Computer draughts tournament, Culemborg (2012.09.16)

[Ed Gilbert]

Preliminary announcement for the open Dutch championship rapid Polish Draughts for computerprograms 2012.

Day of the tournament: Sunday the 16th of September 2012.
This date was chosen as the favourite by 7 of the 8 responses to our inquiry for the day of play.
Responses came from Tornado, Dream, Jonah (made by Johan Taeke Dekker, participated already in 2008), GWD, Sjende Blyn, TD King and Cerberus.

Location: PW Consulting, Stationssingel 25 Culemborg, The Netherlands.

Time: 09:30 Installing the hardware, 10.00 Start first round.


Rules will be the same as during the tournaments of recent date (last year we played 9 rounds with playing-time 20 minutes per round per participant; after 75 moves arbitration if necessary).
Probably Jaap Bus will again be the referee (at the moment he is recovering from a gallstone-removal-operation).


If you want to participate, please report by answering this e-mail. If you are among the above mentioned responses, please answer as well.
Please report your participation before the 9th of September.


frgr


Leo Nagels

[MichelG]

I'll join as well with dragon.

Michel

[BertTuyt]

As I don't expect that my new Engine (Version 12 , already ) is ready (debugged and fully tested) in a few weeks, I'm afraid i will not participate

Bert

[Krzychumag]

I apologise that am asking, perhaps somebody knows. There is transmission On-line through the Internet from these tournament.

[BertTuyt]

I'm afraid not, but I'm not 10% sure.
At least it was not done before to my knowledge.

Bert

[Krzychumag]

I'm afraid not, but I'm not 10% sure.
At least it was not done before to my knowledge.

Bert

Bert thanks for information.

-- Krzysztof.

[BertTuyt]

I'm curious what happened during the Open Dutch Computer Draughts today.
Hope that some one has a clue, or that one of the participants can post a short note.

Bert

[MichelG]

I'm curious what happened during the Open Dutch Computer Draughts today.
Hope that some one has a clue, or that one of the participants can post a short note.

Bert

Dragon won I think these are the results:

Code: Select all

dragon       12 / 7 games
sjende blyn   9
td king       8
gwd           8
torn          7
dios          7
dream         3
cerberus      2

Michel

[BertTuyt]

Michel congratulations !!!!

Could you post some hardware/software and further match details?

Bert

[Krzychumag]

Michel congratulations !!!!

I have such a Michel question as for these tournament. Write what time of these tournament was and with what version of program Dragoon you played. Whether you are planning new versions of program.

[MichelG]

Michel congratulations !!!!

I have such a Michel question as for these tournament. Write what time of these tournament was and with what version of program Dragoon you played. Whether you are planning new versions of program.

Thanks!

We played at 75 moves in 20 minutes. I used dragon v4.1. Hardware:
6-core intel 980 at 3.3 Ghz, 64 bit windows 7
16 GB cache for databases (8 piece). 1 GB hash tables
Speed: 5-6 million evaluations/second.

I am still working out the last bugs in the program. I think i got them all fixed now except one. Once that is done i will make a new version.

While Dragon won the tournament convincingly, i think there is still much room for improvement. Both damage and Kingsrow appear faster and stronger and i am working on a dragon v5 to increase its node/second count. At the moment i am re-writing the movegenerator using new bitboard structures.

The first results for the movegenerator look promising; it is running perft (bulk-counting) at 130 Mnps at the root position.

Michel

[Rein Halbersma]

The first results for the movegenerator look promising; it is running perft (bulk-counting) at 130 Mnps at the root position.

Michel

Hi Michel,

Could you also report "nodes per second" instead of "leafs per second"? With nodes I mean the number of recursive calls to perft():

Code: Select all

int perft(int depth)
{
    ++node_counter;

    // rest of code
}

Rein

[MichelG]

Hi Michel,

Could you also report "nodes per second" instead of "leafs per second"? With nodes I mean the number of recursive calls to perft():

Code: Select all

int perft(int depth)
{
    ++node_counter;

    // rest of code
}

Rein

That would be nps/branching factor right? 130/6.4=20 Mnps. To be complete, below is my perftbulk code:

Code: Select all

INT64 perftBulk_bb3(BITBOARD3 *bb3,BITBOARD3 *movestack,int depth,int color)
{
	int nmoves;
	INT64 sum=0;
	int nr;
	BITBOARD3 *msn;

	nmoves=movelist_bb3(color,bb3,movestack);
	if (depth==1) return(nmoves);

	msn=&movestack[nmoves];
	for(nr=0; nr<nmoves; nr++) {
		sum+=perftBulk_bb3(&movestack[nr],msn,depth-1,color^1);
	}
	return(sum);
}

[Krzychumag]

It thanks for information Michel. And when you are going to drop the program Dragon v.5 and whether the base of ends in the program will be the same as in the version Dragon 4.

[Rein Halbersma]

That would be nps/branching factor right? 130/6.4=20 Mnps. To be complete, below is my perftbulk code:

Yes, that's correct. But having the counter inside the recursive function makes it easier to output this and also makes it better comparable to the regular search() NPS (basically you leave out the eval and reductions stuff).

Code: Select all

INT64 perftBulk_bb3(BITBOARD3 *bb3,BITBOARD3 *movestack,int depth,int color)
{
	int nmoves;
	INT64 sum=0;
	int nr;
	BITBOARD3 *msn;

	nmoves=movelist_bb3(color,bb3,movestack);
	if (depth==1) return(nmoves);

	msn=&movestack[nmoves];
	for(nr=0; nr<nmoves; nr++) {
		sum+=perftBulk_bb3(&movestack[nr],msn,depth-1,color^1);
	}
	return(sum);
}

I also make the optimization to return the bitcounts of the successor bitboards for depth==1.