Shartranj!

[adams161]

To clarify as to why i added shtaranj without putting extensive effort into it, I never found a substantial shatranj community on icc and the top players werent very good. pulsar can easily compete for the number 1 or number 2 spot on iccs shatranj best list which might only have 40 active plyers that play shatranj on icc in tottal.

I probably should add support for the bare king rule, But right now its not much more than a material count evaluate. Most all humans will find pulsar a challenge. The book was written becasue in any wild without a book the computer cant play on icc because it will repeat lines. But if you look at the book its extremly shallow and just designed to randomize its moves on move 1.

Pulsar has some wilds that some effort has been put into. Its not the best at any wild. the current goal of the last year has been to make essentially a decathalon opponent. An opponent that can play the icc wild 29 which picks at random from say a dozen wilds, and like any decathalon player it will have events ( wilds ) its good at, some its fair, and some its poor.

there have been programs in the past ( never released ) that could play all the wilds on icc. Currently no program is capable of playing wild 29 on icc ( random wilds ) 'help wild29' on icc. I think i'm getting close to where pulsar can play.

Mike

[adams161]

I checked on icc. Since it got removed from wild 29 ( random wilds ) it seems nobody is playing shatranj. What is needed perhaps are some computer opponents on icc to generate interest. Here is the icc best list for icc wild 28 or shatranj. Normally it would list 20 players but there are not 20 players with 6 games in last month.

>Best list for Wild type 28:
>1749 cornfused
>1647 monkeyman722
>1623 blitzspieler
>1595 FeC-Urza
>1568 OxidisedLizard
>1406 chesswhiz
>1353 frank001
>Number of established and active players: 7 aics%


this is pulsars rating ( now inactive ) on playinasec which was handicapped to only search to depth 3 i believe ( playinasec never plays at full strenght and moves instantly)

>Shatranj (w28) 1859i 46 0 4 50 1859 > win loss draw total rating

so you see even playing in handicap mode out of 50 games it never lost.

I never did bother to test at full strenght.

Mike

[hgm]

Indeed, I noted Shatranj is virtually unplayed. I looked for weeks several times a day, in order for an opprtunity to test WinBoard. Only after 3 weeks there was a game which I could use to test ICS 'observe' mode, and one of the players was kind enough to play an unrated Shatranj game against Fairy-Max. (Which worked as far as WinBoard was concerned, but Fairy-Max forfeited on time.)

The match Fairy-Max - Pulsar2007-9e-33 ended in 29-3.

Fairy-Max did not lose a single game out of its 96 games agains Pulsar and Dabbaba. Yet, I goot the feeling that in the middle game, Pulsar is tactically equal to Fairy-Max. In many games it can force Fairy-Max to a negative evaluation. But than it suddenly makes an unsound sacrifice, and the game is lost. Usually Pulsar's score increases to +2 or even +4 after such a losing sacrifice (and Fairy-Max jumps to +2 or so).

So I think there is something badly wrong with Pulsar's evaluation, in particular with its piece values (since there seems to be little else, and since the magnitude of the effect is so big). I put up all the games at

http://home.hccnet.nl/h.g.muller/games/STT1.zip

Unfortunately, Pulsar violates WB protocol so badly that its depth/score information does not make it into the PGN.

[adams161]

You know i had to look to see what pulsar's material values are for shatranj. I think i just plugged in some values based on the knight and rook being 'valuable' and that was the end if it.

this is what pulsar is using:

pawn 100
knight 350
bishop 250
rook 600
queen 200

bishop and queen have different names and mobility in shatranj but you get the idea probably.

any thoughts?

Mike

[adams161]

looking at my moblity and piece placement values were i have them, they are probably to agressive as well and need to be controled. pawns seem to get up to 60 points to advance making a pawns pottential 160, and a knight can get up to 30 points for being in the center. other pieces dont get a bonus based on placement on the board.

rooks seem to get high values up to 30 points if they control a lot of spaces, and rooks seem to have open file and on teh 7th rank bonuses.

queens and bishops are dampened with no bonsues at all just raw material.

king can get up to 60 points for being in center during an endgame.

so it looks like the rook knight score is inflated, and the pawns may be to hyper to advance with up to 60 points for moving to 7th rank. but with their base of 100 they go to 160 but can only realize being 200 for being a queen.

queens and bishops have little value.

king may have to much value in end game for being in the center considering many shatranj games dont end in mate.

Mike

[adams161]

Its not enough to log on and sit their for 30 minutes and wait. Having programmed in wilds its critical to get the wild players to play you. To make the non wild players willing to take a chance at playing.

keys to getting games

1) long logins On icc you can type set noa 1 ( for noautologout ) if you have a free computer account. Log on, issue a seek with gameend, set noa 1 and see how many games you get in 24 hours.

2) talk to the players. As much as you can be around teh engine while its playing and if the engine gets a tell respond. If they say why cant i move my pawns 2, tell them!

when players see that seek on their icc seek graph for hours on time you begin to develope people who know that your engine is available for play. This is very differnt than 'hunting fo games' i.e. logging in and waiting for a game and logging out.

On fics its much tougher. well they dont offer shatranj and have no plans to offer any more wilds it seems. On fics teh default newbie formula is something that only allows blitz and standard, not wild. They say newbies woudlnt know how to play. Well the solution is to do similar to blitzin and have a warning box that says are you sure you want to play this type of game its wild. But fics has not given control to the interface writers. They unilatterlly exclude wilds. Average fics player has a haze idea what wild is. He doesnt see any wild seeks on his seek graph or sought.

I ran wild tournaments on icc as a tomato manager. Its possible to seel anythying to that hungry and eager crowd looking for a 'game' of some sort. I ran wild 9 tournaments. 2kings. Ideally you need a few people interested, (your regulars) and others just willing to give it a try. willing to learn.

Mike

[hgm]

Well, in Fairy-Max the values are mainly also blind guesses . The only piece that I really measured the value of, is the Ferz (Queen), in a context of normal Chess pieces. But I evaluated it as a pair, and as it is a color-bound piece, there probably is a pair bonus involved, of which I have no idea about the magnitude. (Fairy-Max is not really ideal for measuring this, as it does not score Bishop pairs in its eval.)

The values I use are

Code: Select all

P =  50
N = 325
B =  70
R = 500
Q = 150

The R and N value are simply copied from normal Chess, as these are the same pieces as used there. The Ferz (Q) was measured, as I said. P and B (Elephant) are educated guesses.

This leads to very large disregards for Pawns, which can be clearly seen in the games: nearly always Fairy-Max has 3 or more Pawns fewer than the opponent in the late middle game, but slightly more valuable pieces. I considered increasing the Pawn value somewhat to suppress this behavior. But OTOH, it seems to win all games where it does this. So who am I to say that it is a mistake?

[hgm]

OK, Mike sent me a new version of Pulsar, that knows the opening position and has altered piece values, and knows the baring rule (Pulsar2007-9e-34).

I set it up for a gauntlet against Fairy-Max and the previous Pulsar version, 32 games each. Games can still be watched at

http://80.100.28.169/gothic/shatranj.html

[hgm]

In your implementation of the baring rule, you created a fatal bug:

Pulsar now takes the last piece of the opponent with his King, even if that piece is defended. E.g.

[d]8/8/7b/4K3/2kQ4/8/8/8 b

Black (Pulsar) plays Kxd4 here, presumably because he thinks this wins by baring the white King. You should test the condition for bare King only after the move of the side for which you test it.

Indeed Pulsar also has a problem with mate distance. It consides positions where it can be mated in 1 (rather then where the King can be captured in 1) as illegal positions. As a result, it claims stalemate in positions where it is to be mated in one, if not in check. (Not that that matters much for the result, of course. The way it handles the baring rule is much more detrementel, even compared to not knowing it at all.)

[adams161]

I shoudl have that barring bug fixed today. I wanst able to get any games that went to endgame to test yesterday.

How is the match going otherwise? your link for the match was not working.

Mike