Internet engine matches

Discussion about development of draughts in the time of computer and Internet.
Post Reply
Krzychumag
Posts: 145
Joined: Tue Sep 01, 2009 17:31
Real name: Krzysztof Grzelak

Re: Internet engine matches

Post by Krzychumag » Sun Aug 19, 2012 09:22

Michel tested in programme options PERMANENT BRAIN during game. Very a lot of mistakes are.

1. The program isn't saving moves for black and white during the game.

2. During the game when I played black while hitting the pawn, the program made a move but left the pawn on his place.

Ed Gilbert
Posts: 752
Joined: Sat Apr 28, 2007 14:53
Real name: Ed Gilbert
Location: Morristown, NJ USA
Contact:

Re: Internet engine matches

Post by Ed Gilbert » Tue Aug 28, 2012 13:49

It's available for download now:

http://mdgsoft.home.xs4all.nl/draughts/beta4.1.html
Hi Michel,

I installed it on 3 computers (XP 32-bit, Vista 64-bit, Win7 64-bit), but could not get it to run on any of them. On the 64-bit machines no user interface appeared, and after about 30 seconds I got a popup message from Windows that the program had stopped responding. On XP the interface window is displayed, but there is some error message about unable to connect to the engine. I have .net 4.0 installed on all 3 machines. Is there some other requirement for it to run?

-- Ed

MichelG
Posts: 244
Joined: Sun Dec 28, 2003 20:24
Contact:

Re: Internet engine matches

Post by MichelG » Wed Aug 29, 2012 13:22

Ed Gilbert wrote:
I installed it on 3 computers (XP 32-bit, Vista 64-bit, Win7 64-bit), but could not get it to run on any of them. On the 64-bit machines no user interface appeared, and after about 30 seconds I got a popup message from Windows that the program had stopped responding. On XP the interface window is displayed, but there is some error message about unable to connect to the engine. I have .net 4.0 installed on all 3 machines. Is there some other requirement for it to run?

-- Ed
I suspect dragon has problem creating a memory mapped file due to insufficient user rights. I haven't had the opportunity yet to look into it further.

Michel

BertTuyt
Posts: 1316
Joined: Wed Sep 01, 2004 19:42

Re: Internet engine matches

Post by BertTuyt » Tue Sep 04, 2012 19:22

I worked on a more robust/reliable Search-Routine based on PVS, MCP, LMR and FHR.
As already posted, this is the base of the Damage changes/improvements.
In a later phase i will activate some other enhancements.
Next to that I also made some Evolution Function changes (like Breakthrough Routine).

As the prove of the pudding is in the eating, Damage played the "traditional" 158 game match against KingsRow.
Game/Match settings as usual 10 min Game, 1 core and no pondering.

The Match Result:
KingsRow 15 Win, Damage 5 Win, Draw 137, Unknown 1

At least not a result which is too bad, and it seems that Damage is improving step by step....

Now need to analyze the 15 Games lost by Damage, if they reveal some common clue. :)

Keep you all posted.

Bert

Rein Halbersma
Posts: 1628
Joined: Wed Apr 14, 2004 16:04
Contact:

Re: Internet engine matches

Post by Rein Halbersma » Tue Sep 04, 2012 19:58

BertTuyt wrote:I worked on a more robust/reliable Search-Routine based on PVS, MCP, LMR and FHR.
As already posted, this is the base of the Damage changes/improvements.
In a later phase i will activate some other enhancements.
Next to that I also made some Evolution Function changes (like Breakthrough Routine).

As the prove of the pudding is in the eating, Damage played the "traditional" 158 game match against KingsRow.
Game/Match settings as usual 10 min Game, 1 core and no pondering.

The Match Result:
KingsRow 15 Win, Damage 5 Win, Draw 137, Unknown 1

At least not a result which is too bad, and it seems that Damage is improving step by step....

Now need to analyze the 15 Games lost by Damage, if they reveal some common clue. :)

Keep you all posted.

Bert
That are encouraging results! Apropos keeping us "posted": why don't you post those winning/losing games for our entertainment? :-)

More generally, I think that the computer draughts world is already so small that keeping stuff closed is not really productive, it's much more fun sharing your ideas/code (at least for me).

Rein

BertTuyt
Posts: 1316
Joined: Wed Sep 01, 2004 19:42

Re: Internet engine matches

Post by BertTuyt » Tue Sep 04, 2012 20:30

Rein, thanks for your reply.

You are right the Draughts World is very small, at least I guess only (or less) then a hand full of programmers post on a regular basis.
No problem to post a file with the win/lost games, I was not sure if anyone was interested, outside Ed.

With respect to sharing ideas, also no issue.
Thats why i indicated the elements my Search is based on which was PVS, MCP, LMR and FHR.
If this is not sufficient, I will share all details you need.... :D

Bert

BertTuyt
Posts: 1316
Joined: Wed Sep 01, 2004 19:42

Re: Internet engine matches

Post by BertTuyt » Tue Sep 04, 2012 20:39

Here they are... :D

If someone would read/analyze the WL-games and provide some hints for me/Damage how/where to improve, this is highly appreciated.

Bert
Attachments
WL dxpgames September-2012.pdn
(17.81 KiB) Downloaded 116 times

TAILLE
Posts: 968
Joined: Thu Apr 26, 2007 18:51
Location: FRANCE

Re: Internet engine matches

Post by TAILLE » Wed Sep 05, 2012 23:30

Hi Bert,
BertTuyt wrote:I worked on a more robust/reliable Search-Routine based on PVS, MCP, LMR and FHR.
As already posted, this is the base of the Damage changes/improvements.
In a later phase i will activate some other enhancements.
Next to that I also made some Evolution Function changes (like Breakthrough Routine).
I easily imagine we all work on all these algorithms in order to try and build the best search routine.
As you know I do not use the PVS but a MTD-f best like procedure. I never tried the MCP procedure.
I tried several implementation of the LMR and FHR procedures. My current conclusion is that LMR might become unuseless if you choose an agressive FHR procedure. I never tried to implement the MCP procedure but here again, depending on how agressive is your FHR procedure, I believe the MCP cannot improve your search.

BTW, in any case the goal of a reduction is to accept to not solve some kind of problem in order to solve many others, more frequent.

As an exemple of problem difficult to resolve for a computer see the following position:
Image
White to play and win

For a human it's not very difficult (a human see here the oppotunity to block completly the position) but for a computer .... ?

My feeling is the following : if your program is able to find this problem that means that your restriction/pruning mechanism is perhaps not agressive enough and you program may be weak in real games!

What is your feeling?
Gérard

Ed Gilbert
Posts: 752
Joined: Sat Apr 28, 2007 14:53
Real name: Ed Gilbert
Location: Morristown, NJ USA
Contact:

Re: Internet engine matches

Post by Ed Gilbert » Thu Sep 06, 2012 04:00

I agree with you Gerard. A program that uses no search reductions may solve these kinds of problems easily but will do poorly in real games. It is important to use more CPU cycles on the better lines and aggressively reduce inferior lines.

-- Ed

Ed Gilbert
Posts: 752
Joined: Sat Apr 28, 2007 14:53
Real name: Ed Gilbert
Location: Morristown, NJ USA
Contact:

Re: Internet engine matches

Post by Ed Gilbert » Thu Sep 06, 2012 21:41

I installed it on 3 computers (XP 32-bit, Vista 64-bit, Win7 64-bit), but could not get it to run on any of them. On the 64-bit machines no user interface appeared, and after about 30 seconds I got a popup message from Windows that the program had stopped responding. On XP the interface window is displayed, but there is some error message about unable to connect to the engine. I have .net 4.0 installed on all 3 machines. Is there some other requirement for it to run?
With some help from Michel I was able to get the beta version of dragon running on my computers. I have to run it as administrator. I tried DXP, and found some problems with dragon running as a Follower, which Michel is fixing, but I was able to play some DXP GAYP games with dragon as Initiator. We should to be able to run a match of 2-move openings with his next update.

-- Ed

BertTuyt
Posts: 1316
Joined: Wed Sep 01, 2004 19:42

Re: Internet engine matches

Post by BertTuyt » Sun Sep 16, 2012 11:28

Herewith the results of the most recent match Kingsrow - Damage.
As usual 1 core search, no pondering, 10 minutes game - time.
Kingsrow 3 wins, Damage 7 wins, Draw 148.

Bert

BertTuyt
Posts: 1316
Joined: Wed Sep 01, 2004 19:42

Re: Internet engine matches

Post by BertTuyt » Sun Sep 16, 2012 12:07

Here are the WL Games.

Bert
Attachments
WL dxpgames 16-Sept-2012.pdn
(11.69 KiB) Downloaded 115 times
WL dxpgames 16-Sept-2012.pdn
(11.69 KiB) Downloaded 111 times

Rein Halbersma
Posts: 1628
Joined: Wed Apr 14, 2004 16:04
Contact:

Re: Internet engine matches

Post by Rein Halbersma » Sun Sep 16, 2012 13:11

BertTuyt wrote:Herewith the results of the most recent match Kingsrow - Damage.
As usual 1 core search, no pondering, 10 minutes game - time.
Kingsrow 3 wins, Damage 7 wins, Draw 148.

Bert
So Damage beat Kingsrow?? That's a milestone!

Ed Gilbert
Posts: 752
Joined: Sat Apr 28, 2007 14:53
Real name: Ed Gilbert
Location: Morristown, NJ USA
Contact:

Re: Internet engine matches

Post by Ed Gilbert » Sun Sep 16, 2012 14:32

Congrats Bert. I knew it was only a matter of time before you guys would catch up.

Was that match using 6-piece db kingsrow and 7-piece damage like the others?

I am looking forward to trying your new damage.

-- Ed

BertTuyt
Posts: 1316
Joined: Wed Sep 01, 2004 19:42

Re: Internet engine matches

Post by BertTuyt » Sun Sep 16, 2012 15:07

Hi Ed, yes it was.

So Kingsrow had a disadvantage basically, although I'm not able to quantify it, I guess that with a 7p against 7p Kingsrow and Damage are almost equal (or when one program performs better it is Kingsrow). The number of draws already reveals that it is a close call.

Anyway there is still much work to do here.
I need to implement the parallel search, better time control, pondering, resolve the bugs in the outpost and locks, and the search still goes crazy at times.
Also i still need to activate some code which i implemented during holiday (related to search extensions), which is not tested so far.
And last but not least, i need to compile a 8p DB to be really competitive.

So I'm sure that in the end Kingsrow is the better program, but at least Damage is making steps in the right direction.
I'm not sure what Damy, Maximus, Dragon, Sjende Bleyn is doing, but it feels that Damage is in the top3 , which is ok for me.

Bert

Post Reply