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  
Parent - - By leavenfish (***) Date 2012-08-16 07:36 Edited 2012-08-16 07:39
Look, 48 hrs of 'computer's churning over a move' is just that. Give it up. Don't keep shifting your argument  - now it's to the time a correspondence player (if that's what you call computer aided 'help chess') spends on a game vs an OTB game. I was probably playing 'mano y mano' correspondence chess before you were born (back just prior to 1980 btw), so don't lecture me on correspondence chess. And it's not you...I don't know you from Adam, but you butted in my and Bob's discussion on a point and and my likening computer aided chess to 'swimming with jets'. That's your preogative and you were welcome to do so...but you always seem to want to disagree with me and uphold some freaky 'computers are supreme' ethos when your 'help chess' is a completely different animal than real one on one chess...I've heard that from you in other threads and frankly it just gets tired...but that's your preogative as well. But it is also my prerogative to call a halt to it when it starts to get lame. Do you 'computers are supreme' bit all you want and if others want to discuss it, so be it. I'm just not going to. If you want to consider that 'being emotional', so be it...

There is an old saying that 'we are all entitled to our opinions, but that doesn't mean they are all equally valid'. Go back and look at some of the nonsense you have said and I have quoted you on and sleep on it. But, maybe you have a computer do that for you as well and think your sleep is better and more restful that what we humans do. :wink:
Parent - - By Uly (Gold) Date 2012-08-16 14:50
You still haven't explained what do you do in your computer unassisted games that can't be done in the computer assisted games. And why it can't be done in computer assisted games.

Correspondence time controls give enough time for everything, you can get off the car and run marathons, you can take off the jets and go for swimming, whatever it takes to find the best move, including ditching all engine analysis and going for your move because they have no clue.

If you played 'mano y mano' (AKA "mano a mano" in Spanish) so much that you could never learn how to implement engine help correctly on your games, please don't assume all the people would just put computers to churn on a position over a move. It requires creativity to know how to use engine's output correctly.
Parent - - By leavenfish (***) Date 2012-08-17 00:42
"You still haven't explained what do you do in your computer unassisted games that can't be done in the computer assisted games. And why it can't be done in computer assisted games."

"Can't"???? Dude, I never claimed that....don't try and drag me into an argument that exists only in your mind. This is over.
Parent - - By Homayoun_Sohrabi_M.D. (***) Date 2012-08-17 00:57
I think the corr guys are a good thing for engine authors because they need to buy all the engines and bring some money in.   The OTB guys on the other hand are a lot of times satisfied with the strongest free engine available or just purchase one engine for analysis.
Parent - - By tomgdrums (****) Date 2012-08-17 02:11

> I think the corr guys are a good thing for engine authors because they need to buy all the engines and bring some money in.   The OTB guys on the other hand are a lot of times satisfied with the strongest free engine available or just purchase one engine for analysis.


I would concur with this!

Although I have suffered from a bought of engine mania for the past three or so years.  :)  But I only play OTB (or unassisted real time chess on ICC or Playchess) and I am now (finally) seeing how I don't need all the engines.

I agree with Uly that for OTB it is good to have an engine that "agrees" with your own taste in chess.  But I also do think it is good to have an engine or two that don't agree with my own taste in chess.  It makes me look at different options and explore positions in post mortems a bit deeper.  Which would hopefully make me a better rounded player.

I am currently in a phase of checking my own post game analysis of my games with three engines.  And it is proving fruitful!
Parent - - By Homayoun_Sohrabi_M.D. (***) Date 2012-08-17 04:05

> I am currently in a phase of checking my own post game analysis of my games with three engines.  And it is proving fruitful!


Cool!:cool:

The only thing that would help my game is TIME, like having a 30 hour day.   I need to improve my openings (I'd like to read a few pages of the new Watson book), do a few hundred tactics problems a day, improve my endgame...I'd also like to look over some GM games...and maybe play a few games every day and...:wink:
Parent - - By tomgdrums (****) Date 2012-08-17 12:18

>> The only thing that would help my game is TIME, like having a 30 hour day.


You know what has helped me with time is Peshka!  At first I didn't like the software so much but they overhauled it and there are a lot of great training courses. Now it works great and is very efficient!   And since all the courses are in one place it is easy to get in 30-45 minutes of chess training with my morning coffee before I head off to work.  A little tactics, a little endgames etc. etc.  Very efficient.

Sure, I would love to have the time to get out a nice board and set up all training positions and push some wood around but alas that doesn't happen much.  :)

So Peshka is the next best thing.

And for going over a master game I have found the Chesstiger app for Ipad to be the best and most efficient thing!  It supports variations so you easily check out any annotated variations without having to move pieces around and resetting the board (on a real board)  Again, the real board is more aesthetically pleasing but life is life.  :)
Parent - - By Homayoun_Sohrabi_M.D. (***) Date 2012-08-17 12:24
I didn't know about Chesstiger app for iPAD, I'll check it out.  Thank you.
Parent - By Barnard (Bronze) Date 2012-08-18 01:22
from what i know,if you try to find ''chestiger'' for ipad,you wont find it:

Christophe Theron has changed the name of the product (app) to ''Chess Pro - with coach'',at least for ipod
Parent - - By Uly (Gold) Date 2012-08-20 22:04

> But I also do think it is good to have an engine or two that don't agree with my own taste in chess


That may be good if you have exhausted your own ideas and the engine's ideas on analysis, and you want more. However, I found out that it's more time effective to just play more OTB games in that time instead of doing further analysis.

So:

Analyzing with optimal single engine + playing more OTB games in that time > Analyzing with several engines but playing less OTB games.

Looking at too many engine's ideas can be distracting, better to focus with a single engine on the time you spend on OTB games. I think the main improvement will come from the games, so at the point you spend more time analyzing postmortem than playing games is time that could be used better. I don't even analyze my won games.
Parent - - By tomgdrums (****) Date 2012-08-20 22:10

> <span class="htt">In Response to</span> Uly


> But I also do think it is good to have an engine or two that don't agree with my own taste in chess
> That may be good if you have exhausted your own ideas and the engine's ideas on analysis, and you want more. However, I found out that it's more time effective to just play more OTB games in that time instead of doing further analysis.
>
> So:
>
> Analyzing with optimal single engine + playing more OTB games in that time > Analyzing with several engines but playing less OTB games.
>
> Looking at too many engine's ideas can be distracting, better to focus with a single engine on the time you spend on OTB games. I think the main improvement will come from the games, so at the point you spend more time analyzing postmortem than playing games is time that could be used better. I don't even analyze my won games.


Oddly enough I have found my chess game improving (finally) by playing one good long OTB game a week at the chess club and doing a solid post mortem on that game.  And then spending the rest of the week doing a post-mortem on a master game or two.  (focusing on specific concepts)
Parent - - By leavenfish (***) Date 2012-08-21 01:11
"Oddly enough"?

Just curious...why do you put it that way?
Parent - - By tomgdrums (****) Date 2012-08-21 01:23

> "Oddly enough"?
>
> Just curious...why do you put it that way?


Because most of the time people will tell you to play a LOT to improve.  And there is truth to that but I have also noticed that when I play a little bit fewer games that are more meaningful (ie:  Longer games) that my game gets better.
Parent - By leavenfish (***) Date 2012-08-21 03:52
'Most of the time'...people don't know what they are talking about.:wink:

A lot of bad habbits repeated is only going to accentuate those bad habbits. What you are doing is going to be more productive, plain and simple. Keep it up!
Parent - By leavenfish (***) Date 2012-08-18 01:09
We boys do love our toys, that is true! I can't count the number of engines I have...the latest Houdini, Komodo, Rybka, and older HIARCS, Fritz...several older ones and some free ones as well.  They really don't help my OTB game, but they are great to play around with.
Parent - By Bob Durrett (***) Date 2012-08-09 00:07
A related issue is: What is the best way to use hash tables in the proposed application? This may be application-dependent.

In your multi PV mode, there is the question as to how the engine is utilizing hash tables and whether or not it would use the hash tables in the same way in a normal mode (not multi PV).

The whole idea of using hash tables, if I understand this correctly, is to avoid duplication of analyses. For example, if a particular position is reached in different analysis lines, the findings for that position should not have to be re-calculated. Instead, just refer to the hash table and get the desired information immediately.

The definition of "Hash Table" may vary with programmer. One programmer may have a different idea about hash tables than another. The way hash tables are used may also not be the same from one engine to the next unless they are clones. Of course, if one programmer has merely used the code of another then there may not be any difference.

If you are going to evaluate multiple lines from the same position, then a position encountered when analyzing one line may recur in some of the other lines. Use of hash tables should make multi-PV mode more efficient.

I am probably boring the engine programmers to tears by saying all this. I hope not.

Bob Durrett
Parent - - By Uly (Gold) Date 2012-08-09 01:08
So your opinions above are based on never experimenting on what engines suggest after hours of analysis and comparing it to what one finds after interactive analysis and seeing how much better they could spend their time?

Fast time control games and corr time control games are different beasts, I think that an engine optimized for correspondence would not do very well at blitz and would have lower elo on the rating lists, because finding the best move takes time. I wouldn't be surprised if such an engine wouldn't show any move choice until it had a clear preference, and until all its PV moves shown would be high quality, and had a plan, e.g. "look, this line goes to a position with much better chances for me than the root, and after thorough analysis of every move in it, I couldn't find better alternatives". For drawn positions the engine could go either for sharp variations "look, this line goes to a position that is equal to the root, but I have many alternative moves along the way, making it harder for my opponent to predict my moves, so at least playing it out would be easier for me" or complex variations "look, this line gives plenty of alternatives to both sides, with decent end-positions for all lines, which is the one that gives the highest chances for a decided games, but I have this emergency line to play in case the other ones fall down after more analysis, while I don't see one for my opponent, which could potentially give excellent chances." Etc.

There's just no time in fast time control games to do things like that, and the PV moves of the engines are crappy in general, I've managed to analyze my games all along taking only the move choice of the engine and ignoring the rest of the PV, for this reason. And it doesn't get better with more time, after hours of analysis, the PV of the engine is still low quality, and it takes a few minutes to refute!

So, please, play some correspondence games, gain experience on the issue, and then, we can have a conversation that refers to what to do when you have days to choose a move and when the strategy of the engines for short time control games leads to huge wasted time.
Parent - - By Bob Durrett (***) Date 2012-08-16 16:12
If the community of chess programmers ever does become interested in creating a chess engine optimized to play correspondence chess, and if they produce several or more such engines (all distinct, not clones), then the next logical step would be to stage engine vs engine tournaments played at correspondence chess time limits. If there was enough interest, a World Champion Correspondence Chess Engine could be awarded to the strongest correspondence chess engine.

I see no merit in entering a (yet to be designed) correspondence chess engine into a fast time controls tournament. That would be pointless.

I, with my magic crystal ball, can also see future correspondence chess tournaments with mixed participation; some human and some engine.

:-)

Bob Durrett
Parent - - By Uly (Gold) Date 2012-08-16 17:38
Well, ask the programmers, I wouldn't be surprised if they told you and claimed that the behavior of the engine is already optimal, and that what works at short time controls keeps working after hours and after days. Some people believe this, with they thinking that an unattended Rybka Cluster would be a beast at corr time controls.

>I, with my magic crystal ball, can also see future correspondence chess tournaments with mixed participation; some human and some engine.


As I told you in a different branch this is already happening, with non-centaurs players that only play engine moves, and that have nice results against other non-centaurs players that only play engine moves but that have slower hardware, or a worse book, and also getting nice results against corr players that mainly play their own moves which are inferior to the engine's (I played such opponents that surprise me with moves no engines would ever play, and then they go and lose the game), which means the non-centaurs (engines) have no incentive to try harder. The problem corr chess faces is the very weak opposition that it sometimes has, against such opposition more effort is not more rewarded than few effort, so mediocrity is allowed to succeed.

I still believe the top corr players that have low draw percentage achieve so with high centaur interaction, and human input.
Parent - - By Bob Durrett (***) Date 2012-08-16 18:40
Here are two distinct thoughts:

> Well, ask the programmers, I wouldn't be surprised if they told you and claimed that the behavior of the engine is already optimal


When I was active at CCC, many years ago, I read many bulletins discussing the problems with incorporating too much knowledge into a chess engine. Of course, they were talking about engines designed to win the fast time control engine vs engine tournaments.

Nevertheless, there was quite a bit of discussion there, at that time, about ways to incorporate knowledge into an engine since doing so was a relatively new idea at that time. I think that discussion subsided as the programmers eventually learned how to do it and just how much was practical in a competitive engine.

If the time controls were changed to correspondence chess time limits, then the tradeoffs between amount of knowledge used versus the time required to do so would be "a whole new ballgame." IMHO.

> I still believe the top corr players that have low draw percentage achieve so with high centaur interaction, and human input.


My intuition tells me that you are right about that. I just don't have the concrete facts to support my intuition on this. It seems that there is not much data and how would you measure it?

Bob Durrett
Parent - - By Uly (Gold) Date 2012-08-16 23:26

> If the time controls were changed to correspondence chess time limits, then the tradeoffs between amount of knowledge used versus the time required to do so would be "a whole new ballgame." IMHO.


Agreed, but the opinions that matters are those of the programmers, the people that could make an engine optimal for correspondence chess have to believe in it. Also, I think this isn't only about knowledge, that only effects evaluation and I think the main issue here is that engines search poorly when they have too much time to analyze.

If there's a deep tactic that leaves you losing to a king attack 12 moves down the line, no knowledge in the world is going to help, you need to do your search, and you are doing something wrong if this variation is being pruned at depth 24.

>It seems that there is not much data and how would you measure it?


I think it's easy to measure, you just need to get in contact with such top corr chess players and convince them to run a test with them. Correspondence time controls, the best they got against unassisted engines. If the unassisted engines hold their own then that's your proof that all human input at this level is just an illusion and that engines are already optimal, with the great performance of the players explained because their common opposition was just weaker than unassisted engines.

I'd expect the centaur to crush the engines because of all their weaknesses talked about in these threads and elsewhere, though.
Parent - - By Bob Durrett (***) Date 2012-08-17 01:47

> I think the main issue here is that engines search poorly when they have too much time to analyze.


That is an interesting idea. But . . . don't sell the programmers short. There are some very smart chess programmers out there that can "do anything" almost if they are sufficiently motivated. They are not gods, but they are very good at what they do. (Some of them.)

What we "need" to do is to realistically estimate the amount of time that would be available to a correspondence chess engine. For example, is seems reasonable to assume that different humans do things differently (this is another "basic principle").

Do all engine-assisted human correspondence chess players use their chess engines the same way? For someone who is constantly jumping around in the position, the current "fast" chess engines may indeed be optimal.

On the other hand, for the person that lets the engine "think" for hours at a time, then the current engines might not be the best choice (unless, as it is today, there is no choice).

> you just need to get in contact with such top corr chess players and convince them to run a test with them


is that all?  :-)

In summary: How long does a top correspondence chess centaur allow the engines to think before the position currently being analyzed is replaced with a new one?

Bob Durrett
Parent - - By Uly (Gold) Date 2012-08-20 22:24

> How long does a top correspondence chess centaur allow the engines to think before the position currently being analyzed is replaced with a new one?


Well, you need to get in contact with them and ask them.

In my case, I normally don't let the engines on a position for more than one minute, as I think their problems start as soon as that and the human can already decide on a better way to use the next minute around. The human also needs more time to grasp the analyzed variations, the main reason I don't go for 30 seconds per position is that that would be twice as much what I need to grasp, and I'd need it to grasp it twice as fast, so one minute seems to be a good amount because going faster gives too many positions, but probably stronger players could even look at more positions.
Parent - By Bob Durrett (***) Date 2012-08-20 23:31
Clearly, then, the current engines are best for you.

Bob Durrett
Parent - - By leavenfish (***) Date 2012-08-09 00:03 Edited 2012-08-09 00:05
Bob: "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."

This caught my eye.
I want to relay a story that I think helps illustrate where I am comming from with my previous comments.

A few years ago, Scottish IM (maybe he had recently made GM, I do not remember) Johnathan Rowson played a match with GM Michael Adams. Adams pretty well had is way with Rowson. After the match the thing that stuck out most to Rowson from their post-mortems was just how...little Adams saw. While Rowson was calculating many variations and doing so to considerable length, Adams studied the positions, picked maybe just 3 or 4 lines and calculated just 3 or 4 (I don't recall right now...but its was very little) moves deep.

What I am getting at is that the better a player you are, the less you calculate and the more your knowledge of the game and ability to evaluate positions correctly leads you thru a game. That is why I say that someone who is actually very good at chess (as opposed to 'very good' and guiding their computer slaves round) will use an engine mostly to check lines for tactical flaws and he would largely avoid most suggestions from an engine. Most are not worth the effort to consider which is why I think your "at least ten alternatives" remark is not accurate...well, for someone who does understand the game well as opposed to someone who may not and relies on an engine and largely give a 'yea' or a 'nay' to its lines.

A little data on myself. I've been playing tournaments for nealy 30 yrs. I quit correspondence play when my ICCF rating was at 2398...in good part because it was less a game of man vs man and more a game of man vs computer engines and giant databases and frankly, it did make it harder to win when you could be pretty sure you were playing a number of 'B' level players guiding around engines. I've toyed with the idea of playing more ICCF tournament...they pretty well encourage computer aided play these days. I might do it since there is really no other game out there save for over the board. My OTB rating has fallen...and fallen...and fallen all the way to about 2050 USCF. I only play 1 or 2 tournaments a year and of course can not exactly stay 'tournament tuff' with that, but I enjoy it when I have the time.
Parent - By Bob Durrett (***) Date 2012-08-09 00:18
You just posted a great bulletin! I love it when the human side is given, even though this is a "Rybka" forum and rybka is not human. :-)

There are many reasons why people's ratings decline as people get older. Perhaps one of the more common reason is that one's priorities change.

That has certainly happened to me.

As I said before, "Chess is a great game!" There are many variants of chess and "one man's meat is another's poison" but it should never be necessary to give up all forms of chess.

Incidentally, I have heard the Rowson/Adams story before but perhaps with different characters.

Sincerely,

Bob Durrett
Parent - - By Kappatoo (*****) Date 2012-08-09 18:07

> What I am getting at is that the better a player you are, the less you calculate


This is certainly not true. Just look at press conferences with players such as Anand, Morozevich, Ivanchuk, Kasparov, Kramnik, ... They reproduce variations which are sufficient to fill books.

There are a few things to note here:
The first point concerns style. Adams is a brilliant positional player who relies much more on intuition than on calculation. (This does not mean, however, that he is not good at calculating! It is all relative to other top players. I would bet that he calculates better than an average GM.)
Recently, I saw a post-mortem analysis of Morozevich and Carlsen after one of their games last year in Biel. It seemed that Carlsen had hardly calculated anything during the game (which is most likely not really so, cf. below), while Morozevich kept rattling off very long and complicated lines.

Secondly, it seems to me that there is such a thing as 'post-game amnesia'. After a game, the tension drops off, you are tired and so on. It often happens that you can no longer remember variations even though you are sure that you did calculate them during the game.
(There is also 'post game pseudo amnesia: Players sometimes don't want to talk too much about what they calculated, for some reason or other, and thus confine themselves to general considerations.)

Finally, what I think is absolutely correcy about your observation that being able to know when to calculate and which lines to calculate is an immensely important skill. (I guess every chess player has made the experience of ignoring a move or rejecting it immediately, and then to realize very quickly that it is terribly strong after one's opponent has played it.)

Oh, and I also agree that in most positions, there really aren't ten candidate moves.
Parent - - By leavenfish (***) Date 2012-08-10 00:42
As you agree with most of my observations so let me just address the one where I said "What I am getting at is the better a player you are, the less you calculate".

As a rule, I will stand by that. I would agree however that as a blanket statement it is not always going to be the case from game to game - there are simply times when one MUST calculate. I certainly would not deny that...I see it enough in my own games! You may be refering to the game where Carlsen's King went for a walk on move 11. Yes, you do have to calculate rather a lot when your king gets exposed that early.  However, in the considerably majority of positions that arise during a game, a good player will get FAR MORE milage out of being able to evaulate a few short lines properly than if he indulges in lots of 'brute force searching' because he doesn't 'undertand' what is most important in a position.

In any case ,GM Andrew Soltis in his book Studying Chess Made Easy, devotes a chapter to this aspect of the game. The title of the chapter is : Two-and-a-half move chess. In it he maintains that "calculating more than 2 1/2 moves ahead becomes important only when the position is very tactical".. He notes that "NO one knows what the precise relationship is between the ability to look ahead and overall playing strength .But research on chess-playing computers gives us a rough idea. When the early program Belle could see 4 full moves into the future it was playing at master strength.".

One other observation of his: "The player who can see 5 moves a head - but misjudges the positions he has visualized - will lose over and over to a player who can only see one and a half move ahead but evaluates correctly".. It is just simply true.

By the way, the first example he gives is Dominguez - Morozevich, Wijk ann Zee 2009 where Moro lost...to a 2 move tactic.
Parent - - By Barnard (Bronze) Date 2012-08-10 01:03
well,i know a game that Anand lost in 6 or 8 moves if im not wrong,he simply followed a game,without evaluating it,and he was lost after that moves

since i never took a look at that game with Moro,can you tell me at what moves he made that blunder?
Parent - - By leavenfish (***) Date 2012-08-10 03:12
Why do you automatically assume it was a blunder?

http://www.youtube.com/watch?v=XfqHWEqNmfA

Aside from the flurry around move 15 in a line that was not played....it's mostly short lines being evaulated.

Actually, it occurs to me you are not talking about the King walk in Carlsen - Moroszevich Biel 2011, but instead the game in Soltis's book:

http://www.chessbase.com/news/2009/wijk/games/giddins05.htm

Snatching the g5 pawn led to a 2 move tactic, if that is what you are asking about.
Parent - By Barnard (Bronze) Date 2012-08-10 16:04

>where Moro lost...to a 2 move tactic


lossing in a 2 move tactic,is a blunder
Parent - - By Kappatoo (*****) Date 2012-08-10 08:29
I still don't buy this. I can see in my own games that it is critically important to calculate a lot. I looked at games with many very strong players. I never tried to study this systematically, but it seems clear to me that the stronger they are, the more they see during a game.
Just recently, Gata Kamsky said that he is very dissatisfied with his game in recent years because he has become lazy: He often relies on general considerations instead of calculating, as a result of which his playing strength suffered.
Chess is a very concrete game. I don't know what you mean by 'very tactical', but most positions are such that calculating helps. Otherwise, computers wouldn't be so strong.
I don't know why you mention Belle here. It is clear that humans will never be able to calculate that many moves.
Parent - - By leavenfish (***) Date 2012-08-10 12:05
Lets define our terms then. What to you mean by 'a lot' as in "it is critically important to calculate a lot" ?

If you don't understand why I mention Soltis's reference to how many moves Belle calculated (FOUR MOVES...I don't know why you consider that humans 'will never be able to calculate that many moves', re-read the post. It is clear why he mentioned it.
Parent - - By Kappatoo (*****) Date 2012-08-10 12:45
Define? In which unit?

Well, of course humans can calculate four moves. But I would guess Belle calculated all, or nearly all, variations up to a depth of four moves. This is A LOT. And it is something humans are not able to do.
Parent - By leavenfish (***) Date 2012-08-11 03:56
"In which unit"???? You are the one who said "it is critically important to calculate a lot". So, tell me how many moves you think a GM considers in an average position and how deep on average he anaylses those lines. Come on, give it a stab.
Parent - - By Regularuser (***) Date 2012-08-10 09:44 Edited 2012-08-10 09:47
It is simply untrue in my experience that strong players calculate less.

It is true that strong players need to calculate less than weaker players in the sense that they can use their positional understanding to evaluate some positions where a weaker player would have to analyse further to know who stood better.    But that does not mean that they actually calculate less - they focus their calculation in the critical areas and in practice analyse a lot more.    Strong players are typically better at positional understanding and calculation, and in my experience they use both. 

I know it is written in many books that in certain positions a GM would not not have calculated much and that might be true in those specific positions, but don't be deceived by this.   I have played several strong IMs and a few GMs over the years and done post mortems with them, and trust me they calculate plenty through all phases of the game.  It is not necessarilly about calculating big combinations or complicated tactics (though they do that well too), it is often about looking after all the little details which they do so thoroughly and accurately.

Watch the videos of a Kramnik post-mortem to see how much he calculates.   It always takes my breath away about how many lines he has looked at and how quickly he can rattle them off.

As for, "calculating more than 2 1/2 moves ahead becomes important only when the position is very tactical", this is simply nonsense, probably written to sell a book - after all the idea is very appealing to a certain class of chess player.
Parent - - By leavenfish (***) Date 2012-08-10 12:12
Then again, lets define our terms.
For one, I do not say they calculate "less"...less than....who?

"trust me they calculate plenty through all phases of the game"
and roughly how many moves on average would that be. With the exception of positions that simply DEMAND a lot of calculation, just how many moves deep on your average position do you think this is and how many moves do you think they actually bother to analyse?
Parent - By Kappatoo (*****) Date 2012-08-10 12:51
I think it is almost impossible to give a reasonable answer to these questions. At least, it would take serious empirical research. I can give you an estimate concerning the maximum depth calculated by strong humans. But the average? Sometimes, in fact often, a couple of plies is sufficient. On other occasions, there are variations which are 20 or more plies deep.

Concerning 'less': You said that stronger players calculate less. When I look at post mortems of Kramnik, Ivanchuk, Morozevich ... it is obvious that they calculate many times as many lines as I do. And I think I calculate comparably much for a player of my strength.
Parent - - By Regularuser (***) Date 2012-08-10 14:12
+1 on the points that Kappatoo made.

Like Kappatoo, I can't give a meaningful answer to average depth but wil say the following:

- if they haven't worked everything out in advance or are occasionally making a pure positional move, IMs / GMs will look a lot more than 2.5 moves ahead (i.e on almost every move)

- "Demand" is a difficult term in this context.   Maybe you are thinking of overtly tactical positions where everyone would agree there is no choice but to calculate.   But there are other positions where it appears to be optional, but calculation is necessary to avoid/gain a meaningful shift in the position (say over 0.25 in comp terms) , or even squeeze a win or avoid a loss - this is actually a big proportion of the positions encountered in a game.   

- Whilst this point about "demand" is true in the middlegame, it shows up even more in the endgame.   Look at the rook endings as a specific example- many amateurs would not calculate much in many rook endings because in their eyes the position does not demand it, and/or they are simply not practiced at calculating in endgames - accordingly they play the position largely on general principles.   GMs know a lot more about them than amateurs so can avoid some calculation, but I can tell you from personal experience that GMs calculate a lot more than amateurs in endings.  For his level, GM Keith Arkell is considered an expert at rook endings, and he definitely calculates an awful lot in endings.

- it is not just a question of depth, it is a question of breadth too: depth alone is a poor indicator.   Accepting humans prune more and better than computers, and that in many postitions there will be only 2 or 3 good candidate moves, one thing I think distinguishes strong players is that they prune less severely than weaker players further down the calculation tree - they can look ahead to see move that a weaker player will find surprising in advance but may spot quite easily once they arrive at the position.   They also calculate with fewer errors than weaker players.

- to illustrate the point of depth/breadth, the longest line I have accurately calculated OTB in a middlegame is over 20 moves (i.e over 40 plies) but apart form spotting a couple of sacs it was an easy line to calculate because it was largely forcing (i.e. the branching was low).   That may be comparable to the max depth of many GMs for a middlegame but proves nothing.   But I definitely don't calculate as much or as well as any GM I have ever met (sadly!).

- I have never analysed with Michael Adams.   When he was still an IM, I have stood at the side of his board watching and listening whislt he did a post-mortem with his opponent, with Danny King and a few others joining in.   He obviously calculated a lot in those days.

But from your comments you must have been a reasonably good OTB player.   Did you ever play a strong IM or GM and what was your experience in doing the post-morten with them?
Parent - - By leavenfish (***) Date 2012-08-11 04:35 Edited 2012-08-11 04:38
"if they haven't worked everything out in advance or are occasionally making a pure positional move, IMs / GMs will look a lot more than 2.5 moves ahead (i.e on almost every move)".

Don't introduce memorized opening variations, that's a different animal, nothing to calculate anyway. And purely positional moves count just as much as tactical ones...but we could omit both for sake of agrument and consider the rest.

Neither of you seem willing to want to try quantifying your ascertions that your average GM calculates a lot more than I tend to believe. Now, I am not saying Soltis's 2 1/2 moves is gospel, probably closer to 3 1/2 IMO...actually, if I've mistated it, I don't think he actually says a GM does that on average. Let me clarify what he says:

"You already know that when you consider a candidate move you should try to forsee your opponent's best reply. This means seeing one full move into the future. The 2 1/2 - move guideline means that in the majority of cases you can confirm that the candidate is good by seeing no more than another one and a half moves beyond that" pg 124 I do maintain that this is going to be far closer to the truth than what you seem to think. A GM probably spends more time (barring time pressure) simply studying the position, lifting a very small number of candidates and then looking 2 1/2 to 3 1/2 avg depth...not more as a rule, unless the position simply demands it. Calculation is an error prone exercise. It's more important to analyse short lines and make a proper assessment rather than longer lines where you could miss more. Why do you think GM's play bullet or blitz SO GOOD...they certainly aren't calculating deep. They just pick the right things to caluculate, make short calculations and evaluate the lines.

Yes, I've played one GM multiple times (no wins, no draws...just some close ones) and several IM's, 1/2 out of the last 2. In that game where we split the point, I calculated only very short lines and doubt he did much different.

Frankly, they usually speak in generalities when discussing the games.

Here is the game while I'm thinking of it. The pentultimate round and I wager that he offered the draw because...well, he is worse, had been effortlessly outplayed when he went to too great a legnth to trade some key pieces, got his queen misplaced...and just did not see a favorable trend likely to arrise after I clamped down on his weakness. My only real calcuation came at move 11 and 12...and that wasn't deep at all.

Actually, he told a friend who was incredulous that he would offer the draw, that he felt he could win no more than 25% of the time from the final position if he had just played to 'let his rating points speak' and continue.

[Event "19th Battle of Murfreesboro"]
[Date "2011.08.20"]
[Round "4"]
[White "Me"]
[Black "IM"]
[ECO "A47"]
[WhiteElo "2026"]
[BlackElo "2416"]
[PlyCount "38"]
[EventDate "2011.??.??"]
[SourceDate "2012.02.20"]

1. Nf3 Nf6 2. d4 e6 3. e3 b6 4. Bd3 Be7 5. O-O c5 6. c3 Ba6 7. Re1 Bxd3 8. Qxd3
Qc8 9. e4 Qa6 10. Qd1 O-O 11. d5 d6 12. dxe6 fxe6 13. e5 dxe5 14. Nxe5 Qc8 15.
Qe2 Na6 16. Bg5 Nc7 17. Nd2 Qe8 18. c4 Rd8 19. Rad1 Rd6 1/2-1/2
Parent - - By Regularuser (***) Date 2012-08-12 08:23 Edited 2012-08-12 08:34

>Don't introduce memorized opening variations, that's a different animal, nothing to calculate anyway.


I wasn't taking about memorised opening, I was talking about things calculated previously in the game that the player feels no need to recheck.

>Neither of you seem willing to want to try quantifying your ascertions that your average GM calculates a lot more than I tend to believe.


I could not even take a stab at how much I calculate on average without going through a load of my games move by move.  But I don't need to know how much a GM calculates and how how much I calculate to know that a GM calculates more.   Let me give you an analogy.   Two people have a pile of rice each.   We can all see which pile is bigger.   We don't have to be able to say how many grains of rice  is in each to know that one is bigger than the other.    When I have done post-mortems with strong players, I know they have typically calculated more than me, even if I cannot tell you my average depth and width and their average depth and width.

>I do maintain that this is going to be far closer to the truth than what you seem to think. A GM probably spends more time (barring time pressure) simply studying the position, lifting a very small number of candidates and then looking 2 1/2 to 3 1/2 avg depth...not more as a rule, unless the position simply demands it. Calculation is an error prone exercise. It's more important to analyse short lines and make a proper assessment rather than longer lines where you could miss more. Why do you think GM's play bullet or blitz SO GOOD...they certainly aren't calculating deep. They just pick the right things to caluculate, make short calculations and evaluate the lines.


I think there is some truth in this and I previously agreed that GM's will prune out some nonsense lines that weaker players will look at.   But my contentions are that:
- they don't spend long on the general study of the position before moving onto the calucation phase (and then iterating back to study after calculation)
- they calculate more short lines than weaker players (looking after the details)
- when they need to go deep they do so, and on average they achieve more depth than amateur players.
Actually how well they look after the tactics in blitz games shows how much they calculate at quick speed, and if they do this at blitz you can bet they are doing it many times over normal time controls.

>Frankly, they usually speak in generalities when discussing the games.


That would be my experience if its a 3 minute chat at the end of the game, but not my experience if it is a 30min post-mortem.

>Here is the game while I'm thinking of it. The pentultimate round and I wager that he offered the draw because...well, he is worse, had been effortlessly outplayed when he went to too great a legnth to trade some key pieces, got his queen misplaced...and just did not see a favorable trend likely to arrise after I clamped down on his weakness. My only real calcuation came at move 11 and 12...and that wasn't deep at all.


Firstly, congrats on holding someone of this strength to a draw.   And yes I can see that you did not need to calculate much to get to this position.   But your opponent seems to have played very weakly for a 2400:  he seems to have calculated next to nothing hence getting a realy bad position so quickly, so I am not sure what conclusion we can draw from this.   As an aside, I am surprised you took the draw - I would have played on for sure.
Parent - By leavenfish (***) Date 2012-08-12 15:47 Edited 2012-08-12 16:54
Post mortems are usually filled with players searching for the truth or some approximation of it...considering lines...it's not an exhibition of "I calculated all this, so perpare to be amazed..." It's more a 'well, you could have tried...x and lets see...y and z could result and I think I could have would up with a good kingside attack because your last move would have removed your knight from f6'. Things like that. Or 'well, I was thinking about maybe Rb1 with the idea of b4 and a minority attack'...and the opponent responds, ' yes...that would be an alternative. If you were to do that, I would probably play ...f5 with the idea of f4, hitting your f2-e3-d4 pawn chain', etc...and, what is that...it is really an anaylsis session about things that never happened, but could have. And when it comes to lines in a book or a video...well, the games have been looked at since then, checked by computers and prepared for presentation. As I think Soltis or someone else I was recently reading said, GM's often do like to 'show off'...and that makes sense if only because they want to present a good product, with lines that might dazzle the reader/viewer. I think that's why Kasparovs videos and books go in for longer lines (plus, like Alekhine, he does like to show off...and Alekhine was even caught inventing the course of games to make himself look good....), while Karpovs...well, usually very sparse annotations, more 'to the point' lines and verbage. Kaspy tries to explain his ideas in variations, Karpov (like say, Capablanca), presents the essential ideas and generally short lines.

"I think there is some truth in this and I previously agreed that GM's will prune out some nonsense lines that weaker players will look at."  

I am glad we agree there. They therefore do not calculate those lines...which weaker players will calculate, move after move during a game. That adds up....

"But my contentions are that:
- they don't spend long on the general study of the position before moving onto the calucation phase (and then iterating back to study after calculation)
"

   Each move does have the potential to change the character of the position...new possibilities, things evolve. You will often see them consider moves a weaker player would never consider...and it's usually because they thought about the position more before moving along to calculation. It is vitally important to discern what you need to consider. About every 3 moves a good player is going to draw up a 'to do list' based on these new positions. There is a lot more to do that just start calculating after a cursory glance at the position.

"- they calculate more short lines than weaker players (looking after the details)"

agreed...about the more short lines part. The weaker player will extend those lines out more, if only because weaker players 'can not make a proper judgement' after just looking short and have feel they need to 'go further' to dry and come up with something more concrete...a win of a pawn, a gashing of the pawn structure around the king. That is kind of analagous to computer engines, isn't it? Their ability to judge a positon correctly after just a few moves is poor...so they go deeper. A better player will often know intuitively that a line he has looked at a few moves deep are likely to result and b'good things' are likely to await, and head that way. Conversely, a line he has looked at 2 or 3 moves deep and thinks 'this is getting messy' or 'I don't like moving that piece away from my king', he would abandon, unlike a weaker player who will calculate deeper to try and reach a conclusion. That's where mastery of the game comes from.

"- when they need to go deep they do so, and on average they achieve more depth than amateur players."

Agreed as well. As I opined, they do so really only when the position demands it...of course, he may find himselves in situations with lots of time and might calculate some largely extraneous lines to verify his thoughts. I don't deny that. You know, it probably has happened more and more as we get into more modern times where 'home study' can end at move 20 (especially in super gm play where they have prepared for certain opponents they play several times a year, but that's an exception that proves my rule) and in the past when the game was less played out, GM's found new positions arise MUCH earlier in the game. The then had to fit more of the 'real game' where there are more options, into less time and again relied less on caculating 'long' lines.

"Actually how well they look after the tactics in blitz games shows how much they calculate at quick speed, and if they do this at blitz you can bet they are doing it many times over normal time controls."

To some degree, I can agree with you there, but again, mostly only when the postion demands long lines. But I always seem to amaze lower rated players when they watch my blitz games. Mostly I am checking very short lines...the rest, is on intuition (I've seen similar situations before, this is what you do here, etc) or purely positional grounds (the knight needs to go to d4 and the bishop if attacked should fall back along the f1-a6 diagonal, not the b1-h7 diagonal where it would be out of play), etc. I must think it a simple extrapolation that GM's do that as well...only much, much better. I see lots of 1 min games on ICC...some might be turkeys and at the end when there is no time, blunders can happen, but many most all agames among the best players are really better than anything an expert is likely to play given much more time. There simply is no time for calculating long variations. It is then very short variations of multiples candidates  - and I am sure the number of candidates shrink as the time limits go down - and lots of intuitive play that results in what these GM's actually end up playing.

I agree with you on my game....but I took the draw because I've only drawn this guy once before (several losses, some embarrasing!) and it was the pentultimate round. 3 1/2 out of 4, and getting an easier oppoenent in the last round was just too appealing. That and the fact that I don't play but a tourney or 2 a year and he plays almost very week...so if tactics arose, I would not be terribly sharp, feel uncertain and more likely to screw up... I'm not sure there was a lot to calculate on either side so I agree with you there. I do NOT think his calculation was at all bad and for the very reason you agreed with - there wasn't much to calculate. No, he aimed for good exchanges, but misjudged what happens if I decline (ex, the queen trade, which left it out of play when he got behind in development...he loves to simply and outplay in the endgame, and is quite good at that).
Parent - By leavenfish (***) Date 2012-08-11 04:49
Regularuser: As for, "calculating more than 2 1/2 moves ahead becomes important only when the position is very tactical", this is simply nonsense, probably written to sell a book - after all the idea is very appealing to a certain class of chess player.

You really think that if the position isn't tactical...pretty calm, that you need to caluclate a lot more than that?? I think it much more beneficial to carefully study the position, pick a few candidate moves and do short analysis to see which offers the best way to proceed. We may just have to agree to disagree on this I am afraid.

I think John Nunn (no calculating slouch...of course, he played sharp agressive lines) had it right when he says: Don't Anaylse Unnecesary Tacits (DAUT). It's an error prone enterprise. Avoid long lines unless they are absolutely necessary and learn to evaluate positions better.
Parent - - By Regularuser (***) Date 2012-08-12 07:06
A belated reply.....    I think this is probably true.

Permanent Brain makes the point further on about sacrificing depth to get more options looked at along the way.   Of course it is a valid point, so there is a trade off.   The trade off will have been optimised for fast games (this is how developers test) but I suspect that this is not ideal for correspondence play based upon long searches.   The long searches bit is important, as you don't have to use those to play correspondence chess, but I suspect that for optimal results it would be wise to do so at least occasionally.  Of course it would take a long time to do a meaningful test to prove/disprove this opinion.

Zappa seems to prune much less that other engines.   With Zappa it is defintely better to give it a long time to search.

What would be really good would be to have a UCI parameter that influences pruning, so that with any engine we could determine the depth/width trade-off.
Parent - - By Bob Durrett (***) Date 2012-08-12 09:18

>> What would be really good would be to have a UCI parameter that influences pruning, so that with any engine we could determine the depth/width trade-off.


Yes! That is really a creative idea!  Congratulations. :-)

Create one chess software (engine plus GUI) which is adjustable so as to make it best for whatever application the user wishes at the time he/she is using it.

It would be good if this adjustment would be very easy so the user could use it for different applications is a single sitting.

:-)

Bob Durrett
Parent - By Uly (Gold) Date 2012-08-12 10:56

> Create one chess software (engine plus GUI)


Still waiting for a GUI that is optimized for analysis, it seems it would be much easier to create than an engine optimized for corr chess :smile:
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