Moderator Elections

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

User avatar
Rebel
Posts: 7034
Joined: Thu Aug 18, 2011 12:04 pm
Full name: Ed Schröder

Re: Moderator Elections

Post by Rebel »

Pali wrote: Mon Apr 29, 2024 3:36 pm Perhaps not relevant to this specific subject but I suggested this on my chat with Rebel/Ed Schroder on discord:

- Having a pinned post on testing methodologies used by modern engines. This could link to OpenBench github, FishTest github, JW's SPRT github, Neural Network Trainers etc.
- Explicitly inform users that SF clones (and clones in general) are very likely to be either worse or neutral. It's really easy to contact SF maintainers, information on which SF clones serve a purpose (e.g. I see Huntsman being brought up for mate finding) and which clones do not. Keep in mind that SF is GPLv3, if a clone contributed Elo, the changes would be ported to SF as they have full legal rights to do it.
- Flag SSS testing? Perhaps this outside the scope of talkchess moderation but it's also supposed to be informing to the average reader. I saw someone post 14 wins, 11 losses, 47 draws as evidence that a version is better the other day. Someone who's not experienced could easily fall for this kind of statistic.

Keep in mind, I do not know the limitations of what can and can't be done on talkchess and as such my phrasing may be off or I may be suggesting something that may not be possible.
Thanks Pali for constructive ideas.

I am thinking of a sticky thread about chess programming from the ground up divided into 3 sections : newbie | advanced | expert. Something like -

Newbie -
1. I heard a lot good things about Vice.
2. Focus should be on bug-free.
3. Perft
4. Compiler choice

Advanced -
1. Introduction to cutechess
2. introduction to "how many games" to proof a change

Expert -
Introduction to SPSA

And of course an introduction to NNUE.

Can we count on the cooperation of Discord members as a mutual project with proper credit given to the contributors?

------

The other issues then, Talkchess since its existence is a place for every computer chess lover who wants to share and the only limitation is the charter they should keep and we don't want to change that or become the forum for discord. People who play only 100 games know very well that is not enough, they have been told dozen and dozen times during the years. They don't have the hardware you have on OpenBench or on Fishtest and we don't want to take away their pleasure because they have limited hardware.

Then we have the well known CCRL and CEGT rating lists since 2006, even they can't produce the needed games as everybody can see looking at the error bars. Not perfect, but I don't want to miss them. Same problem, not enough computer power like you have. They are very useful for starters since they test every engine.

Derivatives - I am not into the number of legal derivatives, but tell me, has there been one SF derivative SF profited from?
90% of coding is debugging, the other 10% is writing bugs.
Pali
Posts: 29
Joined: Wed Dec 01, 2021 12:23 pm
Full name: Doruk Sekercioglu

Re: Moderator Elections

Post by Pali »

Rebel wrote: Tue Apr 30, 2024 10:21 am Thanks Pali for constructive ideas.

I am thinking of a sticky thread about chess programming from the ground up divided into 3 sections : newbie | advanced | expert. Something like -

Newbie -
1. I heard a lot good things about Vice.
2. Focus should be on bug-free.
3. Perft
4. Compiler choice

Advanced -
1. Introduction to cutechess
2. introduction to "how many games" to proof a change

Expert -
Introduction to SPSA

And of course an introduction to NNUE.

Can we count on the cooperation of Discord members as a mutual project with proper credit given to the contributors?

------

The other issues then, Talkchess since its existence is a place for every computer chess lover who wants to share and the only limitation is the charter they should keep and we don't want to change that or become the forum for discord. People who play only 100 games know very well that is not enough, they have been told dozen and dozen times during the years. They don't have the hardware you have on OpenBench or on Fishtest and we don't want to take away their pleasure because they have limited hardware.

Then we have the well known CCRL and CEGT rating lists since 2006, even they can't produce the needed games as everybody can see looking at the error bars. Not perfect, but I don't want to miss them. Same problem, not enough computer power like you have. They are very useful for starters since they test every engine.

Derivatives - I am not into the number of legal derivatives, but tell me, has there been one SF derivative SF profited from?
While the steps are open to discussion, I do think the Newbie/Advanced/Expert is fine. One has to get a working chess implementation before testing patches and probably an engine with a decent amount of features before doing any form of SPSA.

Yes, pretty much every project I'm aware of is either under MIT or GPL and a good amount of people would be willing to help.

This is fair, I get how long it takes as I've had to run SPRTs on 4-6 cores for a good while. I would still side with having those tests marked with a "Small Sample Size" but that is up to you. I think the things we were able to agree on is already the majority of what most engine developers wanted.

I am all for CCRL and CEGT, engine devs who are at the top of the list already are aware of the compression that happens. The vast majority of criticism I've seen towards CCRL has been about the clones that were tested, but this is not relevant to talkchess.

Derivatives - I am not aware of any SF derivative that has contributed Elo. Forks intended to be used for Fishtest patches are obviously not "derivatives". There are useful derivatives/clones though (Fairy Stockfish).
Viz
Posts: 100
Joined: Tue Apr 09, 2024 6:24 am
Full name: Michael Chaly

Re: Moderator Elections

Post by Viz »

From what I can recall all ideas from sf derivatives kinda sucked.
smatovic
Posts: 2769
Joined: Wed Mar 10, 2010 10:18 pm
Location: Hamburg, Germany
Full name: Srdja Matovic

Re: Moderator Elections

Post by smatovic »

FYI:

All new forum issues fixed, most feature requests solved, Linux server hardened
in a 1st and 2nd step, two months w/o DDoS attack, crawlers and scrapers active,
load/memory/traffic look good to me.

We are green to hold moderator elections from my side.

--
Srdja
User avatar
Antihelion
Posts: 30
Joined: Tue Mar 26, 2024 8:21 pm
Full name: Lyndon S. Sears

Re: Moderator Elections

Post by Antihelion »

DDOS attack? I wonder who's behind this.
smatovic
Posts: 2769
Joined: Wed Mar 10, 2010 10:18 pm
Location: Hamburg, Germany
Full name: Srdja Matovic

Re: Moderator Elections

Post by smatovic »

Antihelion wrote: Wed May 01, 2024 4:28 am DDOS attack? I wonder who's behind this.
2020 to 2024 IP blocking and issues with Cloudflare CDN to prevent presumed
DDoS attacks on TC, causing collateral damage on ChessUSA shop hosted
together on same server
viewtopic.php?p=962928#p962928

Retrospectively, it is not clear, if these were real DDoS attacks against TC, or an overload by bad scrapers and good crawlers, TC is of value as data treasure for some players.

--
Srdja
User avatar
Rebel
Posts: 7034
Joined: Thu Aug 18, 2011 12:04 pm
Full name: Ed Schröder

Re: Moderator Elections

Post by Rebel »

Pali wrote: Tue Apr 30, 2024 11:38 am
Rebel wrote: Tue Apr 30, 2024 10:21 am Thanks Pali for constructive ideas.

I am thinking of a sticky thread about chess programming from the ground up divided into 3 sections : newbie | advanced | expert. Something like -

Newbie -
1. I heard a lot good things about Vice.
2. Focus should be on bug-free.
3. Perft
4. Compiler choice

Advanced -
1. Introduction to cutechess
2. introduction to "how many games" to proof a change

Expert -
Introduction to SPSA

And of course an introduction to NNUE.

Can we count on the cooperation of Discord members as a mutual project with proper credit given to the contributors?

------

The other issues then, Talkchess since its existence is a place for every computer chess lover who wants to share and the only limitation is the charter they should keep and we don't want to change that or become the forum for discord. People who play only 100 games know very well that is not enough, they have been told dozen and dozen times during the years. They don't have the hardware you have on OpenBench or on Fishtest and we don't want to take away their pleasure because they have limited hardware.

Then we have the well known CCRL and CEGT rating lists since 2006, even they can't produce the needed games as everybody can see looking at the error bars. Not perfect, but I don't want to miss them. Same problem, not enough computer power like you have. They are very useful for starters since they test every engine.

Derivatives - I am not into the number of legal derivatives, but tell me, has there been one SF derivative SF profited from?
While the steps are open to discussion, I do think the Newbie/Advanced/Expert is fine. One has to get a working chess implementation before testing patches and probably an engine with a decent amount of features before doing any form of SPSA.

Yes, pretty much every project I'm aware of is either under MIT or GPL and a good amount of people would be willing to help.

This is fair, I get how long it takes as I've had to run SPRTs on 4-6 cores for a good while. I would still side with having those tests marked with a "Small Sample Size" but that is up to you. I think the things we were able to agree on is already the majority of what most engine developers wanted.

I am all for CCRL and CEGT, engine devs who are at the top of the list already are aware of the compression that happens. The vast majority of criticism I've seen towards CCRL has been about the clones that were tested, but this is not relevant to talkchess.

Derivatives - I am not aware of any SF derivative that has contributed Elo. Forks intended to be used for Fishtest patches are obviously not "derivatives". There are useful derivatives/clones though (Fairy Stockfish).
Cool....send you a PM.
90% of coding is debugging, the other 10% is writing bugs.
User avatar
Rebel
Posts: 7034
Joined: Thu Aug 18, 2011 12:04 pm
Full name: Ed Schröder

Re: Moderator Elections

Post by Rebel »

Viz wrote: Tue Apr 30, 2024 11:46 am From what I can recall all ideas from sf derivatives kinda sucked.
That is my impression as well.
90% of coding is debugging, the other 10% is writing bugs.