Winboard Forum

Hi all ,

we discussed adding a Blitz rating list to CEGT but could not really come to an agreement.

Considering that we have a lot of engines in that are supporting only Winboard time controls I proposed something like 40/4 repeated, but other testers would prefer 5+0 or even 6+4 or more, what in my opinion then would not be comparable to our usual time scheme (40/40 adapted).

So my question to authors and other testers would be:

1. Should we add such a rating list in the future or is Uri the only one interested in this? Of course such Blitz games would be run on only two or three computers in order of not reducing the amount of games with our usual time control 40/40 repeated too much.

2. Are there any new/additional testers willing to help us with a bigger Blitz rating list here?

3. Are there engines in the CEGT list not supporting increment?

4. What other advantages/disadvantages do you see for time controls x in y repeated, x+y and x+0?

5. What really is considered a Blitz game when we adapt to 2 Ghz? Is for example 40/8 repeated still Blitz or rapid (active) chess?

6. Would it be preferable to have games for many engines with faster Blitz 40/4 or to test only 20 or 30 engines with longer Blitz time control?

Advice as always is welcome and maybe would help us to come to an agreement.

Best Regards
Heinz

Hi all ,

we discussed adding a Blitz rating list to CEGT but could not really come to an agreement.

Considering that we have a lot of engines in that are supporting only Winboard time controls I proposed something like 40/4 repeated, but other testers would prefer 5+0 or even 6+4 or more, what in my opinion then would not be comparable to our usual time scheme (40/40 adapted).

So my question to authors and other testers would be:

1. Should we add such a rating list in the future or is Uri the only one interested in this? Of course such Blitz games would be run on only two or three computers in order of not reducing the amount of games with our usual time control 40/40 repeated too much.

2. Are there any new/additional testers willing to help us with a bigger Blitz rating list here?

3. Are there engines in the CEGT list not supporting increment?

4. What other advantages/disadvantages do you see for time controls x in y repeated, x+y and x+0?

5. What really is considered a Blitz game when we adapt to 2 Ghz? Is for example 40/8 repeated still Blitz or rapid (active) chess?

6. Would it be preferable to have games for many engines with faster Blitz 40/4 or to test only 20 or 30 engines with longer Blitz time control?

Hi Heinz,

I can give you my opinion about 4. and 5.
I believe anything except x+0 is fine, if one realizes that it just
gives an idea about the used exactly time control and not more.
IMHO results for x+0 are just random noise, as the games can
turn at any point after move 50-60, because of extreme time trouble.
Some may say it will still give something valuable for exactly
the given tc, but I doubt it for x+0(any +-/5 advantage can turn
into the opposite like a tossing shell in the sea).

I am doing a test currently and it is shocking how different results
are between e.g. 40/1 40/2 40/4 and 40/8.
(If I had more CPU I would also play 40/16 40/32 and 40/64)
Even same positions produce very different results over a big set of games.
Often a program even chooses a more successfull move at lower depth -
it's really horrible...

For 5. according to FIDE rules anything below 15min per side is still
considered as Blitz, thus 40/8 is not yet rapid chess, if I assume
an average game length of 65-70 moves.

Best regards,
Guenther

Hi Heinz!

Heinz van Kempen wrote:So my question to authors and other testers would be:

1. Should we add such a rating list in the future or is Uri the only one interested in this?

I think it is a great idea, if it doesn't take too much time and resources from the testing at slow time controls.

Like Guenther, I hate sudden death time controls without increments. x + y (with y > 0) and x moves / y minutes are both OK to me, although I slightly prefer the former.

Tord

Heinz van Kempen wrote:1. Should we add such a rating list in the future or is Uri the only one interested in this? Of course such Blitz games would be run on only two or three computers in order of not reducing the amount of games with our usual time control 40/40 repeated too much.

It will be fantastic. Anyway, AEGT/CEGT is the best rating list (IMO).
But do you only want to use engines from CEGT? It will be much better use engines from AEGT also.

Heinz van Kempen wrote:3. Are there engines in the CEGT list not supporting increment?

I know 3 engines that don't support increment: Amayan 1.595, Nejmet 3.07, Dragon 4.6.

Heinz van Kempen wrote:4. What other advantages/disadvantages do you see for time controls x in y repeated, x+y and x+0?

x in y too ugly (it is only my taste). x + y is nice with y may be zero. I don't see disadvantages of x + 0. If some engines have poor time handling it is not a blow of time control.

Heinz van Kempen wrote:5. What really is considered a Blitz game when we adapt to 2 Ghz? Is for example 40/8 repeated still Blitz or rapid (active) chess?

Words 'blitz', 'active' and 'long' have not any sense for computer chess. Only words: "faster" - "slower", "shorter" - "longer".
For the sake of truth, in my opinion, time contrlol "one move per year" is still extremly bullet for modern PC, because engines will do a lot of stupid mistakes.
IMO, long enough is time control that allowed TSCP be at the same level (or higher) than Shredder. Because, I am sure if given enough time TSCP will not made any mistakes.

Heinz van Kempen wrote:6. Would it be preferable to have games for many engines with faster Blitz 40/4 or to test only 20 or 30 engines with longer Blitz time control?

More games, more engines => shorter time control.

Guenther Simon wrote:I can give you my opinion about 4. and 5.
I believe anything except x+0 is fine, if one realizes that it just
gives an idea about the used exactly time control and not more.
IMHO results for x+0 are just random noise, as the games can
turn at any point after move 50-60, because of extreme time trouble.
Some may say it will still give something valuable for exactly
the given tc, but I doubt it for x+0(any +-/5 advantage can turn
into the opposite like a tossing shell in the sea).

I don't saw big random noise with x + 0. The same noise as for longer games x + y or x in y.
And with current hardware games can turn at any point after start of game, because of extreme time trouble, even if SSDF time control used.

Guenther Simon wrote:I am doing a test currently and it is shocking how different results
are between e.g. 40/1 40/2 40/4 and 40/8.
(If I had more CPU I would also play 40/16 40/32 and 40/64)
Even same positions produce very different results over a big set of games.

Very interesting to see your testing.
My test shows only small difference and only for some engines it is noticable.

Hi,

I would find a blitz rating list interesting too. I'm curious to see whether there are differences between different time controls. To make it "real blitz" and to have a real difference between the list and to add more engines I opt for 4+2 perhaps 4+1. I think increment is good to avoid a lot of lost on time and strange endgame blunders. If there are a few engines which does not support increment skip them.

regards Joachim

Hi Heinz,

Heinz van Kempen wrote:1. Should we add such a rating list in the future or is Uri the only one interested in this? Of course such Blitz games would be run on only two or three computers in order of not reducing the amount of games with our usual time control 40/40 repeated too much.

I think it would be very interesting as it could help answering the old question: "is time control so critical for testing?".

4. What other advantages/disadvantages do you see for time controls x in y repeated, x+y and x+0?

My opinion is that x/y makes no sense mathematically, as it introduces an artificial discontinuity (e.g. why not 39 or 41?). The only reason why it's used is that mechanical clocks do not handle increments. I suggest 4+2.

Of course I would understand that you prefer 40/y, for direct comparison.

Fabien.

Hi all ,

thanks all for your comments. I will answer in a more detailed way to all the proposals on Monday evening and just wait if there will be more proposals.

The CEGT team is still fairly small and although most are using more than one computer our main concern is not to neglect our 40/40 main project.
So if no new testers can be gained we will only use two or three computers for Blitz. To have a lot of games for a lot of engines over the months anyway it would be better to play faster Blitz like 4+2 or 40/4, but exactly here those testers interested in running Blitz tournaments and having the machines available do not agree and want more time even for Blitz. Hope there will be either a change in opinion or additional testers.

More tomorrow evening.

Best Regards
Heinz

Igor Korshunov wrote:
IMO, long enough is time control that allowed TSCP be at the same level (or higher) than Shredder. Because, I am sure if given enough time TSCP will not made any mistakes.

Tscp cannot search more than 32 plies even in the longest line so no
time control that is long enough will help it to get shredder's level

Fabien Letouzey wrote:Hi Heinz,

Heinz van Kempen wrote:1. Should we add such a rating list in the future or is Uri the only one interested in this? Of course such Blitz games would be run on only two or three computers in order of not reducing the amount of games with our usual time control 40/40 repeated too much.

I think it would be very interesting as it could help answering the old question: "is time control so critical for testing?".

4. What other advantages/disadvantages do you see for time controls x in y repeated, x+y and x+0?

My opinion is that x/y makes no sense mathematically, as it introduces an artificial discontinuity (e.g. why not 39 or 41?). The only reason why it's used is that mechanical clocks do not handle increments. I suggest 4+2.
Of course I would understand that you prefer 40/y, for direct comparison.

Fabien.

I see nothing wrong mathematically with x/y

basically fischer time control is also x/y with y=1

I see no mathematical reason why y=1 is better than y=40
I think that better time management in x/y games is a mathematical problem that most programmers even did not try to handle and if I choose y=100 most programs will play faster relative to x minutes for all the game.

In Movei I did something about this so you will not see this stupid behaviour.

Uri

Uri Blass wrote:Tscp cannot search more than 32 plies even in the longest line so no
time control that is long enough will help it to get shredder's level

Do you sure that 32 plies without unsound pruning (like others programs doing) is not enough to play perfect chess? I am not.
But, anyway it is not a problem, because you can easily change this constant to 1024

Uri Blass wrote:basically fischer time control is also x/y with y=1

I can't understand this.
Are you talking about fisher control 0 + y ? I don't like such controls.
I prefer bigger x and smaller y.

Heinz van Kempen wrote:The CEGT team is still fairly small and although most are using more than one computer our main concern is not to neglect our 40/40 main project.
So if no new testers can be gained we will only use two or three computers for Blitz.

Hello Heinz,
I am also interested in blitz games, especially to find out the relationship with games played at long time controls. At present I have a machine that could be used for testing, but I don't know for how long it will be available...

Hi Uri,

Uri Blass wrote:I see nothing wrong mathematically with x/y, basically fischer time control is also x/y with y=1. I see no mathematical reason why y=1 is better than y=40

Wow, I would never have expected that from you!
You should be teaching me ...

It's true that the *debugged* version of increments (adding increment BEFORE the move of course, not after it; is this bug from Fischer?) is equivallent to 1/y + main time.

But yes, 1 is better than 40 (and any other value) and leaves no discontinuity. E.g. with 40/y, constraints for playing move 40 and move 41 are completely different. Just an example, there should be no need for me to explain!

Fabien.

Hi Igor,

Igor Korshunov wrote:I prefer bigger x and smaller y.

I agree; x leaves room for smart time management. y should only be large enough to reduce endgame blunders to a minimum.

I think y should be proportional to x to allow comparison with longer games but the question is what should this ratio be. I use y = x/2 only because it allows 2+1 which I use for testing. Most interfaces unfortunately forbid sub-second increments ...

Fabien.

Igor Korshunov wrote:

Uri Blass wrote:basically fischer time control is also x/y with y=1

I can't understand this.
Are you talking about fisher control 0 + y ? I don't like such controls.
I prefer bigger x and smaller y.

You are right that it is not exactly equivalent but basically it is equivalent to type of time control that is used without fischer.

take for example 5 minutes per game+3 second per move time control.

it is equivalent to 300 seconds for the first move+3 seconds for every new move so you have 300 seconds/1 move+3 seconds/1 move

120 minutes/40 moves+60 minutes/20 moves time control was used in the past in human tournaments so having different time control for the first set of moves is not something new.

Fischer time control only means that the first set of moves also have a single move instead of 40 moves when the second time control that repeats itself also have a single move.

Uri

Fabien Letouzey wrote:Hi Uri,

Uri Blass wrote:I see nothing wrong mathematically with x/y, basically fischer time control is also x/y with y=1. I see no mathematical reason why y=1 is better than y=40

Wow, I would never have expected that from you!
You should be teaching me ...

It's true that the *debugged* version of increments (adding increment BEFORE the move of course, not after it; is this bug from Fischer?) is equivallent to 1/y + main time.

But yes, 1 is better than 40 (and any other value) and leaves no discontinuity. E.g. with 40/y, constraints for playing move 40 and move 41 are completely different. Just an example, there should be no need for me to explain!

Fabien.

basically fischer time control is a subset of the possible time control of x1 seconds/y1 moves+x2 seconds/y2 moves.

I understand that there is less work if you only think about optimization of time management for fischer time control but I am not sure if this is a reason to prefer fischer time control.

Note that I have no objection to fischer time control if it is also used in the long games for comparison.

There is a problem to compare engines at different time control if you have 40/40 and 5+2 because I am sure that there are engines that prefer one time control and there are engines that prefer different time control not because of the fact that one is longer but because of time management.

comparison between fischer 2+1 time control and 40/2 time control may be also interesting and I think that the results may help authors to know if they need to improve time management in one of the time controls.

Uri

Igor Korshunov wrote:

Uri Blass wrote:Tscp cannot search more than 32 plies even in the longest line so no
time control that is long enough will help it to get shredder's level

Do you sure that 32 plies without unsound pruning (like others programs doing) is not enough to play perfect chess? I am not.
But, anyway it is not a problem, because you can easily change this constant to 1024

1)Of course if you modify tscp it is strong enough in long time control(you need more modification otherwise it may crash because it assume that the number of moves in the move list is not bigger than some constant that is not big enough when you search 50 plies forward).

2)This long time control has no practical importance because it is so long that you cannot finish a single game in your life(even 1 move per year is not enough)

If there is an evidence that a program is stronger for correspondence games(of course you cannot have enough games to prove it but there can be evidence that may lead me to believe it) then it is interesting because correspondence players may prefer to use it for their games but if some program is stronger at 1000 years per move then this information is uninteresting and with branching factor of 6 even with no pruning program that search 10 plies in 3 minutes will be unable to search deep enough even in 1000 years(it may get only 20-21 plies in 1000 years that is not enough to be close to solve chess).

Uri

Alessandro Scotti wrote:

Heinz van Kempen wrote:The CEGT team is still fairly small and although most are using more than one computer our main concern is not to neglect our 40/40 main project.
So if no new testers can be gained we will only use two or three computers for Blitz.

Hello Heinz,
I am also interested in blitz games, especially to find out the relationship with games played at long time controls. At present I have a machine that could be used for testing, but I don't know for how long it will be available...

Hi Alessandro ,

thanks for your offer. All help is welcome and as well in AEGT as in CEGT there is no obligation to test for a longer time. Just write to hvankempen@web.de and we could agree how to start.

Although there are always problems with different opinions concerning engines to take in and time controls, testing in a team is generally nice with a lot of interchange of ideas. Moreover it is the only way to accumulate an amount of games with better statistical values for a bigger group of engines.

My proposal would be to start with the best amateur engines and then going down in our list and add the commercials later. So for example set up a tournament with the best 24 amateur engines and run it until all have 500 games and later add one by one via gauntlets. All opponents should then be in a range of +- 250 ELO. This would be a systematical approach. Of course there are also other possiblities.

So again: all people who want to help, also if it is only for a few weeks, are highly welcome.

Best Regards
Heinz

Igor Korshunov wrote:

Heinz van Kempen wrote:1. Should we add such a rating list in the future or is Uri the only one interested in this? Of course such Blitz games would be run on only two or three computers in order of not reducing the amount of games with our usual time control 40/40 repeated too much.

It will be fantastic. Anyway, AEGT/CEGT is the best rating list (IMO).
But do you only want to use engines from CEGT? It will be much better use engines from AEGT also.

Hi Igor ,

thanks for the compliment. Anyway in my opinion there is nothing what could not be improved. The main thing is to gain help for such a project and AEGT already showed that this is not so easy, as most tournament organizers already have fine tournaments they want to stick to and giving all the rules yourself is of course easier than agreeing a compromise with others.

I also agree that a Blitz timecontrol (and especially a faster one) would be an opportunity to include more engines being also strong. WildCat 4 for example is already strong enough to compete with the lower end in the CEGT rating list. We only feared that there would be an update just when we started to test it for a few weeks. The next version will be definitely in.
Same for Jonny and some others.

Best Regards
Heinz