I have some specific questions. I was wondering what you guys think:
1) In your opinion, at slow time controls (at cc level, 1-2 days a move) does the same rating list hold? (when more time is given, results may be surprisingly different) Is Houdini still the best? (positional understanding may gain more in value as time increases) .any ideas? Did any one study this yet?
2) Komodo: main strengths over Rybka?
3) Houdini: main strengths over Rybka?
4) iDeA: which engine is best?
5) Endings: which engine is best? Is Shredder not the best anymore?
6) Why is Houdini better than Rybka? (positional? tactical? increased depth?)
More important is ..
1, How to maintain tension where every move is most important...
2.How to get onto those positions where you with better interactive analysis have chances of winning.
3.Certain engines tend to go for straight play where gain is found but after hours of interactive deep analysis leads to certain draw...
( Which must be avoided ?!?! i am not obvious which sign fits here.)
4.Depending on your opponent's level of play you should decide where you will have better chances..
And for me ( i do not have grudge against ) Houdini 2.0c is only for endgame ...
Opening and Middle game phases are not for H2.0c which is better in fast time controls..
I really appreciate the best effort in those losing positions where H2.0c can put ..
( Here lots of CC player do not agree with me.. and it does not bother me..)
I prefer Rybka in the opening and Middle but not in endgame where it's exaggerated evaluation is bad...
However i would like to put a caveat here,to get best out of R4 you need big hardware at very least a quad.
>Hi, I am correspondence chess GM
Hello Dr. Turgut.
Hope you get your questions answered.
Good luck in your games.
Hope you had a good time at the Chicago Tournament.
>Komodo: main strengths over Rybka?
Better understanding on positions where there's no clear path to take and you need a plan. However! You should avoid such positions or situations on your correspondence games, since you should have a plan and know what to do way long before the position is reached, if you reach a position where Komodo is useful, you've done something wrong previously already.
>Houdini: main strengths over Rybka?
Speed. Better evals in positions where Rybka is over-optimistic and shows very high scores but they're drawn. Houdini probably also has a tactical advantage. These factors don't offset Rybka's advantages over Houdini, which are better granularity, analysis of material imbalance positions, more accurate evaluation of queen's value, seeing the difference between two similar, non-transposing plans, only showing close 0.00 scores for really drawish positions instead of giving this to most moves, better long-term eval and rightly showing high evaluations for advantageous positions while Houdini dismisses the position as drawn, among others.
I don't use Idea. I don't think there's an engine generally good in all endgames, and Houdini isn't better than Rybka for interactive analysis of corr games.
Move A: 0.21 score
Move B: 0.22 score
Move C: 0.23 score
Move D: 0.24 score
Move E: 0.25 score
Move F: 0.26 score
Move A: 0.20 score
Move B: 0.20 score
Move C: 0.20 score
Move D: 0.21 score
Move E: 0.21 score
Move F: 0.21 score
The latter is what I observe in Houdini in such positions after interactively refuting the variations, Houdini gives up and says "Gah! All moves are pretty much the same, go with your preference as the 6th move is just 0.01 worse than my original preference." Please note Houdini's scores may show actually a more accurate eval of the moves (say, if you played infinity games the first three moves would end with the same performance, and the next set would also end with the same performance that is a bit better), but for analysis this is not useful because you're only going to play one game, and you want to maximize your chances of best performance. Throwing a dice and picking a move will not do.
This is something that can't be fixed by just multiplying Houdini's scores for some factor.
In the end, granularity equals how much the engine is willing to defend its move preferences instead of giving up and seeing all the moves as equal.
IDEA has built-in loss of granularity due to the way it stores scores, so I guess IDEA users wouldn't notice Houdini doing worse than other engines at this, but the other mentioned problems added up will still have Houdini scoring more moves the same than with other engines.
how you achieved the title of correspondence GM without knowing it?
There was correspondence chess long before there was any chess software.
Some have excelled even without computers.
Once you get a title, such as GM, you keep it for the rest of your life.
but if he is a correspondence GM without engine,with engines,he can easily go to the higher positions at the site he chooses to play correspondence chess
Tansel, who achieved his GM-title in 2008/2009 when Rybka was the only engine who could seriously be consired (+ Shredder in the endgames), obviously didn't play too much in the last 3 years (only 10 games) and just use[s|d] Rybka... and as there is much talk about Houdini, Komodo, Critter and others these days he most likely has started this thread to inform himself.
well,in that case,his question is plenty of sense...but i think it can be a good idea to tell prior to ask the question,when he achieved the totle,and what engines he used,so we had a basis to give him a better answer
> I notice that it took him five years to reach his current level.
While its more a matter of number of games than years, it will be a lot more difficult these days... the competition-level is -due Rybka/Houdini & hardware & opening-books and other publicly available resources) a lot higher then five years ago which results into a lot more work and still a very high draw-level... that's also why most top-players which earned their ELO "in the good old days" just play very rare games nowadays as their ELO is likely to drop instead to rise.
This is similar to the question as to the extent to which a high OTB rating (IM, GM) helps a chess engine programmer to develop a stronger engine.
Do we know how well our new friend (Tansel) has done in OTB chess?
I would like to ask him how he did it. I hope he continues posting here.
> the extent to which prior achievement in OTB chess (IM, GM) helps someone to rise quickly in this "new" kind of correspondence chess (where all the available resources can be used).
It's obviously no disadvantage to understand the game , but its no necessity either when you're willing to compensate any missing understanding with a plus in other resources (willing to work, hardware, opening-books, ...). I can't give you any formula to answer your question, i think the best one can do is play and play and play... and try to improve one's own play with the experience gathered.
> Do we know how well our new friend (Tansel) has done in OTB chess?
His FIDE-rating is 2294: http://ratings.fide.com/card.phtml?event=6300332
This is just my feeling, however. It would be nice to obtain some concrete evidence.
My question presumed that both players would have the same resources.
Assume that a 1400 elo player with lots of experience with chess engines does have the same performance of a 2300 elo player with little experience with chess engines.
Then, all the 2300 player has to do is gain more experience with chess engines and he will have a better performance than the 1400 elo player. And it's much easier for the 2300 elo player to gain experience with chess engines than it's for the 1400 elo player to increase his rating.
A 1400 elo player may have enough knowledge to improve on the engine's plans by simply noticing a Bishop is stuck in a corner and that line should be avoided, while the 2300 elo player could be able to improve on the engine's plans by improving its play in more subtle situations where his knowledge applies.
> It would be nice to obtain some concrete evidence.
A concrete evidence for something so obvious like there is a sun rising up every day? Its like asking for an evident proof that a certain position is won... reality is that such an evident proof often would fill millions of paper-pages with analysis&lines. Of course chess-knowledge ("ELO"; there are for sure a lot unrated and yet strong people out there) helps analyzing chess-games, and of course the impact of other resources -like i pointed out above- can't be neglected either. As Lobostepario/Gino pointed out in the other thread the average FIDE-ELO of the top ICCF-players is below 2200, still there are a few FIDE-GMs like Ulf Andersson at the top of the list... others could do the same, but they simply don't care about corr-chess.
> A concrete evidence for something so obvious like there is a sun rising up every day?
Perhaps my choice of words was not good since I obviously did not express the idea I intended.
What I was seeking was not proof or convincing evidence to prove a point.
Instead, I was trying to say that I would like to examine in more detail the ways in which a strong OTB master might approach correspondence chess; differently from the way a much lower rated OTB player (like me) would do in correspondence chess.
It goes without saying that there would be differences. Debate on that point would be pointless. But, what are the specific differences? That is my interest.
1) I have not played much cc in the last 1-2 years (my last tournaments was WC 24 Final)
2) I am familiar of strengths and weaknesses of old programs (Rybka, Junior, Fritz, Naum, Zappa etc) (all my own theories) But I have no experience in Komodo, Critter an Houdini. I am trying to learn and understand these programs now.
3) cc is completely different. In order to be very good, you need to be able to say "these engines are not correct, and I will go my own way." (otherwise game ends up drawn) But copying the engine is no fun.. there is no point playing correspondence chess then...One needs to have a "feeling" to decide when exactly the computer is not finding the best... hard to explain this as a science.. But this is where human+computer combination excels. But, this is risky and takes a lot of responsibility. Due to this reason, I am trying to learn their strengths and weaknesses.
a)Critter 1.6a "..as long as you know how to use the session file correctly to speed up analysis...." what do I need to do?
b) No clear path: here I differ in opinion: I think that this is best for cc. I don't want a clear path. This is where human can make a difference. less calculation, better for human..
c) iDeA is a very useful tool. I used it a little, but I think it is extremely powerful for cc.
a)As far as I can understand, you do not think Houdini is very good for cc, but you like it for endings. is that correct? So, for cc, Houdini may not be the best. Any one else thinks same?
b) Rybka 3 Dynamic: this is an interesting program. i also liked Human version.
6) Shredder: i have a hard time believing any engine surpassed this program. I will check it with Stephan. I am very impressed with this program.
a) thanks for the technical help! (i am following him from 3 years behind. I just built something similar to his 3 year old SR2 computer)
b) OTB: I am hardly playing, I am taking my son to chess tournaments now. My 8 year old son beat a 2263 (2163 fide) USCF master this weekend! (He is playing for US team in Slovenia for 8 year old World Championship next month). he is 1744 rated already.
8) Komodo: GM Kaufman says it is already equal to Houdini (non MP version) in private matches. Kaufman is one of the best players in the world understanding material unbalance. (If you have not read his article in ChessLife 4-5 years ago, you should read it. A classic. One of the best articles ever written about computer analysis, and material unbalance)
9) Hiarcs. This was also a very good program. Can it hold on longer time controls? (maybe not fast enough, but this understands chess)
10) mindbreaker: thanks for the ideas. i will work on them.
thanks everyone! please add your opinions as I find them very valuable.
You should wait a few weeks since I will be publishing an interview with the new ICCF World Champion Ron Langeveld in the ICCF website.
He will be discussing some of his analysis methods, I think this will be very informative
Take care my friend,
As far as i know they (normally top) hide the truth..
Doubt there is a single secret that makes someone a World Champion. But it will be revealing to pick his brain at least once.
If you please and he too.
Will you tell him to message me ?
> a)Critter 1.6a "..as long as you know how to use the session file correctly to speed up analysis...." what do I need to do?
After installing the engine, change these options:
Tick the "Use Session File" box.
Change SF size to 64 (Note: 64MB should be enough space till Richard releases next Critter. Keep in mind what you set here will be added to Critter's Memory usage.)
Set SF Write Depth to 11 (Unless you analyze lots of positions with a depth that is lower than 19, in such case you'd like to use a lower number.)
Tick the "Resolved Score Drops" box (critical for optimum performance of Session File on Fail Lows.)
Save this as a personality so you only need to do this once.
After doing this, Critter will now remember all positions that you give it to analyze, and its scores, having a great speedup when revisiting positions, and even a greater one when remembering refutations. Also, when you go back to previous positions, Critter will automatically look for alternative moves at lower depth, so that you don't have to use MutiPV or Exclude Moves for this: if Critter didn't change its move choice, there is no better move, and if there's a better alternative, Critter will switch to it automatically (this is more powerful than it sounds.)
However, Critter has its nuisances, you can't just set it up and analyze normally as you would with other engines. For instance, backtracking information backwards is tricky, if you just leave Critter analyzing a position for 8 hours and leave, when you come back and find a line of your own that is better, Critter will not add it to Session File unless its relative depth is +1, which may take another 8 hours, and as the time accumulates, eventually you'll have to analyze an irrelevant position for hours and hours just so the new information gets added to the file.
For this reason you need to analyze to some average depth with Critter, and never go over this depth unless it's needed for Critter to learn new info.
Critter thinks the best line is 1.a a 2.a a with score 0.20 at depth 22
You know this line is bad, you have move 1...b!!, and you show it to Critter.
Critter thinks 1.a b 2. b b 3. b has score 0.05 depth 22.
Now, when you go back to the first position, two things may happen:
1. Critter knows 1. a b because it learned it, and will either show a move better than a automatically, or will show 1. a b as best up to depth 22.
2. Critter extended 1.a a 2.a a too much and has learned it with a depth higher than 22, so it'll think it's best, unaware of your refutation.
If 2) happens, you need to go to 1. a b again and reach depth 23, as you go back, Critter will have learned correctly and will show the 1) case as above.
And that's about it. The engine is storing in harddisk what you analyze, and it's very effective, I just have to wonder why engines that do this are so rare.
> b) No clear path: here I differ in opinion: I think that this is best for cc. I don't want a clear path.
Komodo is useful in finding good paths where there are unclear ones, if you don't want a clear path, you'll probably find Komodo useless.
> c) iDeA is a very useful tool. I used it a little, but I think it is extremely powerful for cc.
And I think it's extremely dangerous, specially if it gives a false sense of security while it's hiding an important move that is the opponent's best plan.
>9) Hiarcs. This was also a very good program. Can it hold on longer time controls? (maybe not fast enough, but this understands chess)
I'll plug an ad and claim the Hiarcs Paderborn 2007 version is the most useful for correspondence games.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill