Too many crafty posts!

[bob]

Sound familiar?

Sure, it reminds me of the old days when people would get bent out of shape over ChessMaster personality posts.

My personal opinion is that there is no such thing as too many posts on a topical subject. When people are foaming at the mouth over Rybka or Crafty or Junior or Naum or Sjeng or whatever, I think it is fantastic. It means that people are getting a great deal of enjoyment out of it and have passion for the subject. If the subect is not interesting to people, then we won't have lots of posts on that topic.

So if you are not interested in a subect, do what I do. Skip over that post and read the next one that looks interesting.

a version a day keeps the doctor away...

Or makes you pull your hair out. But I think we are about past that stage. The threads seem to be working for everyone, the windows and linux memory leaks (two different issues completely) are history, and it now appears to compile cleanly for everyone so far as I know... Back to a more sedate release schedule.

I might use a "+20 Elo" window from here on, except for bugs. So that once we gain +20 elo over the last 22.n version, we release 22.n+1. That at least gives a target and it usually takes a couple of weeks (at a minimum) to produce that kind of improvement...

Are you holding back any secret versions pending the next CCT or other big tournament?

Not as of yet, but we generally stop releasing 1-2 months prior to an event, and then release the latest right after the event.

[bob]

Bugs like memory leaks should be fixed as soon as possible, certainly. But perhaps you should name bugfix versions differently, because for example there is a big difference between 22.1 and 22.2, but small differences between the more recent versions. Maybe they would be better named 22.2, 22.2b, 22.2c etc, or 22.2, 22.22, 22.23 etc, or something else. So that users would have a better estimation of how different the versions are and how likely it is that their playing strength is also different.
Just my opinion, of course. Besides, on the other hand, maybe this is not so important.

that might not be a bad idea, but I try to limit "bug fix" releases another way, namely by testing carefully. the smpnice=1 is still the issue here as stopping and re-starting the threads leads to the problems with memory leaks of late. Without that (and I don't run like that ever during my testing) none of these problems would have happened...

[bob]

Sound familiar?

Sure, it reminds me of the old days when people would get bent out of shape over ChessMaster personality posts.

My personal opinion is that there is no such thing as too many posts on a topical subject. When people are foaming at the mouth over Rybka or Crafty or Junior or Naum or Sjeng or whatever, I think it is fantastic. It means that people are getting a great deal of enjoyment out of it and have passion for the subject. If the subect is not interesting to people, then we won't have lots of posts on that topic.

So if you are not interested in a subect, do what I do. Skip over that post and read the next one that looks interesting.

a version a day keeps the doctor away...

Or makes you pull your hair out. But I think we are about past that stage. The threads seem to be working for everyone, the windows and linux memory leaks (two different issues completely) are history, and it now appears to compile cleanly for everyone so far as I know... Back to a more sedate release schedule.

I might use a "+20 Elo" window from here on, except for bugs. So that once we gain +20 elo over the last 22.n version, we release 22.n+1. That at least gives a target and it usually takes a couple of weeks (at a minimum) to produce that kind of improvement...

I really like your new testing methodology. I think it will bring you great success to measure everything so carefully.

I personally believe this testing methodology is going to produce more improvement than anything we have done in the past. Now it is easy to verify that _every_ change is for the better, else it can be tossed. Tracy and I do most of the coding changes, and I can't tell you how many ideas we try, ideas which make perfect sense to a chess player, yet the testing proves "this is worse". But every now and then, one of the changes improves play. During the "slack time" when we are working on changes, I can throw in a half-million games to tune a specific evaluation term.

[James Constance]

Sound familiar?

Sure, it reminds me of the old days when people would get bent out of shape over ChessMaster personality posts.

My personal opinion is that there is no such thing as too many posts on a topical subject. When people are foaming at the mouth over Rybka or Crafty or Junior or Naum or Sjeng or whatever, I think it is fantastic. It means that people are getting a great deal of enjoyment out of it and have passion for the subject. If the subect is not interesting to people, then we won't have lots of posts on that topic.

So if you are not interested in a subect, do what I do. Skip over that post and read the next one that looks interesting.

Agree. Actually it doesn't bother me in the slightest, but I thought I ought to keep up the good traditions of CCC by commenting upon it and perhaps requesting a new sub-forum!

[Peter Skinner]

Sound familiar?

Most posts here are due to bug fixes or compile issues.

I personally have found several that Robert has managed to fix. With the current way that we test, releases should now flow every few weeks (possibly 4-6 weeks?).

[vb4]

Hi Bob,

So it takes atleast a couple of weeks for a +20 Elo increase? Great. So if we allow 4 weeks / +20 ELO in 2 years we're looking at a +520 in ELO !! I cant wait. I wish ! Only kidding Bob, you have done a incredible job developing Crafty. Keep up the good work.

Les

[bob]

Hi Bob,

So it takes atleast a couple of weeks for a +20 Elo increase? Great. So if we allow 4 weeks / +20 ELO in 2 years we're looking at a +520 in ELO !! I cant wait. I wish ! Only kidding Bob, you have done a incredible job developing Crafty. Keep up the good work.

Les

the tuning goes reasonably quickly, although there is a finite amount of stuff to tune. But we also have more evaluation to add. I wanted to do something similar to fruit/glaurung/etc with respect to passed pawns and give credit for controlling squares in front of the pawn. Another thing is an open file adjacent to a passed pawn with a rook or rooks is another feature we don't recognize yet. And we have other plans for king safety. We already have some things like long-range attacks of pieces against the king, but we plan on doing more...

There's probably a few search issues left to tune as well. We no longer use fractional extensions and only extend on giving check. I have a couple of ideas about more aggressive LMR for some types of moves. Lots to try, and now we can make a very accurate determination of whether something works or not, no longer having to assume / guess...

[Laskos]

Sound familiar?

Sure, it reminds me of the old days when people would get bent out of shape over ChessMaster personality posts.

My personal opinion is that there is no such thing as too many posts on a topical subject. When people are foaming at the mouth over Rybka or Crafty or Junior or Naum or Sjeng or whatever, I think it is fantastic. It means that people are getting a great deal of enjoyment out of it and have passion for the subject. If the subect is not interesting to people, then we won't have lots of posts on that topic.

So if you are not interested in a subect, do what I do. Skip over that post and read the next one that looks interesting.

a version a day keeps the doctor away...

Or makes you pull your hair out. But I think we are about past that stage. The threads seem to be working for everyone, the windows and linux memory leaks (two different issues completely) are history, and it now appears to compile cleanly for everyone so far as I know... Back to a more sedate release schedule.

I might use a "+20 Elo" window from here on, except for bugs. So that once we gain +20 elo over the last 22.n version, we release 22.n+1. That at least gives a target and it usually takes a couple of weeks (at a minimum) to produce that kind of improvement...

Good. Threfeore if 22.1 is 20 Elo points better than 18.12 I expect to see the next Crafty on forum in 6 years. There will be no complaints about "too many crafties".

Kai

[bob]

Sound familiar?

Sure, it reminds me of the old days when people would get bent out of shape over ChessMaster personality posts.

My personal opinion is that there is no such thing as too many posts on a topical subject. When people are foaming at the mouth over Rybka or Crafty or Junior or Naum or Sjeng or whatever, I think it is fantastic. It means that people are getting a great deal of enjoyment out of it and have passion for the subject. If the subect is not interesting to people, then we won't have lots of posts on that topic.

So if you are not interested in a subect, do what I do. Skip over that post and read the next one that looks interesting.

a version a day keeps the doctor away...

Or makes you pull your hair out. But I think we are about past that stage. The threads seem to be working for everyone, the windows and linux memory leaks (two different issues completely) are history, and it now appears to compile cleanly for everyone so far as I know... Back to a more sedate release schedule.

I might use a "+20 Elo" window from here on, except for bugs. So that once we gain +20 elo over the last 22.n version, we release 22.n+1. That at least gives a target and it usually takes a couple of weeks (at a minimum) to produce that kind of improvement...

Good. Threfeore if 22.1 is 20 Elo points better than 18.12 I expect to see the next Crafty on forum in 6 years. There will be no complaints about "too many crafties".

Kai

I do not follow. the current version is over 100 Elo better than 22.1, based on hundreds of thousands of games played. Don't know about 18.12 and such however. But the current version drubs 20.x's and 21.x's pretty nicely. This version is playing better than Fruit, where older versions (pre 22.2) lost pretty convincingly to fruit on our testing. That elo gain is real. The version I just released to fix a memory leak related to windows NUMA code is 10 elo stronger than the previous version due to tuning some eval terms and then cluster testing the changes.