have,the engine reach only very low depths compared to the engine with the less knowledge,im sure the engine with less knowledge embedded will score very good against it

the problem is mainly resumed to that:
can you have enough time to play stronger based on knowledge,keeping in mind that you dont know how much is ''enough'' traducted to time,and even knowing it,surely the correspondence game will be played faster than the time needed to make prevail the knowledge over the speed?

im an atheist,and i dont trust in anything without proofs,so i wont use none of that pseudo-engines,that are killed at normal time controls like blitz,and they need,from the words of his authors,something that cant be measured in a reasonable amount of time,and that is the concept of ''enough time''
Parent - - By Bob Durrett (***) Date 2012-08-09 02:45
You have many ideas in your bulletin. I will look at each later. Perhaps I should be quiet for awhile and let others respond.

> and i dont trust in anything without proofs


While studying math, I stumbled upon a very disconcerting fact: All of mathematics is based on a set of unproven postulates.

Proofs can only show the consequences of the truth of an assertion. But, if the assertion itself is an unproven postulate, then it is like a house build on a foundation of sand.

Hence, I recommend that you do not put too much faith in proofs.

Bob Durrett
Parent - - By Barnard (Bronze) Date 2012-08-09 02:54

>While studying math, I stumbled upon a very disconcerting fact: All of mathematics is based on a set of unproven postulates


well,i never studied,but even me,know that maths dont need all time to prove anything,sometimes is more than enough prove that it cant occur the opposite,to indirectly,prove it

2 boxes,one with one hat,and the other box,empty...can you prove in what box is the hat?not
so you simply open one,and is empty...can you prove now the hat is in the other box?yes,but deducted indirectly from the empty box

regards
Parent - - By Bob Durrett (***) Date 2012-08-09 03:08
Yes.

Where I live there are sometimes "magic" shows in which the magician uses slight of hand to fool the viewer.

I can imagine a magic show with the two boxes and the hat.

The magician puts the hat in one of the boxes and then shuffles the boxes so that the volunteer does not know which box the hat is in.

Then the magician asks the volunteer to open one of the boxes. The volunteer does and the hat is not in that box.

Then the magician asks the volunteer to open the other box. To everybody's surprise, the hat is not in the other box, either.

What happened? By slight of hand, the magician had removed the hat without anybody seeing him do that.

The volunteer had concluded that the hat was in the second box because he did not find it in the first box and because he ASSUMED, without proof, that the hat was still in one of the boxes.

This illustrates that a proof is only as good as the truth of the assumption.

:-)

Bob
Parent - - By Barnard (Bronze) Date 2012-08-09 03:25
im sorry,but if you are comparing maths and what i described with simplicity because i cant do it better,to prove something with indirects proofs or proving the opposite cant occur,is simply ridiculous

and just an argument to your joking scenario,is that if you make a change to the initial system after the initial settings needed to start trying to prove 'x',it changes the whole system itself,and invalidate all the results after that change and the system itself in the original scenario

regards
Parent - - By Bob Durrett (***) Date 2012-08-09 15:51
Well, I am just happy to be able to talk to you.

Can we talk about chess or chess software now?

Bob Durrett
Parent - - By Barnard (Bronze) Date 2012-08-09 19:19
yes,of course

what engine do you preffer if you play correspondence chess?
Parent - - By Bob Durrett (***) Date 2012-08-09 21:50

> what engine do you preffer if you play correspondence chess?


I have not played correspondence chess for a very long time. However, I played many games before I quit.
There were no chess engines, no personal computers, and no internet when I played most of my correspondence games.
Moves were sent and received by post card. Games typically lasted for at least a year.
Right at the end, I had a chess computer but it played very badly and I did not use it for correspondence chess.
I did not get ChessBase's Megabase game collection until long after I quit playing correspondence chess.
I had some interesting games but they are probably played at a lower standard since I could not check my tactics with a computer.
We moved five months ago and most of my stuff is scattered between book cases and boxes but soon I may try to find some of that stuff. I still have a "Post-A-Log" in which I kept some of my games. I remember a "Golden Knights" tournament I participated in. (USA)
If I were to start playing correspondence chess again, I would have to get used to the chess engines since I never used one in my correspondence games. None of my opponents did either to my knowledge.

I feel the need to greatly improve my ability to do independent analysis again. That takes a lot of self-discipline but this is necessary to think clearly. I guess I would use my existing programs to check the tactics before I "mailed" the moves because I would not want to waste my opponent's time. Currently I have Rybka and Houdini. When Rybka 5 and Houdini 3 come out, I will get them.

I am also writing a chess book (on human psychology) and the research is taking a lot of my time. I do not know how much time I would be willing to spend on correspondence chess. If the book is published, I will then get rich and buy myself an ocean-going yacht. :-)

Bob Durrett
Parent - - By Barnard (Bronze) Date 2012-08-09 22:19
that is very interesting

i never played that such of tournaments (with postal cards) because when i started playing correspondence chess,i was registered at IECG (after moved to LSS),and are web based tournaments

but i think is too much more in the romantic era,play with post cards :smile:

why you dount if the book will be published?im sure it will be a good book,and you will have lots of revenues :smile:

regards
Parent - - By Bob Durrett (***) Date 2012-08-10 00:18
I just joined the Lechenicher SchachServer.

Later, I will study the website.

I should enter a beginner's tournament.

What would you suggest?

Bob Durrett
Parent - - By Barnard (Bronze) Date 2012-08-10 00:40
well,if you are recently registed,you will be allowed to play just 4 games,if Ortwin hasnt promoted you to the next level

so i sugest you that you will register in an Open Tournament

go to new tournament,and you will see the tournaments that you are able to register in

i dont know your strength playing correspondence chess with engine,but usually,the players that register at open tournaments play really poor

if you want to play against relatively strong players,before you register in the open tournament,take a look at the participants

but my ssugestion is that you will register in a tournament that all the players are rated with very low elo(about 800),because that means you will play your games faster,and you will be promoted to higher levels sooner
Parent - - By Bob Durrett (***) Date 2012-08-10 02:03
First, I must learn and understand the rules.

Is checking my moves with a chess engine considered OK, or not?

Bob Durrett
Parent - By Barnard (Bronze) Date 2012-08-10 16:03
yes,engines are allowed to analize your games

EXCEPT in the tournament marked as ''not engines'',you will see it under new tournaments when you will allowed to acces it
Parent - - By Banned for Life (Gold) Date 2012-08-10 01:47
If you expect to be good at modern CC and want to play at a high level, it's critical to come in at a high level. Did you do this? Make sure LSS uses your highest level from previous CC play. Otherwise it will take forever to reach these levels again, and time is working against you...
Parent - By Bob Durrett (***) Date 2012-08-10 02:01
I have no expectations at all. As you say, "time is working against me." I have no delusions about my ability. My purpose is to improve and not to prove anything. It is much too late in my life to care about doing that. Let the youth do that.

Bob Durrett
Parent - - By Bob Durrett (***) Date 2012-08-10 03:30
What is the IECG? Should I join that too???

Bob Durrett
Parent - By Barnard (Bronze) Date 2012-08-10 16:07
no,you cant join it,because it is closed

IECG was the primary site were we played correspondence chess

after a few years,IECG anounced it is closing his service to play,so Ortwin programmed that new server,were all of us moved to play
Parent - - By Banned for Life (Gold) Date 2012-08-10 16:17
To clarify, LSS absorbed the IECG.
Up Topic The Rybka Lounge / Computer Chess / Modern Chess Engines not Optimized for Correspondence
1 2 3 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill