Comments/Ratings for a Single Item
Ah, I was not referred to the page you gave, but followed the instructions on the posting input form, namely writing an e-mail to [email protected]. Unfortunately, the page you referred me to now does not seem to work either. First time I tried it refused my user ID because it contained periods, and now it does not want to accept me because it says 'e-mail address already in use'... But to get back on topic: I am currently trying the Minister/WDN in a normal Chess context on a 10x8 board (so no Capablanca pieces, but two Ministers in stead). It seems that to balance two Ministers, I need Q+R (Q+N seemed far too weak, Q+2N far too strong after ~10 games). That would make them worth approximately 7 Pawns. To get it more precise I have to wait until I have about 100 games.
It seemed the first games of two Ministers against Q+N were a bit lucky. They were losing significantly in 20 games against Q+R, so I went back to Q+N, and had it run overnight. In 115 games this gave a reasonably balanced score of 53.5%. The Pawn-odds score is usually somewhere around 62%, so this could indicate a ~25cP advantage for the Minister. OTOH, the statistical error in 115 games is 4.3%, or 35cP. So the deviation from 50% is not really significant, and indeed, after 90 games, the Ministers were behind, at 47.8%. As Q=950 (by definition) and N=300 (on 10x8), assuming equality would make the Ministers 625 cP each. Taking the 115-game result at face value would give 637 cP (+/- 17cP). I am now running two High-Priestesses against two Ministers.
HG, thank you. I appreciate the effort you've put into this. That there are and can be sets of numbers on pieces to compare and contrast gives designers a valuable tool to use, if it can be developed. I find it interesting that a limited piece with a maximum range of 16 squares and a maximum distance moved of 2 is apparently worth more than a rook on a 10x8. I do expect a similar result with the high priestess. Question: would the high priestess [1] and the minister [1] be adequate/reasonable replacements for both knights and both bishops on a FIDE 8x8 or Capa 10x8? I'd think if they replaced the rooks and gave a pawn, it'd be about even, plus or minus about a pawn, say. The brute force method you are apparently employing must eat up a lot of computer time. I'd love to ask you about all the shortrange pieces I've seen, as there are a good number of them out there. Another common class of pieces, besides sliders and shortrange, is the cannon-type. Have you done anything on these pieces, or any others? I'll see what I can do about getting you signed up as a member. Are you a wiki member? Please email me directly. Joe
Indeed, my method is very time consuming. But I am still developing it, and hope to make it more efficient. (For instance, I have not really established what the optimal time control is to play these games. I am using 40/2' with fairy-Max now, and 4/1' for my better engine Joker, but it turns out that Joker finds the same Pawn-odds score at 40/10' as at 40/2'. If that would be true for all scores, I could speed the process up by a factor 6!) I currently have one dual-core PC entirely dedicated to Chess testing (running 24/7). And I consider fairy-piece values one of the spear points of my research. I have too many spear points, though: I develop two Chess engines, micro-Max and Joker, also in variant versions, am currently the main contributor to the WinBoard open-source GUI (and in particular the only one that cares about variant support). I want to write a Shogi engine, and develop a much faster end-game-tablebase builder. I could use an extra computer, but I want to postpone my next buy until I can get an 8-core Nehalem. Because I want to develop a new SMP algorithm for tree search that scales better on large numbers of cores, and I really needs a machine to test this on. So for the time being, I will have to do with my 2.4GHz Core 2 Duo. But eventually I will do most fairy pieces. I will probably convert Joker to handle fairy pieces as well, as it can reach the same quality play in a fraction of the time. If you have a computer (which is likely, since you post here), you could easily do such tests yourself too, however. It is harder on the PC than on the tester! It is just a matter of letting it run, and once a day look at the result, set up a new position and start a run for that. I am currently running a test with 2 Ministers against 2 High Priests (starting at the A and C positions of Capablanca Chess, so in a context of all normal Chess pieces). This seems a pretty even battle, In fact, after 62 games, the High Priests are even leading (26+ 22- 14=, 53.2%). I guess I will let it run overnight. But they are very close, so the High priestess is also more a Rook replacement than a Bishop replacement. My Rook opening value on 10x8 is only 475. So 625 is even on the strong side for a Rook. (Why do you find that surprising? A Rook on 10x8 has a MAXIMUM of 16 moves, and most will be blocked on a non-empty board. All 16 moves of Minister and High-Priest are unblockable.) I have not tried your leaper replacements for Bishop and Rook yet, but some time ago I very precisely tested a similar piece, F+D, because I wanted it to investigate pair bonuses for color-bound pieces (and F+D is color-bound). A pair of these pieces tested nearly as strong as a pair of Knights, perhaps a quarter Pawn weaker (this was on 8x8, though). This would make it 290, as my Knight is 300 on 10x8. I don't know yet, however, how much of this is pair bonus. But if a color-bound piece with 8 unblockable moves can be wrth as much as a Knight (at least, in a pair), the non-colorbound pieces F+A and D+W must be similar or better, and thus good Knight replacements. It is funny that a pair of the F+D, which is the (color-bound) conjugate of the King, is worth nearly a Knight (when paired), while a non-royal King is worth significantly less than a Knight (nearly half a Pawn less). But of course a Ferz is also worth more than a Wazir, zo maybe this is to be expected.
After 150 games the Ministers lost to the High Priests by 42% (47+, 71-, 32=). That is an excess score for the High Priests of 8%, about 2/3 of the Pawn-odds excess score. That would suggest a High Priest is about 33 cP stronger than a Minister. Because the difference is so small, preliminary tests left me in the dark as to which piece was stronger, so in the test mentioned above I had to run them as exactly equal. This is the safest thing for not producing any self-fulfilling bias in favor of one or the other, and thus reliably determine who is strongest. (Note that the statistical error over 150 games is only 3.56%, so that the probability of an 8% swing (2.24 sigma) between equal pieces is only ~2%.) The disadvantage of exactly equal programmed piece values (625), however, is that any difference in strength is not fully expressed, because the side with the better piece will often trade it for the inferior one, not realizing it is better, annihilating his advantage. So I am rerunning the test now with the High-Priest programmed as 650.
OK, final report of the pilot tests (~100 games in a Chess context). The re-run of M vs HP with HP set at 650 did not make the HP mre valuable compared to when the engines thought they were equal. On the contrary, the result was 45.4% (51+ 64- 26=) for the Ministers, against 42% earlier. The 4.6% excess score for the High Priestesses corresponds to about 35 cP, or 17 cP a piece. That confirms the value 650 (against 633 for the Minister). Perhaps the first time the High Priestesses were partly lucky. Note that the standard error is 31 cP. Then I tried the 'Elephant' (F+A, or Elepherz, really) and 'Dababbah' (W+D, Wazibabbah), in pairs against a pair of Knights (so that the opponent actually had 4 Knights in the array RNNBQKBNNR against RNEBQKBENR or RYNBQKBNYR). Note that I started the Wazibabbahs at b1 and i1, because the on c1 and h1 there was a terrible development conflict with the Knights. EE vs NN was practically equal (I had already anticipated that, and programmed the value at 290 against the Knight's 300, for a slight discouragement of trading the imbalance away for no reason): 49.5% (39+ 40- 21=). So a _pair_ of E is worth 600. But, being color-bound pieces, I have no doubt that part of this is for the pair bonus, and that an unpaired E is worth less than half of it due to the inability to access the other color. Fairy-Max is not really a good system to test this, as it has no pair bonusses in its evaluation. As the piece itself is less valuable than the Bishop, (a pair of those isworth 750 on 10x8), my educated guess would be that the pair bonus is also smaller, about 30. That would make the E base value 285. YY vs NN gave a clear win for the Knights: 45.5% (37+, 46-, 18=). The 4.5% excess score translates to only 30cP for the YY pair, though. Thus each Wazibabbah comes out at 285. It seems an isolated F+A and D+B are equal in value, and that the F+A achieves its higher value only through the pair bonus. Finally the Commoner, which I already tested before once on 8x8. Again I tested a pair of them against two Knights. As expected, also on 10x8 a very clear win for the Knights: 40% (35+ 46- 14=). That corresponds to 70cP, or 35cP per Commoner, which thus comes out at 265 (very close to the preset of 260 I gave it, so no need for a second iteration). Note that this low Commoner value falsifies about every speculation that has been made about it (usually presented as 'King end-game value'), which all put it above the Bishop, around 400. It is not an effect of the bigger board, as I found nearly the same value on 8x8. This is opening value, though, and Betza has mentioned that the opening value of the Commoner is indeed very low, but that it gets very strong in the end-game. So I am testing that now, By setting up varies Pawn chains of 7 or 8 Pawns on 2nd and 7th rank (making sure each side plays each Pawn setup equally often in both directions), plus Kings on i1/i8, and then giving one side Knights on b1 and f1, and the other Commoners on b8 and f8. No other pieces are on the board, and I would say this definitely qualifies as end-game. 13 games is a bit early, but the Commoners are not doing very well here either (61.5% for the Knights, so far). So the preliminary conclusion of the piece values in a Chess context is: HP=650 M=630 N=300 E=285 (pair bonus=30) Y=285 G=265 P=85 It seems that all pieces with 8 targets cluster around 290, and those with 16 targets around 640, with very litle spread. (I measured an F+D earlier, and it came out at 580 for a pair, also fitting this pattern). For the Commoner this is a surprise (although it is not completely clear why others over-estimated it so much, as it is simply another piece with 8 targets, like the Knight, and much slower at that...).
The Commoners did make a strong come-back in my GG vs NN end-game test, and finally won a 133-game match by 55.3%. This shows that indeed they do perform vbetter in the end-game than in the opening, compared to Knights. Normally I would say that the 5.3% excess score corresponds to 40cP, so that each Commoner is 20cP than a Knight, i.e. 320. (In stead of their opening value 265.) Here that would be really premature, as I have not determined the end-game value of a Pawn. (My intuition tells me it must score better than the 62% for Pawn odds in the opening, as the best you could hope for in the latter case is to preserve the advantage to the ending through game stages that are still full of tactical surprises, which could easily erase the advantage. Unlike a Piece advantage, a Pawn advantage in general does not automatically grows during the progress of the game. You can only start building out a Pawn advantage when the board gets empty enough that it can be converted to a passer that can be safely pushed through enemy territory. So an end-game Pawn might give an excess score larger than 12%, and hence the 5.3% would correspond to less than 40cP.) But no matter what the end-game Pawn value is, it is clear that the G-N difference changes sign. This could be due to a drop in value of the Knight as well, and is not necessarily proof that the Commoner rises in value. According to Larry Kaufman's analysis of 8x8 games, a Bishop or Rook rise in value compared to other pieces (including the Knight) as well. He expresses this as a function of the number of Pawns present, and the B-N swing can be 50cP in going from positions with all Pawns to positions without Pawns. This is similar to what we observe for the G-N value. It would take far more end-game test matches, involving all piece types, to determine this. (In particular Q vs 3 minors would be useful.) Whatever the case, the relative Commoner increase is a comparatively small effect. It never ever even gets close to the commonly encountered estimate of 400.
You can upload the implementation of this game using my piece values for the time being from the following link: http://home.hccnet.nl/h.g.muller/GrtShatranj.zip Just unzip it anywhere you like, (the zip archive contains a single folder GrtShatranj with everything you need in it), go to the folder, and click the shortcut named 'Great Shatranj'. That should startup WinBoard in Great Shatranj mode, using Fairy-Max as engine. (WinBoard actualy thinks this is variant capablanca, but legality testing is switched off, so that doesn't matter.) If you want to paste in positions through a FEN, you should use the following letters (I was limited to what Fairy-Max accept, as I still have to make the piece identifiers programmable there rather than the fixed set PNBRQKEFWACHG): P = pawn N = knight E = elephant (AF) W = dababbah (DW) C = commoner (FW) M = minister (DWN) H = high-priestess (AFN) K = king For some reason it does not start to play automatically if you make a move; you will have to use the 'mode -> machine black' (or white) menu to make it play. BTW, there is no guarantee what the other variants will do in this Fairy-Max version, which I use to test strange pieces. So don't use the 'new variant' menu!
The way I read it, Minister and High Priestess are regulare FDW and FAN, which can jump to all their distant destinations... I do like Great Shatranj for the consistent application of a single idea, mnamely to replace all distant slider moves by a single 2-jump. But I don't think it stands any chance of replacing the Mad Queen game. Slider moves are simply too interesting to discard completely. I might include Great Shatranj and its pieces as standard variant in the Variant Server.
HG, George, sorry for taking so long to reply to your comments. I was looking over comments about piece strengths, and ran across several here by Mr Muller, and one by Mr Duke. [I suspect I addressed at least part of this in previous comments off-topic, but I should/need to tie it all together here.] I appreciate the comments you gentlemen have made. HG, your work on piece values is especially welcomed and I again would like to thank you for the time you have put into this question. The amount of computer work you have done is truly amazing. George, thank you for the comment and rating; HG has the right of it, however; the FAN and the WDN are the real pieces, with both the knight and either the alfil or dabbaba leap. In fact, this game design started my fascination with short range pieces that could leap. I find it surprising that seemingly no one else has used these pieces in games. I find them beautiful and logical pieces, as well as rather obvious, because they are really just cut-down, or better, shatranjized, BN and RN pieces. I'd be honored to have them appear on a variant server. EDIT: My memory is even worse than usual; Joshua Morris uses both pieces in Kozune [posted onsite].
John, thanks for the comment and rating. I believe I see the idea behind calling the NAF a Big Ferz; the NWD is not exactly a Big Wazir, though. Based on analogy, your Big Wazir would seem to not properly exist on a standard 2D board [although there is undoubtedly some strange geometry where it could exist as the counterpart to the Big Ferz]. I suspect the closest to the Big Wazir that might work on a 2D is the double DW, a warmachine that may move as a wazir or dabbaba twice in one turn, thus covering 4 squares in each of the 4 orthogonal directions. [I do use this piece in Grand Shatranj, as the lightningwarmachine.] It is not an exact analogy with the NAF, but does hit 16 squares, also. Interesting question; anyone else have any ideas for a 'Big Wazir' piece.
I am currently playing a round of this game with Joe at the Game Courier. And I really like it. In its simplicity, it has expanded the strategy needed to prosecute a decent game. A player cannot rely on a single line of assault to accomplish the mate, they will need to think in terms of a series of battles to reduce and penetrate the opponent's ranks. Without sliders, the players need to closely engage one another. This can create several areas of serious contention on the field. And each might equally lead to success, so that the opponent risks catastrophe if each are not taken seriously(particularly in the opening). Right now, Joe and I(or at least I am) are testing to determine the effectiveness of Pawn strutures against this large variety of leaping piece. So far, they seem to hold up well. Though the other pieces can quickly bypass them. In itself this is not a bad thing since the opponent can simple maintain a strong defense, and not readily abandon their Pawns. Those players who are familiar with the Mad Queen variant will find much that is familiar. They will not find this game difficult to learn, though application of the Mad Queen's common strategy may prove disastrous.
The relative attractancy of this sort of game lies just in this, namely that it is slow. One needn't exhaust one's brain. Yet it is complex enough to create strategical complexity. When chess was slow, in medieval times, it was extremely popular, also among women. The later faster game was more about performance and was regarded as less enjoyable by many people. However, the more devoted gamesters became even more enchanted. /Mats
The game can definitely let a player recover from an error. Maintaining the exchange ratio appears to be necessary. I anticipate that the current game I'm playing will exceed 70 turns. Looking at the potential of a hundred-turn game will often scare the impatient. ;-) Has anyone determined the various combinations of pieces needed in the endgame? I like the pattern that a Minister and Dabbaba take when cornering the opposing King.
Larry, Mats, thanks for your comments. You bring up a lot of topics. First, some statistics: of 10 games completed, 8 were won by resignation, one by time, in 33 moves [although the position was very poor for the loser - it appeared a resignation around turn 40 was clearly possible], and one by mate - with a breakaway Minister and High Priestess literally cornering the king, in 69 moves. The resignations lasted from 11 to 79 turns. So, for all games, # turns/game = 11, 20, 21, 33, 43, 45, 49, 50, 69, 79. While it's not a huge sample, it looks like hundred-turn game would exist, but be rare, and if the numbers hold approximately true, most games will end within around 50 turns or so. I don't know the numbers, but isn't this fairly close to FIDE norms, at most a bit longer, not a lot? Hey, Larry, has it occurred to you that you might just play a pretty good game of shortrange? Being a Jetan master and all? ;-) You came out of that slightly premature attack very well, and our game right now is almost at a standstill. We've achieved a balance of forces across most of the board. But the final battle is not even showing an outline yet; it's just the first faint stirring of pieces, with no form or center. And this speaks to both Mats comment about the appeal of the game being its slowness and your comment about having a chance to recover. A so-far common pattern in these games is that many show a major battle involving, and costing, about half the pieces per side, occurring early midgame. Then the 2 sides re-group, and often a final key battle is fought, with one side generally coming out of that the clear winner. The short games are when one side is clearly losing after the initial battle, and resigns. [Or when one side gets blitzed by the other through differential skill in handling these almost familiar but rather tricky pieces.] The 40-50 turn games are when one side has either lost the 2nd battle, or never got its pieces together enough after the first battle to fight effectively. I suspect the longer the game the better the players are, or at least they're very evenly matched, at whatever skill level.
Larry, if you go through the comments for this page, you'll find some comments by HG Muller on piece values there. In general, he values the minister at 6.33 pawns and the High Priestess at 6.50. The guard, he values for the endgame at 3.2, but more like 2.8 at setup. And he makes the comment that all the pieces in this game that are analogs to pieces that can mate in regular Capa can mate in this game. Start about in the middle of the 30 comments. Mats, you've made an interesting point in saying that one needn't exhaust one's brain in this game, which fits kind of next to Larry's comment about being able to recover in this game. In FIDE, tactics, from the nature of the pieces being generally infinite sliders, is always active. While you certainly could use a strategy, it's 'positional play', aka: tactics, which often determines the game, and always has at least an indirect effect on the outcome. In Great Shatranj, strategy is always active, but tactics tends to happen more sporadically, with intense bursts for 5 - 10 turns at a time, followed by a bit of strategic picking up of the pieces. In closed games, given that no piece here has a blockable move, I suspect the tactics would be more varied, intense, and far-reaching. These pieces are made for close-in, with wide, short footprints. The High Priestess attacks 8 forward squares [and 8 rearward], every turn, unblockably. Does the B+N?
Another nice Shatranj variant from Joe, this time on 10x8.
I'd tentatively estimate the piece values as P=1; N=3.38(=3.5 approx.); E=Y=2.695(=2.75 approx.); Guard(approx.=K's fighting value)=3.2; HP=MI=7.075(=7 approx.); R=5.5.
Hi Joe,
Just saw you post in another thread and hadn't seen you in a while. Good to see you drop by! Hope all is well :)
I've been meaning to ask you if you'd consider simplifying the promotion rule for this game. I remember that, after some discussion, the promotion rule for Modern Shatranj was simplified. I'm thinking the same reasoning may apply here. These promotion rules are somewhat difficult to implement when programming this game. Also, since the number of promotions to specific pieces is limited, the FEN notation for a game would need to be expanded somehow to keep track of which promotions have already been used. It seems like unneeded complexity to me but happy to hear your thoughts.
Cheers,
Greg
Also, since the number of promotions to specific pieces is limited, the FEN notation for a game would need to be expanded somehow to keep track of which promotions have already been used.
That is not true, is it? You can see what is on the board, and at all times would be allowed to increase that to what you had in the opening position (which can also assumed to be known). And you cannot promote to an Elephant on the shade where you already have one.
Pawns may promote to lost pieces, with restrictions. One pawn may promote to either (lost) member of the following 4 pairs of pieces: dababbas; knights; elephants; minister-high priestess. If promoting to a lost elephant, the pawn must promote to the opposite color of a surviving elephant.
I took this to mean that only one pawn may promote to a lost piece in each pair. In other words, once a pawn has promoted to knight, you cannot promote another pawn to a knight even if you lose another one. Upon further reading, however, I see it is not entirely clear. But the word "either" must have been chosen for some reason.
Are you doing this game for your program Chess V Greg. First off, I notice on your 'person info' page, the link to Chess V ....
http://sourceforge.net/projects/chessv
this doesn't seem to work.
I have actually done a zrf file for this, I did it years ago, but I notice it wasn't released, the file is titled 'great shatranj -test' lol, so I must of forgot. I see that I have made pawns able to promote to General, Minister or High Priestess :)
Joe, you need to change your pawn promotion rules, stop making life difficult for everyone :) .. and I sent you an email too, look in your junk email section hehe.
25 comments displayed
Permalink to the exact comments currently displayed.