OpenChess

dorsz wrote: Wed Jul 05, 2023 8:48 pm The downside of keeping an engine private is small number of users.

I doubt the purpose of a learning engine was to get tons of users. They are enough of them who don't understand the learning feature. I even know authors of learning engines who didn't know what experience data were in their own experience files !

Hello Friends,

Let me say please a few words over these issues!

1st of all, I see that
Computerchess is really in danger..forget everything
Even SF based ones are started each other cloning!
What is that ?? e.g according to my latest testings:
https://sites.google.com/site/computers ... -july-2023
Polyfish 210623 is almost identical to SF-PB 190623!

And the most important question is coming:
Who cloned firstly?..for example if we check the dates,
SF-PB is firstly released, where Polyfish is released later!
Really sad...I feel like I am a naive tester...not sure about
All of you? but someone have to stop these funny jokes!

On other hand,
Not so good step e.g taking so much from free Stoclfish
And later making as private and called as my own engine!
Ask yourself...if Stockfish was as private engine...then
I hardly doubt that current 'Eman 9.2 thread' would exist!
Is it clear ? or if not? just do not hesitate to ask me)

In other words, (sure I'm referring about SF based ones)
I think that without Stockfish's permission: no anyone has
Right to make as own private closed engine, but who cares?
And unfortunately not many pay too much attention over!

Btw, some useful data, where it's clear that Eman is based on SF:
Plus for anyone missed: Polyfish and SF-PB are almost same in playing...
1. Engine Polyfish 210623 and SF-PB 190623 show a similarity of 99.00%
21. Engine Eman 8.91 and RapTora 2.2_Eduard show a similarity of 69.87%
29. Engine Eman 8.91 and RapTora 2.2_Anton show a similarity of 69.47%
40. Engine Eman 8.91 and LittleBeast 2_sl show a similarity of 69.17%
55. Engine Eman 8.91 and Stockfish 16 show a similarity of 68.47
70. Engine Eman 8.91 and Stockfish 15.1 show a similarity of 68.07%


Greetings

Sedat Canbaz wrote: Wed Jul 05, 2023 10:21 pm Polyfish 210623 is almost identical to SF-PB 190623!

Yes, I think both have sources so you don't have to guess that.

Sedat Canbaz wrote: Wed Jul 05, 2023 10:21 pm And the most important question is coming:
Who cloned firstly?..for example if we check the dates,
SF-PB is firstly released, where Polyfish is released later!
Really sad...I feel like I am a naive tester...not sure about
All of you? but someone have to stop these funny jokes!

Who really cares who was the first? These engines are just Stockfish with an ability to use book. Even their authors said that.

dorsz wrote: Wed Jul 05, 2023 10:29 pm Yes, I think both have sources so you don't have to guess that.

Hmm... but I do not see the point here...because in my understanding
E.g working on same source is just meaningless, really waste of time..!
What is advantage between both? nothing...exc. different names !

dorsz wrote: Wed Jul 05, 2023 10:29 pm These engines are just Stockfish with an ability to use book. Even their authors said that.

No... I have to disagree here...for example
Mostly my tests and bench results indicate that
Mostly SF-PB vers are not exactly same as Stockfish!

Why I say like this?
E.g SF-PB is performed 12 Elo better than Stockfish:
https://sites.google.com/site/computers ... t-nn-cs-33
You may know, 1000 games (per player) is serious data...
I mean here... the error margin can not be as reason...

Plus as other reference, during my beta chess benchmarks testings:
https://sites.google.com/site/computers ... benchmarks
I found that SF-PB is not same as Stockfish dev.(released on close dates)
In other words, SF-PB performed and solved better...

If still not so clear,
Similarity results can be as other reference about SF-PB is not exactly same as SF
https://sites.google.com/site/computers ... -july-2023

Hope helps...!

Greetings

Hi Mr. Sedat
Thanks for your comment. But I can understand something here. You mentioned that working on a same source code is wasting of time. Then you mentioned that Sfpb performs 12 elo better than stockfish in your tests and it seems that polyfish doesn’t have this advantage in your tests. Now just suppose that polyfish has been created earlier than SFpb (it is a probability). Now can we say working on same source code or similar engines is just wasting of time?
Best regards

Homayoun wrote: Thu Jul 06, 2023 5:55 am Hi Mr. Sedat
Thanks for your comment. But I can understand something here. You mentioned that working on a same source code is wasting of time. Then you mentioned that Sfpb performs 12 elo better than stockfish in your tests and it seems that polyfish doesn’t have this advantage in your tests. Now just suppose that polyfish has been created earlier than SFpb (it is a probability). Now can we say working on same source code or similar engines is just wasting of time?
Best regards

Hello Mr. Homayoun

A good question...and thanks for your comments too )

Well, I see that there are some confusions between us
But I think that it's quite normal...we are not all same )
Anyhow, I will try to explain with more details...

1st of all, I don't claim that
All SF-PB and Polyfish vers are based on exactly same source/code!

But in this case..
If both engines are based on same code and has same playing style,
Then as a tester, simply I say again: it is just a waste of time !!
Why I have to spend so many efforts, CPU time.. yes, my work is free,
But not so cheap... )

Continuing...

But what is clear that, e.g according to my latest similarity's results:
https://sites.google.com/site/computers ... -july-2023

We noticed that
1. Engine Polyfish 210623 and SF-PB 190623 show a similarity of 99.00%
So..here is clear that both are almost same, at least on playing style!
Btw, as you may know, SF-PB's main difference (than Polyfish) is that
SF-PB vers are as small exe..I mean SF-PB is using external Evalfiles
Btw, Kayra, LittleBeast, SugaR etc. are relesing is same way... great!'

On other hand, (regarding my organized tours)
My goal is that to test engines with more different playing styles!
But what a pity that, nowadays my wish/goal is became too hard!
Almost daily SF based releases.and we don't know which are very close..
In other words,
I've no much free time (before starting a tour) to use similarity tools!
To understand me, what I mean...you have to be in my shoes..!)
Test 40-50 engines in a tour... and later we can talk... )

One thing more, this is other question of course.. but anyhow )
I found that the way of SF-PB's releasing system is so good!
Because I spend less time for downloading...plus I don't spend
Much time to delete all rest unwanted/not suitable exe files...
Even if I miss, these small exe files don't take too much disk space)!
E.g nowadays,unfortunately many are releasing with over than 200 MB
After extracting...these large eng files are going over than 1 GB...and if
We count also my tested books sizes...all my PC is full with chess files )

Regarding why 12 Elo difference..just a moment please,
I think that here you are missing very important things..
The tested engine versions (in my tours) are not same vers as
E.g with the ones, which I tested on my recent Simex2's results!

So we have no any data about the their similarities..
I mean regarding on older tours.g based on same code or not...
Sure Ii they are based on exactly same code and created in same way
Then...we may ask ourselves, about why there is 12 Elo difference..
But if not... then SF-PB seems to be different and stronger!
If nothing else ...at least under these conditions...

Greetings

It is only natural that SF-PB and Polyfish has a similarity of %99, since both of them are basically unchanged Stockfish with extented book features. Authors state that in their release notes.
On the other hand, scores like %60-70 are also natural in the NNUE era. If it was the classical evaluation era, people would call them clones of each other but the fact that they're playing much higer level (in long time controls, chess is pretty much solved, said larry kaufman) and evaluations are based on the same net, it is normal.

And one interesting subject is sometimes we see exactly a same engine with a fixed source code while compiling by two different persons shows different performances in all the tests. I remember this when SFpb was compiled by both Skynet and chessman. I am sure that you remember those days. I want to say that it seemed even the playing style was also a little different although it was only one source code and only one engine SFpb. This is the point that I can’t understand why, as a veterinarian who knows almost nothing about chess programming and making engines.
Best regards Mr. Sedat

Homayoun wrote: Thu Jul 06, 2023 7:50 am And one interesting subject is sometimes we see exactly a same engine with a fixed source code while compiling by two different persons shows different performances in all the tests. I remember this when SFpb was compiled by both Skynet and chessman. I am sure that you remember those days. I want to say that it seemed even the playing style was also a little different although it was only one source code and only one engine SFpb. This is the point that I can’t understand why, as a veterinarian who knows almost nothing about chess programming and making engines.
Best regards Mr. Sedat

Not at all.. Mr. Homayoun

Oh..yes... you have right...

What I can say more, according to my experience:
Actually...let's say the target is to build a good rating list
And just in case.. we used to test exactly same copy of engines
Then we can not expect exactly same identical Elo points...
Sure as same words are going for opening books as well

But what I know, usually
the strongest books, engines are getting closer or at higher places!
E.g as other examples,
https://sites.google.com/site/computers ... t-nn-cs-32
I do not think that RapTora 2.2, LittleBest 2, Brainlearn SugaR, etc
Are managed occasionally to be ranked at one of Top ranks...
I mean this is not as a luck for sure!

On other hand,
Some engines are doing better under Bullet, some better under Blitz etc..
And if I am going to give examples: Eman 8.91's performance seems to be so...
E.g under Bullet, Eman is performed about 20 Elo less than RapTora:
https://sites.google.com/site/computers ... t-nn-cs-33
But under Blitz conditions, Eman performed as one of best:
https://sites.google.com/site/computers ... t-nn-cs-34

Greetings

moonstonelight wrote: Thu Jul 06, 2023 7:36 am It is only natural that SF-PB and Polyfish has a similarity of %99, since both of them are basically unchanged Stockfish with extented book features. Authors state that in their release notes.
On the other hand, scores like %60-70 are also natural in the NNUE era. If it was the classical evaluation era, people would call them clones of each other but the fact that they're playing much higer level (in long time controls, chess is pretty much solved, said larry kaufman) and evaluations are based on the same net, it is normal.

Yes...for NNUE 3700+
In case similarity about 60%-70% can be said high..but I think it's acceptable
Even if we do not accept... then I am afraid to say that we'll have troubles
To find less than these numbers...

But in case of close to 99% similarity,
Then no words to say more...exception:
Someone is cloning each other work...sad..

Greetings