Movei "peppers" Fruit

[Dr.Wael Deeb]

Hi,

here's the pgn ...

Probably you posted all the games in good spirit, but it is no good
idea to clutter the forum with lots of games from a PGN file.
Moreover it seems you hit a max number of allowed lines for posting
and your post therefore misses nearly all interesting games(21-50).
I looked at some of the games already, because now they are available
for checking, but it would have been much better they would have been
checked by Rudolf before posting the 'results'.
I see some strange depths from time to time, which even my old PIII
reaches with Fruit in fractions of the used time. Also Fruit is often
out of book after move 4 or 5(no idea what book was used, but it seems
not Fruits original book from Marc)
It seems something used a lot of CPU on Fruits machine. Some real debugs/logs
would be nice, because they could reveal at what speed
Fruit was running too reach that low depths. May be the cause was a
GUI hickup or the autoplayer or the live tool ate Fruits cpu power...

Guenther

I saw that this opening book was used:

HS-BOOK 2.0

[Graham Banks]

I saw that this opening book was used:

HS-BOOK 2.0

Hi Gerhard,

the HS-Book 2.0 is my preferred book under this GUI for CCRL testing. It's a very fair and balanced book.

Regards, Graham.

[ThatsIt]

I saw that this opening book was used:

HS-BOOK 2.0

Hi Gerhard,

the HS-Book 2.0 is my preferred book under this GUI for CCRL testing. It's a very fair and balanced book.

Regards, Graham.

Hi Graham !

i think you've mixed up the answers ?!

Best to you !
G.S.

[genorb]

Hi everybody,

just to mention that it seems that the new version of movei gain a lot on slower time control.

On my website I am testing engines with essentially 2 time control

bullet: 1min. + 1sec./move
blitz: 40 moves/5 min. with repetition

However when I include a new version of a program (in the blitz competitions), it plays 2 games against all other engines at time control 5min. per game.

for the bullet I get this result up to now (I give also the result for Fruit)

Name Elo + - games score oppo. draws

Fruit 2.3.1 2656 40 38 252 68% 2516 21%
Movei 0.8.438 2394 46 47 160 37% 2492 26%
Movei 0.8.403 2348 31 32 378 35% 2474 20%

So this is a gain of 46 "Elo" (but the errors are still large, I agree). The difference with Fruit is 262.

But for blitz I get this very good result

Name Elo + - games score oppo. draws

Fruit 2.3.1 2685 50 49 156 67% 2551 24%
Movei 0.8.438 2518 82
Movei 0.8.403 2378 36 37 266 39% 2463 24%

So the gain is now 140! The difference with Fruit is now only 167.

Sorry I don't have here (at work) the detail about the games played by the last version of Movei in blitz (The rating comes from my website, and the number of games from memory, it could be 86, but not more). But I recall that Movei 0.8.438 played only games with time control 5 min/game while the previous version of Movei played also a lot of game with time control 40 moves/5 min. (actually something like 266-82=184).

For your information I use the book HS-Book 2.0 for all engine limited to 10 moves. Detail about hardware, if important, can be find on my website (see my profile).

So I expect that the rating of the new version of Movei will even grow when I will make it play soon at slightly longer time control (namely 40 moves/5 min. instead of 5 min/game).

So even if in my tests Fruit stay much stronger, I do not know what would happen for very slow time control like 40/60'+40/40'+30'

Uri you made a great job!!

Regards,

Fabian

[Graham Banks]

I saw that this opening book was used:

HS-BOOK 2.0

Hi Gerhard,

the HS-Book 2.0 is my preferred book under this GUI for CCRL testing. It's a very fair and balanced book.

Regards, Graham.

Hi Graham !

i think you've mixed up the answers ?!

Best to you !
G.S.

Yes - my apologies Gerhard. My post should have been directed to Wael.

Regards, Graham.

[Dr.Wael Deeb]

http://f23.parsimony.net/forum50826/messages/172150.htm

[genorb]

http://f23.parsimony.net/forum50826/messages/172150.htm

Yes but now the time control is faster: Blitz 40/5' repeated

But OK, I think this result is more or less what we expect. By the way, based on those 50 games, what would be the rating gap? (and by the way how do you compute this gap easily? Bayeselo can do it?)

Thanks in advance,

Fabian

[Har]

http://f23.parsimony.net/forum50826/messages/172150.htm

Re: Movei00_8_438_WB -- Fruit 05/11/03, Fehler gefunden.


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


[ SCHACHWERKSTATT ]



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


Geschrieben von Rudolf Klamert am 10. September 2007 22:50:56:

Als Antwort auf: Re: Movei00_8_438_WB im Vergleich zu Fruit 05/11/03 geschrieben von Gerhard Sonnabend am 09. September 2007 10:45:04:

>Hallo Rudolf !
>Wow, 31.0-19.0 gegen Fruit 05/11/03 !!

Hallo Gerhard

Ursache gefunden!

Zwei Dinge sind es, die Fruit 05/11/03 so beeinträchtigt haben,
daß dieses merkwürdige Ergebnis zustande gekommen ist.

1. Der FTPTrigger ist mit einer Minute zu kurz eingestellt, 5 Minuten müssen reichen
2. die Prozessorauslastung ist bei Priorität 8 ca. 45% im Fall der Abfrage und das all Minuten!
3. Abhilfe: die Transferabfrage >=5minuten einstellen! Priorität auf 4

Bei der jetzt laufenden Begegnung wird alle 5 Minuten abgefragt und die Priorität von FTPTrigger ist 4
Im Ruhezustand liegt die CPU-Auslastung bei 1,5% beim Transfer ca. 5%
Die Shredderbases sind vorsichtshalber abgeschalten, zu wenig Speicherplatz.

Bitte Hiarcs 11.2SP bei der Laufenden Partie beobachten.

Mit dieser Einstellung werde ich die Begegnung Movei00_8_438 – Fruit 05/11/03 demnächst wiederholen.

Gruß, Rudolf






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







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


Antworten:


Re: Movei00_8_438_WB -- Fruit 05/11/03, Fehler gefunden. Michael Scheidl 10.9.2007 23:34 (2)
Re: Movei00_8_438_WB -- Fruit 05/11/03, Fehler gefunden. Rudolf Klamert 11.9.2007 07:09 (1)
Re: Movei00_8_438_WB -- Fruit 05/11/03, Fehler gefunden. Rudolf Klamert 11.9.2007 18:27 (0)


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


[ SCHACHWERKSTATT ]



:--------------------------------------------------------------------------------

[Har]

Rudolf Klamert had serious technical problems with Fruit.
He will repeat the match Fruit-Movei.

[Uri Blass]

Thank but I do not think that movei is relatively better at longer time control

Here is another Fruit-Movei match(this time from cegt 40/20)

http://husvankempen.de/nunn/phpBB2/view ... 3&start=15

1 Fruit 05/11/03 1½½½01011½11101½11½1111½1½11½½1111½111111101111111 40.5/50
2 Movei00_8_438 0½½½10100½00010½00½0000½0½00½½0000½000000010000000 9.5/50


There may be a different explanation for the improvement that you see:

Movei prints the pv more often thatn fruit and it may cause some interfaces to make movei slower at very fast time control.

Note that Movei prints every wrong fail high and also it prints the pv again at the end of the iteration when fruit print pv only when it finds a new pv.

version 438 is a significant improvement relative to previous versions
but it is still far away from fruit2.3.1(it is hardly better than fruit2.0)

Uri