DanaSah 6.1 for Windows and Android

[pedrox]

Because the development time of a new version of danasah takes me longer than expected, I will post an intermediate version. You will find DanaSah 6.1 for Windows and Android. There are 2 versions of the engine; one normal and other limited strenght (ls) where you can regulate Elo 0-2500 and has some easy levels. The main difference with the previous version 5.07 is that DanaSah now also incorporates the uci protocol. If you use a uci GUI that not know the command uci_newgame change this in the options (for example this allows compatibility with old Fritz GUI although it will lose some speed).Besides the opening book it has been changed a Polyglot format, although the book distributed remains of Jeroen Noomen and Ed Schroeder. Also many changes in other things.

DanaSah can play normal and Chess960 variants than as xboard as uci protocol. I think can play well on all interfaces that support this protocol (About uci you have to activate the option UCI_Chess960)

Android versions make use of the opening book but not the Scorpio bitbases (I think that there is no engine to play well with them, I see the same problem with DanaSah and Toga2 about Chess for Android, the version included here is compiled with static option, it could crash if egbbso.so file is installed on chess for Android).

Because of the many changes made and adding a second protocol, Chess960, Windows and Android, etc., it takes me a long time to test that everything is correct. So I publish this version so it can be tested, I am more interested in the stability of the engine that in force. It is not necessary to include this version in the rating lists, there are few changes in the search and evaluation, so the Elo must be very similar to version 5.07 at least in the xboard protocol. Note that I only include 64-bit versions for Windows and unoptimized although danasah is not bitboard and not take advantage of 64 bits.

http://www.mediafire.com/download/d43vr ... asah61.zip

[velmarin]

Hi, Pedro
it seems something is not going well on Fritz 12 and uci mode.
After a first game right at the start of the second the first move is illegal, with composite output position.
You need to try with winboard, to isolate the problem.
In console

Code: Select all

uciok
go
info depth 1 time 0 nodes 2 nps 0 score cp 3949 pv b9a8
info depth 2 currmove b9a8 currmovenumber 1
info depth 2 currmove b9a8 currmovenumber 1
info depth 2 time 16 nodes 6 nps 0 score cp 0 upperbound pv b9a8
bestmove b9a8

Yellow arrow is the movement

[velmarin]

Hi, Pedro
it seems something is not going well on Fritz 12 and uci mode.
.....................

Again I made this mistake, sorry, Pedro
Do not disable the:
option name GUI_Has_Ucinewgame_Command type check default true

Now everything is OK.
Disculpa.

[pedrox]

the Elo must be very similar to version 5.07 at least in the xboard protocol

This surprises me. So WB and UCI results are incomparable? What is the reason?

Thanks anyway.

Hi Gabor,

For a long time I have optimized the engine to play with the xboard protocol. Now that I added the UCI protocol I had to change some functions that had been optimized for xboard protocol, perhaps 20%. In this process you can make mistakes and not optimize well for uci protocol.

When I tried the first uci version, this played about 100 Elo points worse than the version xboard (blitz). I have discovered that I had 2 problems, one that I was resetting hash tables between movements because uci protocol sent with each move a command to configure the position (lost 40 Elo). And the second problem is that with every move the uci protocol also sends the list of all the moves and I was setting and making the list with every move (lost 40 Elo). I solved these 2 problems but I still feel that uci version may be somewhat lower as 20 points, at least in blitz. But I'm not a big tester and I can be confused.

Fermin Serrano with Rodin had a similar problem. There is a thread where people speak a little of this topic. I do not want to open a debate between xboard protocol against uci protocol:

http://www.talkchess.com/forum/viewtopi ... highlight=

[pedrox]

Thank you very much for games, Gabor.

This was DanaSah 6.1 xboard vs DanaSah 6.1 uci ?

I'm really surprised by these results. Maybe I now broke something in the xboard protocol.

You time control is much better than mine, I tried to 40/1

[pedrox]

I'm glad it worked. Of all the GUI I've tried only previous versions of Fritz do not have this command (Fritz 15 yes). Having this option in the parameters is ugly but he make programming easier for me.

On previous versions of Fritz engine with every move makes the setup of the starting position and all movements sent by the protocol so he can lose some speed especially in bullet and blitz games.

Maybe I can remove this option in a future version.

Hi, Pedro
it seems something is not going well on Fritz 12 and uci mode.
.....................

Again I made this mistake, sorry, Pedro
Do not disable the:
option name GUI_Has_Ucinewgame_Command type check default true

Now everything is OK.
Disculpa.

[Guenther]

Thanks Pedro.

I'm going to set up a gauntlet between WB Danasah and UCI Danasah, in order to see if there is significant strength difference between them.

Hmm, in a 200-game match under cutechess-cli with 40/4 tc UCI won against WB by 125-75!

You mean 40/4 seconds here right? I guess Pedro means minutes.

[Guenther]

Thanks Pedro.

I'm going to set up a gauntlet between WB Danasah and UCI Danasah, in order to see if there is significant strength difference between them.

Hmm, in a 200-game match under cutechess-cli with 40/4 tc UCI won against WB by 125-75!

You mean 40/4 seconds here right? I guess Pedro means minutes.

Yes, 40 moves in 4 seconds. In such a short time I could not afford longer games.

Well I am absolutely sure Pedro missunderstood you then, because he meant 40/1 minute for his tests.
I have no idea if 40/4s gives reliable results for such a test, but I doubt it.

[pedrox]

This was DanaSah 6.1 xboard vs DanaSah 6.1 uci ?

Yes. Just for reference, here is the cutechess batch file I used. Maybe you find something wrong in it, I am a beginner as far as cutechess is concerned.

cutechess-cli -event Danasah -engine name=D_UCI proto=uci cmd=Danasah_6.1-x64.exe dir=e:\Sakk\Engines\Danasah\ option.Hash=128 option.Ponder=false -engine name=D_WB proto=xboard cmd=Danasah_6.1-x64.exe dir=e:\Sakk\Engines\Danasah\ -each tc=40/12 -draw movenumber=70 movecount=5 score=20 -resign movecount=6 score=700 -tournament gauntlet -openings file=e:\Sakk\PGN-books\HS-Grand_8moves[31437].pgn format=pgn order=random plies=20 start=1 -repeat -games 10 -rounds 10 -pgnout Danasah.pgn -site Szentendre

BTW, when starting the match, cutechess warned me that Danasah has no Ponder option. Odd, isn't it?

Best regards,

When I face the engine against himself I have 2 copies of different engine, for example danasah_xboard.exe vs danasah_uci.exe. Although perhaps this is the same.

I think if you do not use the option of cutechess "ponder" then you need not put option.Ponder = false. I do not put and I have no warning. Although it is the first time I use cutechess 8, the version 6 of cutechess did not ponder.

I tried 100 games 40/4 as you:
Score of danasah61_xboard vs danasah61_uci: 8 - 63 - 29 [0.225] 100
ELO difference: -214.85 +/- 63.56

I am surprised at the result, I'll try to 40/60 which is how I did most test.

[pedrox]

I tried to play the following games:

DanaSah 5.07 vs DanaSah 6.1 uci

Score of danasah507 vs danasah61_uci: 20 - 22 - 58 [0.490] 100

This result is normal.

---

DanaSah 5.07 vs DanaSah 6.1 xboard

Score of danasah507 vs danasah61_xboard: 40 - 16 - 44 [0.620] 100

This result is not normal. I have broken something in the version xboard. I have tried to find the problem but I still have not seen.

-----

I have an previus version DanaSah 6.0 xboard

Score of danasah507 vs danasah605_xboard: 18 - 23 - 59 [0.475] 100

This result is normal.

------

So I have a good thing, a previous version with the 2 protocols that played well. The trouble is that between version 6.0 and 6.1 are big changes (version 6.0 uci have something problems) and might cost me find the error. Thanks for the games. I will try to solve the problem in the coming days.