Which are the chess engines written from scratch ?

[Henk]

Why don't you just test a few top engines written from scratch. Pure originality is almost worthless. Original master pieces are certainly worth while.

[Sedat Canbaz]

Why don't you just test a few top engines written from scratch. Pure originality is almost worthless. Original master pieces are certainly worth while.

A good question

How we can know before especially which are the top 20 strongest scratch engines (without testing under these conditions) ?!

In other words,
Firstly I will test all scratch engines, so in the end of tournament each participant will have min. 100 games per player and later I will continue testing only the top 20 ones

Of course, in case of available a newer scratch version I will test it too

Also, I am interested to test more scratch engines, where still I am waiting for recommendations (regarding which chess engine is written from scratch) by chess friends

Best,
Sedat

[Sedat Canbaz]

Finely Maverick's testing is solved, I have already started testing Maverick 2013-10-10 too

It will be tested on my notebook (i5-3210M @ 2.50 GHz / Windows 8.1 64-bit)

And it's interesting to note that one (1) core speed difference between both processors (QX9650 3.0GHz vs i5-3210M 2.50GHz) is almost identical (checked by Fritz Chess Benchmark)

So, sometimes (when I will have free CPU time) I can test the scratch engines in my notebook too

[bob]

And what about Crafty, Pro Deo, SmarThink, Gandalf, Fritz, Junior, Hiarcs, Shredder, The King... are they written from scratch too or ...?

I think "written from scratch" is ill-defined.

Crafty was based on Cray Blitz. It was not written "from scratch".
Pro Deo might be based in part on Rebel (I don't know).
Current Fritz is based on Pandix.

My private engine is a rewrite of an earlier engine, so it was not written "from scratch".

I view Crafty as a continuation of Cray Blitz. Simply rewritten in C rather than FORTRAN. It certainly meets the "from scratch" criterion, it did not start as a copy of anything at all. There was nothing to copy in 1968.

Dear Robert,

First of all,
I'd like to thank you again for your BIG efforts...

In 1968, I was still not born

For example, in case of not allowing Crafty in a such scratch testing, that would be one of my biggest mistakes in Computerchess
That's why I don't want to make mistake...!!

Btw, Crafty is one of my favorite engines too, I used your engine almost daily and since several years, for my tournaments, for benchmarks...


Greetings,
Sedat

It is one of MY favorite engines as well.

[Sedat Canbaz]

I published SCCT Scratch Rating (a few minutes ago):
http://www.sedatcanbaz.com/chess/?page_id=634

Note that so far approx. 2700 games are already played
And the scratch engine database includes 113 games lost on time
That means in every 100 games, there is possibility to see 3-4 games forfeits on time (the number is not too big, honestly I expected much more...)
Those games which are lost on time are mainly belong to Djinn v1.008 (39 games), without those lost games on time, probably Djinn would be in Top 20

Since today, I started testing only the Top 20 scratch engines

Soon I will test a few scratch engines more (Jazz,Milady,Neurone...)

Games and more info coming soon!

Have fun,
Sedat

[Sedat Canbaz]

I've just calculated the played games by 3 calculation programs and noticed that especially the last 3 places have BIG difference in Elo points
*Note that Tornado is fixed to 2750 Elo

So on which Elo results we should relay, trust...?
And I wonder too about the last 3 places , are they near to human ratings ?

Code: Select all

Calculation by Ordo 0.8:
   # PLAYER                    : RATING    POINTS  PLAYED    (%)
   1 Discocheck 5.2 x64        : 2873.6     130.0     148   87.8%
   2 Crafty 23.8 x64           : 2823.7     124.5     148   84.1%
   3 Texel 1.02 JA x64         : 2779.4     121.0     148   81.8%
   4 Gaviota 0.86 x64          : 2766.6     119.0     148   80.4%
   5 Tornado 4.88 x64          : 2750.0     117.5     148   79.4%
  36 Tscp 1.81 JA              : 1619.9      10.5     140    7.5%
  37 MsKcp 1.4.5 JA            : 1476.7       6.0     140    4.3%
  38 Supra 18                  : 1316.7       2.5     140    1.8%

-----------------------------------------------------------------------------------

Calculation by Elostat 1.3:
    Program                          Elo    +   -   Games   Score   Av.Op.  Draws
  1 Discocheck 5.2 x64             : 2857   81  76   148    87.8 %   2513    8.1 %
  2 Crafty 23.8 x64                : 2804   68  65   148    84.1 %   2514   14.2 %
  3 Texel 1.02 JA x64              : 2776   58  56   148    81.8 %   2515   23.0 %
  4 Cheng 4.0.36a x64              : 2766   65  63   100    67.5 %   2639   21.0 %
  5 Gaviota 0.86 x64               : 2761   61  59   148    80.4 %   2516   17.6 %
  6 Tornado 4.88 x64               : 2750   61  59   148    79.4 %   2516   16.9 %
 36 Tscp 1.81 JA                   : 2084   99 110   140     7.5 %   2521    3.6 %
 37 MsKcp 1.4.5 JA                 : 1984  130  60   140     4.3 %   2524    1.4 %
 38 Supra 18                       : 1925   80   0   140     1.8 %   2525    2.1 %

-----------------------------------------------------------------------------------

Calculation by Bayeselo:
Rank Name                    Elo    +    -   games score oppo. draws 
   1 Discocheck 5.2 x64      2872   77   69   148   88%  2392    8% 
   2 Crafty 23.8 x64         2815   67   63   148   84%  2394   14% 
   3 Gaviota 0.86 x64        2761   63   60   148   80%  2395   18% 
   4 Texel 1.02 JA x64       2760   61   58   148   82%  2395   23% 
   5 Tornado 4.88 x64        2750   63   60   148   79%  2396   17% 
  36 Tscp 1.81 JA            1773   98  114   140    8%  2405    4% 
  37 MsKcp 1.4.5 JA          1666  111  139   140    4%  2408    1% 
  38 Supra 18                1564  126  169   140    2%  2411    2% 

[asanjuan]

Rhetoric was written from scratch ... 3 times!
It started as the final project for the university back on 1999 in Java. It was named JavaChess. It only had alpha-beta, killer moves and a buggy quiescent search. Evaluation was based only in PST. But it was a reeealy slow program.

I ported the code to C++ to gain some speed optimizations, but it was a total failure, because my java classes and structures were very inneficient, so i coded everything again using bitboards, and there is when I renamed the project to "Rhetoric", after a couple of years and the discovery of the fantastic CPW, I could solve some bugs and I could release the first "stable" version (0.10.d).

Then I wanted to implement magic bitboards, so to make things right, and because I knew that move ordering was very slow, I rewrote everything again, changing every single data structure except maybe the move type.

I think that Rhetoric fits in the description of "written from scratch", doesn't it?



Hope to see Rhetoric in your tournament.

Regards.

[asanjuan]

Oh, now I see that is too late.
The tournament has started yet...

[Sedat Canbaz]

Oh, now I see that is too late.
The tournament has started yet...

Thanks for your interest...

No...its not late, I plan to test your engine too (in 1-2 weeks in gauntlet mode against top 20 scratched participants)

Btw, actually for my current scratch testings is more likely to say a rating than tournament

Best,
Sedat

[tmokonen]

Initially, The Baron was based on TSCP. From Baron's changes.txt file:

0.01 (2001-08-29): First version of Baron, only rename TSCP to baron