Truus DXP-Server

Rein Halbersma · Post by **Rein Halbersma** » Mon Apr 13, 2009 14:28

BertTuyt wrote:Ed,Gerard,

see below position (derived from the game Boomstra - Zweerink, after the theoretical continuation 45 ..... 14-19 46 24-20 11-17 47 39-33 17-21 48 15-10 4x24 49 29x20) , in this position Black has to play. Im not sure if evolution function help here.
Anyway for Damage it is a draw ( 19-24), although there is some discussion if 19-23 is a win for Black.

<img src="http://fmjd.org/dias2/save/12396247798.png">

Bert

19-23 is a draw after 43-39, I was too fast to post the analysis, I didn't let Kingsrow calculate long enough

BertTuyt · Post by **BertTuyt** » Fri Apr 17, 2009 19:51

Rein, maybe a strange question.

* But as you are following closely the Dutch NK, and also use computers for analysis.
What is your opinion how the programs would score in this group ??

* Also the similar question to Feike, his son is the sensation so far !!!
I think you are , and should be proud on him.
Does Roel play against BoomstraDam, and how does he score ?

Bert

FeikeBoomstra · Post by **FeikeBoomstra** » Fri Apr 17, 2009 20:38

Of course I am very proud. Roel doesn't like to play against Horizon (as I call the program now). He only wants to play a quicky, but then the computer has a big advantage in finding combinations. So most of the times, he gets an advantage, and then he loose, due to a combination.

What I did to follow the NK: I made a small interface between livedraugths and Horizon, so I can select a game and this game is then real time downloaded, so I can analyze the entire game.

What I observed: in the games, which were considered as "sharp", my score in the middlegame was much too low, often even negative, and only at the end, the score was coming up. So, not enough strategic vision in the program.

Rein Halbersma · Post by **Rein Halbersma** » Fri Apr 17, 2009 21:25

BertTuyt wrote:Rein, maybe a strange question.

* But as you are following closely the Dutch NK, and also use computers for analysis.
What is your opinion how the programs would score in this group ??

* Also the similar question to Feike, his son is the sensation so far !!!
I think you are , and should be proud on him.
Does Roel play against BoomstraDam, and how does he score ?

Bert

HI Bert,

Two points. First, in my very humble opinion, top programs like Kingsrow are almost certainly unbeatable by all humans, with the possible exception of a handful of top grandmasters. I used Kingsrow to analyze many past world championship games and it either reproduces best human efforts or improves them. I know only a handful of very deep combinations and forcings that the program doesn't find immediately (mainly due to its very aggressive search truncations). I might not be knowledgeable enough about the game, but I know of no major positional weaknesses that Kingsrow currently might have. It's just a very balanced program that even at fast time controls plays an extremely solid style of draughts. And humans can forget about winning endgames altogether against programs with more than 7 piece databases

Second, winning against top human competition is quite another thing! Since grandmasters make very few tactical mistakes this means converting into a db win from advantageous late middle game positions. The main problem is therefore to generate interesting games that give opportunities to parlay the computational advantage into a win. This is where one runs into problems: top grandmasters state they could draw at will against almost perfect play, simply by playing very unimaginatively and exchanging pieces as often as possible. This difficulty for superior players to force inferior players into interesting games is the reason why there have been so many rule revision proposals. My guess is that it would take long matches before a win would be seen against top human competition even against the strongest programs.

Overall guess: top programs like Kingsrow would win a few games with no losses and end up near the top of the table. Other programs might follow closely, look at programs like Buggy and Flits that also had quite impressive results against GMs even 5 yrs ago.

Rein

FeikeBoomstra · Post by **FeikeBoomstra** » Wed Apr 22, 2009 13:42

progress 158 games exercise.

I didn't report on my progress, because I was loosing too many games from Truus. She is playing on a "voorpost" quite aggressive, so I needed quite some time to upgrade my voorpost evaluation (for the n-th time). Yesterday I drawed 19 games in a row, so I am making progress. To be honest, the 20th game I lost, but I have solved that problem in the meantime.

So now I start implementing the 158 game concept and keep my fingers crossed.

wellnesswrotter · Post by **wellnesswrotter** » Wed Apr 22, 2009 14:04

so....

that means Truus beats BoomstraDam still???

FeikeBoomstra · Post by **FeikeBoomstra** » Wed Apr 22, 2009 14:18

At second thoughts, I didn't like the name BoomstraDam, so the program is now called Horizon.

I can't say, wait for the 158 games result.

FeikeBoomstra · Post by **FeikeBoomstra** » Sat Oct 17, 2009 19:52

I am still in the process of executing the 158 games. Till now, I have never finished a complete set of 158, as every time I have collected a few lost games, I try to improve the engine, but then you aren't sure the improvement(s) didn't cause effects on the overall performance, so you have to start over again.

This is where I came so far: played 92, won 6, draw 67, lost 17, undecided 2. Not very impressive. But ok, still going.

What I would love to have is an extra function to the truus_dxp_server:
get the PV from truus (just a string copy of the PV panel of truus, including the score etc.)

Maybe Bert & Ed are willing to have a look at it? It would be very helpful to me.

Kind regards,
Feike.

Ed Gilbert · Post by **Ed Gilbert** » Sun Oct 18, 2009 00:58

Feike, I don't know if it is possible to obtain the pv for the server from truus. I ran truus under a debugger and did a search in the truus data space for the pv string that was displayed and could not find it. That probably means that the string is not formatted into a global buffer and will not be very easy to obtain. Maybe Bert knows something about this.

-- Ed

BertTuyt · Post by **BertTuyt** » Sun Oct 18, 2009 13:37

Feike, im convinced it can be done, but ....................... it might be not easy.

When Ed and I starting to discuss the possibilities of a Truus DXP-Server, we based this opportunity upon some assumptions regarding the internal architecture of Truus. As these assumptions, in a later stage, appeared to be valid, implementation (in the end) was "relatively" straightforward.

Now some thoughts about the "PV" and best score. As Ed already pointed out, I tend to agree with the assumption that there is no buffer where the PV is stored. But I basically have no clue. To find out, we need to locate the "Beste Variant" Window Handler, where all the steps are executed. Also there we will find clues about the memory location of Best Move, Best Score and PV.

It might even be (but again no idea so far) that the PV is dynamically constructed from the Hash-Table.

Another difficulty is that the Truus DXP-server use a polling mechanism to detect that the internal board position has changed, which is a signal that the move has played. But as a consequence, and related to the pondering of Truus, the old best Move/Score/PV, is already over-written.

So if we want to obtain this info, we need to switch of pondering (im not sure if we already know how to do this). And even with this change, it could be that the internal architecture does not enable a simple retrieve of the mentioned information.

One could in the end also inject code, to do this task, but then one really has to understand the "purpose" of all routines, and basicallay this is not the way i want to deal with this.

Maybe Ed and I could try to locate the "Beste Variant" Window Handler, but again I don't expect short term results.

Bert

FeikeBoomstra · Post by **FeikeBoomstra** » Sun Oct 18, 2009 14:19

OK, it's just a "nice to have" feature, based on the assumption that the TRUUS score is the statical evaluation of the the position at the end of the PV.
With that feature I could "automaticly" go to the end-position of the PV.

Anyway, thanks for investigating.
Feike.

Ed Gilbert · Post by **Ed Gilbert** » Tue Oct 20, 2009 03:20

Bert sent me some info about Truus' internal representation of the pv and score, and using this I was able to print them to the server's console. It seems to work quite well. I think to make this more useful I should create a pdn logfile of the game in the server and add the pv and score strings as comments in the pdn. At the moment the server polls the pv string every 100msec and prints it whenever it has changed, so there are sometimes 4 or 5 different pv's printed. The last one is of course the one that usually corresponds to the move that was made, but it might also be useful to have the pv's from the shallower searches.

-- Ed

FeikeBoomstra · Post by **FeikeBoomstra** » Tue Oct 20, 2009 09:36

This is a great accomplishment!

What about generating an autonomous message by the dxp server? I would like to have the info at the remote site.

Ed Gilbert · Post by **Ed Gilbert** » Tue Oct 20, 2009 16:42

I will have the server send the pv and search score in a dxp chat message after every move.

-- Ed

FeikeBoomstra · Post by **FeikeBoomstra** » Tue Oct 20, 2009 17:10

great, I'm looking forward to use it.

World Draughts Forum

Truus DXP-Server

whooh