Hello Chess Friends,
Let Me say please some notes more about chess...I think that a such thread is missing, but definitely,It's all about for Computer Chess's progress,...!And I hope you will enjoy reading...)
1st of all,Anyone is free...really..I can't force no any one...And no BIG problem for my side....simply:I am trying just to help...no more..no less....
And for anyone missed,I have some experience in chess...not much...but I have!In short, I realized to share some of my experience with you....
Well, let's give a start:Many testers (including many authors etc) are not aware of thatTheir work is not perfect too (like my work of course) however,As far as possible I am trying to not repeat the same mistakes...
And without to not mention this I can not,What I suggest, before any releases: games, engines, books etcThey have to check their work more carefully tooWhy each time I have to check... ?I can understand if they are beginners, amateurs ?)
So I kindly suggest to all engine programmers, testers, book makers etcPlease check your work more carefully before publishing online... !I am not a beta tester or amateur...actually I was, but in 1997-1998 years...
Continuing...
About Testers/Managers:- Many are not aware even that latest Fritz GUIs do not support book learning- Some are testing clone books or engines...I mean they don't pay attention, sad ( Actually I test SF based engines too, but I try to prefer honest authors engines! But I stay far away from IPPO family ones...Sim tool does not make mistake !) Plus I am doing my best to not make double standards...e.g you may know that Many testers refuse mostly derivatives...I have difficulties to understand them... E.g why Critter, Fire etc. are allowed..but many SF based ones not allowed? I can give more info, but no free time and I suggest to be used Simex or Sim03 - Some testing NNUE vs NON-NNUE..not fair...see below e.g AlphaZero vs SF8- Almost all prefer testing the opening books via Evalfile...I think it's a wrong way.. You can not determine the real book strength...why do you test via assistance ? Not so clear ? just imagine...your goal is to check the endgame knowledge, but You prefer to test the engines via TableBases...)) so what is the point then..?) Plus via Evalfiles, many of the critical weak lines may ended as draw, just saying... If you are online player all right..if is NNUE Test ok, but if Book Test..no comments) Exc. if using Evalfile, then not bad idea the tour name to be as: NNUE + Book Test Otherwise...I can not see any reason and to call it as real Book Tour...hope helps... Btw, here is my old tutorial about how to create and run book players: YouTube- Some prefer very weak short openings..and not so clear the real strength: Details
About pondering...(running matches on same machine),There are a lot of comments on forums...but only commentsTalking and talking...here I will try to talk too but a small note:Plus via testings...also via games too...at least (if nothing else)I have some data to compare...not required new tests again...And now the most important question is coming: Which method is better Ponder ON or Ponder OFF?Ok...1st of all, Chess is thinking in the opponent's turn!Actually it all depends... e.g on high/mid-end machines:Ponder ON sounds not bad, better/stronger games for sure!However there are some issues...which I have to mention:If you have a low-end machine e.g with 2 cores / 4 threadsThen I suggest the engine testings to be as Ponder OFFSure...pondering can be enabled, but it is safer to use 1 Core...Or if the target to produce more games (via parallel matches):Then Ponder OFF can be another good choice as well....On other hand,20-25 years ago... yes...almost all of the machines were 1 CPUIn those period of time, mostly are played via pondering as disabledBut nowadays, many of high-end machines come with many CPUsAnd for maximum performance/strength:Ponder ON seems to be ideal choice but if we use Bullet time control,Then the engines can be tested as Ponder OFF too, we can notExpect much quality games, e.g may appear a lot of blunder moves...But playing serious tours or creating ratings on high-end CPUs:Then...no comments)Meanwhile, a few words about Ponder hits (via Ponder ON),If you are running 3-5 games, yes..this can be as advantage... But at least I can talk about SCCT conditions, Usually I prefer to produce serous amount of games (per player)!So that means, the opponent has same/equal chances to gain...!If X engine suffers to gain, that means it's a Engine's ponder bug..This is also true that as far as possible, I test the stable ones!And I prefer MP+Ponder ON testings as less than 85% CPU usageThat's really very important...in this way, we see better quality games!Just imagine again...some prefer Ponder OFF with maximum cores..But his PC background etc. are full with unnecessary programs...soDo you expect much quality games under such Ponder OFF conditions ?In short, it's safer to be played under 85% CPU usage, it does not matterPonder ON or OFF...even if running without any unnecessary programs...
Btw, one thing more, according to my chess engine testings too,Blitz Ponder ON can be counted in strength as Rapid Ponder OFFFor anyone missed,Learn the real power/strength of pondering ON vs OFF: ResultsAs alternative, Ponder ON vs OFF, 4 Men vs 5 Men etc. Results
About Engine Programmers,- Mostly pay attention only over EvalFile strength, what about stability ?You can create even 5000 Elo engine..but if loses on time...sorry...I would prefer much weaker engine one, which does not loose on time!And why I have to spend time...to cancel and re-start my competitions ?I have no much free time for all, really..the life is short and not only chess!For more information over these issue, which engines suffer: Details- Mostly as NNUE is all right, but some as NON-NNUE suffers: Results
However, in my opinion:NNUE is not everything...that's good...better, stronger games for sureOn other hand, I am a human and I think as human, plus as a tester...Sometimes I don't care much over X NNUE engine, which gains from Previous played data (Evalfiles)..in short, impressive... but boring...!)And in 2022 year, I plan to work mainly over NON-NNUE testings...To be more clear, one of the main reasons is that,If the released engines are come with NN file, which already learned/played...Then why I have to test those chess engines again? for example just in case:Those NNUE engines will start to gain from previous played data (Evalfiles)Frankly, I started to see these kind of NNUE testings as not so much exciting...Plus especially the Top NNUE engines (via NN files) very rarely make mistakesOnly with very fast time controls and very weak openings we can see more wins...Sure the weak engines produce low draw perc. anyhow, better as NON-NNUE!And now, I hope you are able see my points of view...
Meanwhile, just now I remembered the Duel: AlphaZero Vs Stockfish 8Mostly fans were concentrated over AlphaZero's hardware advantageBut day by day the things started to be clear (if nothing else for my side)In short, I think that one of the main advantage was that,AlphaZero played via NN file...where Stockfish 8 played without NN fileIn my eyes, this is a clear big injustice...not fair conditions for sure...!
On other hand, sure this is just my dream )I hope AlphaZero team will challenge e.g Kayra or StockfishBut in case of such Duel, they should not forget to enable Evalfile)And then, I wonder much, what will be the real influence...?As before, AlphaZero will manage convincingly to defeat..?
One thing more, just I'd like to add,Actually, you may know...in FIDE or any other organized GM championships...It's not allowed to look on previous played own data and simply why ?)Because...I think in that case,1st..it will be not so much serious, 2nd will be not much interesting matches...Btw, In this case too..GMs also will produce stronger, better quality games )And draw percentage is expecting to be increased too...as NNUE engines)Even some may appear and to call it as not so original or just as cheating..))
About Book Makers,- It is true that some authors really spend a lot of time, very hard work...But some, sorry to say that, without any serious optimizations are releasing...Ok, no problem...then just don't request over than 200-300 games (per player)Plus just do not request with slower time controls or giving more cores ...Some create less than 5 min (per book) why I have to spend days for testing ?
On other hand, there is no doubt that,- There are many in our chess hobby, who are real experts, great talents... To be honest, usually I am trying to learn from them, and what about you?) In other words, Being wrong is not a shame... the bad thing when we do not learn or teach!
As a last note,We are almost in 2022, it is time for a new revolution !If you will follow my mentioned points is good...if not too bad..And I hope with my current notes,At least some of them will be helpful if not sad, but I can live with that..
Btw,If you have any suggestions, ideas or comments: Contact
I wish everyone all the best in 2022!
Best Regards,Sedat Canbaz
Let Me say please some notes more about chess...I think that a such thread is missing, but definitely,It's all about for Computer Chess's progress,...!And I hope you will enjoy reading...)
1st of all,Anyone is free...really..I can't force no any one...And no BIG problem for my side....simply:I am trying just to help...no more..no less....
And for anyone missed,I have some experience in chess...not much...but I have!In short, I realized to share some of my experience with you....
Well, let's give a start:Many testers (including many authors etc) are not aware of thatTheir work is not perfect too (like my work of course) however,As far as possible I am trying to not repeat the same mistakes...
And without to not mention this I can not,What I suggest, before any releases: games, engines, books etcThey have to check their work more carefully tooWhy each time I have to check... ?I can understand if they are beginners, amateurs ?)
So I kindly suggest to all engine programmers, testers, book makers etcPlease check your work more carefully before publishing online... !I am not a beta tester or amateur...actually I was, but in 1997-1998 years...
Continuing...
About Testers/Managers:- Many are not aware even that latest Fritz GUIs do not support book learning- Some are testing clone books or engines...I mean they don't pay attention, sad ( Actually I test SF based engines too, but I try to prefer honest authors engines! But I stay far away from IPPO family ones...Sim tool does not make mistake !) Plus I am doing my best to not make double standards...e.g you may know that Many testers refuse mostly derivatives...I have difficulties to understand them... E.g why Critter, Fire etc. are allowed..but many SF based ones not allowed? I can give more info, but no free time and I suggest to be used Simex or Sim03 - Some testing NNUE vs NON-NNUE..not fair...see below e.g AlphaZero vs SF8- Almost all prefer testing the opening books via Evalfile...I think it's a wrong way.. You can not determine the real book strength...why do you test via assistance ? Not so clear ? just imagine...your goal is to check the endgame knowledge, but You prefer to test the engines via TableBases...)) so what is the point then..?) Plus via Evalfiles, many of the critical weak lines may ended as draw, just saying... If you are online player all right..if is NNUE Test ok, but if Book Test..no comments) Exc. if using Evalfile, then not bad idea the tour name to be as: NNUE + Book Test Otherwise...I can not see any reason and to call it as real Book Tour...hope helps... Btw, here is my old tutorial about how to create and run book players: YouTube- Some prefer very weak short openings..and not so clear the real strength: Details
About pondering...(running matches on same machine),There are a lot of comments on forums...but only commentsTalking and talking...here I will try to talk too but a small note:Plus via testings...also via games too...at least (if nothing else)I have some data to compare...not required new tests again...And now the most important question is coming: Which method is better Ponder ON or Ponder OFF?Ok...1st of all, Chess is thinking in the opponent's turn!Actually it all depends... e.g on high/mid-end machines:Ponder ON sounds not bad, better/stronger games for sure!However there are some issues...which I have to mention:If you have a low-end machine e.g with 2 cores / 4 threadsThen I suggest the engine testings to be as Ponder OFFSure...pondering can be enabled, but it is safer to use 1 Core...Or if the target to produce more games (via parallel matches):Then Ponder OFF can be another good choice as well....On other hand,20-25 years ago... yes...almost all of the machines were 1 CPUIn those period of time, mostly are played via pondering as disabledBut nowadays, many of high-end machines come with many CPUsAnd for maximum performance/strength:Ponder ON seems to be ideal choice but if we use Bullet time control,Then the engines can be tested as Ponder OFF too, we can notExpect much quality games, e.g may appear a lot of blunder moves...But playing serious tours or creating ratings on high-end CPUs:Then...no comments)Meanwhile, a few words about Ponder hits (via Ponder ON),If you are running 3-5 games, yes..this can be as advantage... But at least I can talk about SCCT conditions, Usually I prefer to produce serous amount of games (per player)!So that means, the opponent has same/equal chances to gain...!If X engine suffers to gain, that means it's a Engine's ponder bug..This is also true that as far as possible, I test the stable ones!And I prefer MP+Ponder ON testings as less than 85% CPU usageThat's really very important...in this way, we see better quality games!Just imagine again...some prefer Ponder OFF with maximum cores..But his PC background etc. are full with unnecessary programs...soDo you expect much quality games under such Ponder OFF conditions ?In short, it's safer to be played under 85% CPU usage, it does not matterPonder ON or OFF...even if running without any unnecessary programs...
Btw, one thing more, according to my chess engine testings too,Blitz Ponder ON can be counted in strength as Rapid Ponder OFFFor anyone missed,Learn the real power/strength of pondering ON vs OFF: ResultsAs alternative, Ponder ON vs OFF, 4 Men vs 5 Men etc. Results
About Engine Programmers,- Mostly pay attention only over EvalFile strength, what about stability ?You can create even 5000 Elo engine..but if loses on time...sorry...I would prefer much weaker engine one, which does not loose on time!And why I have to spend time...to cancel and re-start my competitions ?I have no much free time for all, really..the life is short and not only chess!For more information over these issue, which engines suffer: Details- Mostly as NNUE is all right, but some as NON-NNUE suffers: Results
However, in my opinion:NNUE is not everything...that's good...better, stronger games for sureOn other hand, I am a human and I think as human, plus as a tester...Sometimes I don't care much over X NNUE engine, which gains from Previous played data (Evalfiles)..in short, impressive... but boring...!)And in 2022 year, I plan to work mainly over NON-NNUE testings...To be more clear, one of the main reasons is that,If the released engines are come with NN file, which already learned/played...Then why I have to test those chess engines again? for example just in case:Those NNUE engines will start to gain from previous played data (Evalfiles)Frankly, I started to see these kind of NNUE testings as not so much exciting...Plus especially the Top NNUE engines (via NN files) very rarely make mistakesOnly with very fast time controls and very weak openings we can see more wins...Sure the weak engines produce low draw perc. anyhow, better as NON-NNUE!And now, I hope you are able see my points of view...
Meanwhile, just now I remembered the Duel: AlphaZero Vs Stockfish 8Mostly fans were concentrated over AlphaZero's hardware advantageBut day by day the things started to be clear (if nothing else for my side)In short, I think that one of the main advantage was that,AlphaZero played via NN file...where Stockfish 8 played without NN fileIn my eyes, this is a clear big injustice...not fair conditions for sure...!
On other hand, sure this is just my dream )I hope AlphaZero team will challenge e.g Kayra or StockfishBut in case of such Duel, they should not forget to enable Evalfile)And then, I wonder much, what will be the real influence...?As before, AlphaZero will manage convincingly to defeat..?
One thing more, just I'd like to add,Actually, you may know...in FIDE or any other organized GM championships...It's not allowed to look on previous played own data and simply why ?)Because...I think in that case,1st..it will be not so much serious, 2nd will be not much interesting matches...Btw, In this case too..GMs also will produce stronger, better quality games )And draw percentage is expecting to be increased too...as NNUE engines)Even some may appear and to call it as not so original or just as cheating..))
About Book Makers,- It is true that some authors really spend a lot of time, very hard work...But some, sorry to say that, without any serious optimizations are releasing...Ok, no problem...then just don't request over than 200-300 games (per player)Plus just do not request with slower time controls or giving more cores ...Some create less than 5 min (per book) why I have to spend days for testing ?
On other hand, there is no doubt that,- There are many in our chess hobby, who are real experts, great talents... To be honest, usually I am trying to learn from them, and what about you?) In other words, Being wrong is not a shame... the bad thing when we do not learn or teach!
As a last note,We are almost in 2022, it is time for a new revolution !If you will follow my mentioned points is good...if not too bad..And I hope with my current notes,At least some of them will be helpful if not sad, but I can live with that..
Btw,If you have any suggestions, ideas or comments: Contact
I wish everyone all the best in 2022!
Best Regards,Sedat Canbaz