Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Masters of the Chess: on Chess Capablanca (10 x 8) 40 pieces
1 2 3 4 Previous Next  
Parent - By neoliminal (*) [us] Date 2007-09-18 19:42 Edited 2007-09-18 19:44
Ok, sit the queen in any non corner square on an 8X8 board and a 10X8 board.  How many squares does the Queen have access to?  Shouldn't that show the queen is stronger on a 10x8 board?

And if you can't be civil, don't debate at all.
Parent - - By Octopus (**) [de] Date 2007-09-19 10:36
Hi Ed, you know that I have a different opinion to that. Of course endgame thoughts are very important for theoretical conclusions. But in 8x8 Chess and in 10x8 Chess endgame table bases often are showing, that there is mostly no "logical" behaviour in the process of mating. Thus it mainly depends on tactical effects of complex circumstances how long a mating will need. There seems to be no way to derive merely from a given position how distant it would be away from a mate or draw decision, beside of calculating out the whole thing. Thus it seems not obvious to me why piece values should be derived from such obscure effects.

I have defined and modified an own evaluating model from which I have derived my piece values for SMIRF, where a 10x8 Queen comes out as a piece slightly stronger than an 8x8 Queen. It seems to be a highly trustable model, because it also gives 8x8 piece values very close to its experienced figures.

Please do not "kill" me here for contradicting to your point of view. Sometimes mature 10x8 Chess programs will decide on this ...

Reinhard.
Parent - - By GothicChessInventor (**) [us] Date 2007-09-19 16:45
Reinhard,

I always respected your model because it had a basis in mathematics. There are two ways of looking at pieces that are supposedly "stronger" on a larger board. One could say that pawns are "weaker" on the larger board, therefore, you need "more pawn value" integrated into the number assigned to a piece. If a chess queen on 8x8 = 9.0 pawns but a Gothic Chess queen on 10x8 = 9.5 because the pawn unit is weaker, that is understandable.

My numbers were derived from normalizing an 8x8 chess knight to be worth 3.0 pawns, and then measuring the 10x8 piece strength in terms of 8x8 pawn values.

In order words, if an 8x8 pawn = 1.0, then my 10x8 rook = 4.75, etc.

The values of Mr. Goofy over there state that 1 Archbishop + 1 Pawn = 2 Rooks, which is ridiculous.
Parent - - By Octopus (**) [de] Date 2007-09-19 17:06
Ed, here I also have to disagree. My model gives unnormalized values for a Pawn of: 8x8 -> 3.5 and 10x8 -> 3.6, which is about 3% increased. This is because on the Capablanca board there are only 2 of 10 border Pawns compared to traditional 2 of 8 border Pawns. Therefore this is raising the average value.
Parent - - By GothicChessInventor (**) [us] Date 2007-09-19 20:15
Well, in my estimation, when you have more of something, each something is not worth as much. On a board with 30 pawns per side, I would think losing 1 pawn would not be too much of a deficeit. So, on the board with 10 pawns, each is not as "scarce" as on a board with 8 pawns. A reduction of one pawn is only 10% of the total starting pawn count, on the 8x8 board the same loss is a 12.5% reduction.
Parent - - By Octopus (**) [de] Date 2007-09-19 20:30 Edited 2007-09-19 20:38
If I would accept this as argument, Ed, it must be applicatable for each and every piece, thus it would relatively neutralize itself. But I have to admit, that you are a very creative head. ;-) I would not have been able to produce this wild idea.

In my model the piece values are depending only on their gaits and reachability of board squares. Especially focussing on Pawns and their limited ability to take a journey to the right or the left, there seems nothing relevant to be changed for Pawns at a wider 10x8 board, beside of the fact, that the border is in average more distant to each Pawn. Thus in my view there is no reason to be seen, which should decrease the value of a Pawn on a 10x8 board.
Parent - - By GothicChessInventor (**) [us] Date 2007-09-20 04:10
Then think about this:

Is a Queen much weaker on a 4x4 board? This does not seem to make sense. King + Queen vs. King must have a max win of 3 moves. The Queen can reach more than half of the squares, I don't see how that is a weaker piece.

Let me ask another question:

On what size board would a Queen be twice as strong, at about 18 pawns?

You see what I mean now?

On larger boards, my piece values are computed using 8x8 pawn values. So, on a smaller board, the pieces "gain strength" and so do pawns. On a 4x4 board a pawn on the second rank becomes a Queen with 2 pawn pushes, how can it be weaker on a smaller board? Conversely, on a 100x100 board, how can a pawn be stronger if it must make 98 moves to promote?
Parent - - By Octopus (**) [de] Date 2007-09-20 06:52
Hmm, Ed, I see that while I am open for your arguments and trying to understand and evaluate those, you do not at all think about that, what I have said on this matter. Such a discussion tends to be a little bit one-sided and seems in danger to become uninteresting for me.

Well, with growing board sizes piece values are not growing unlimitly. Instead they are approaching to figures, which would represent a borderless geometry. You could approximate that situation for the big pieces by calculating values from a torus structure.

For the Pawns that situation is different, because their values are overlayed by their ability to final promote. This immanent nature will be unchanged, when the board merely widens by becoming more colums, why I prefer bord sizes like 10x8, which are keeping the 8 rows. You seem to see this yourself, because now you are focussing on different board size examples, where that number would also increase. Indeed, that would change the nature of the game of Chess for the aspect of promotions. But comparing 8x8 and 10x8 Piece values that point is not at all relevant.
Parent - By GothicChessInventor (**) [us] Date 2007-09-20 23:04
The pawn count is relevant.

Many endgame positions are such that promotion of a pawn is REQUIRED for the side ahead by 1 pawn to win.

Say you only had a board that was 5 files wide. You can see, with 5 pawns, you have only 5 "chances" to promote. Therefore, being 1 pawn ahead is meaningless if the pawns can all be swapped off the board. Therefore, you must PRESERVE at least one pawn, and, therefore, the pawn becomes more valueable.

Conversely, on a wider board, each pawn is slightly less important. Imagine a board 20x8 with 20 pawns per side. You lose one pawn during a missed tactic, so what? The pawn count is 20 to 19, no big deal.

Once the pawn count is down to 2 to 1, then it is a big deal.
Parent - By neoliminal (*) [us] Date 2007-09-21 18:19
I agree.  Each piece will reach a maximum value on a infinite board.  That value is probably within a few points of it's current value in FIDE chess, with the exception of the Pawn which has a value that is based entirely on the depth of the board.

The assumption of a board depth of 8 is paramount to this discussion of piece values in Capablanca or Gothic Chess because of the assumed value of 1 for a pawn.  If the board were 10x10, then the pawn value would be reduced, and if you used a value of 1 for it would increase the value of the other pieces proportionally.

The assumption that we are on an 8 depth board should be assume for the conversation.

Increasing the size of the board help any slider that can take advantage of the increase movement (and this is also an assumption of location... more on Gothic Chess Knights in a second).

This means that Rooks, Queen, and Chancellor are all increased in value on a larger board, as they can slide longer distances than they could on an 8x8 board.  Do Bishops also increase in value?  This is an interesting question because they do not gain more single move distance!  They are still limited to 8 maximum squares of movement.  This is an open question to me, and I am still pondering it.

On the value of Knights, their starting position in FIDE chess reduces their value by limiting their initial moves.  In the game of Gothic Chess, the opening moves for the Knight are dramatically increased because of their central positioning.  This increase of value in Gothic Chess is closer to the actual top theoretical value of the Knight than the FIDE Knight, which must move at least once to gain it's real potential.  The question of what the "real" value of the Knight is, ignore starting position, and if it increases on a larger board is interesting.  Since a Knight has a maximum of 8 possible squares.  The reduction of movement (and therefor value) for the Knight is almost always board edge.  Hence on a larger board, the value of the Knight increases to the theoretical maximum value based on a limitless board.
Parent - - By neoliminal (*) [us] Date 2007-09-18 19:52
And you wont answer this because you know what it shows.  The majority of those moves are King moves... getting pushed from one end of the board to the other.... Nothing at all to do with the strength of the Queen.
Parent - - By GothicChessInventor (**) [us] Date 2007-09-18 20:57
A published artificial intelligence author has told you that your values don't make sense.

A famous chess player has told you that your values don't make sense.

Example after example has shown that your values don't make sense.

You are not only ignorant, you refuse to learn from "those that know", making you the most pigheaded person I have encountered.

Your values are wrong. I will not waste any more time with you.

The End.

Have a great life with your superior, unproven values.
Parent - - By neoliminal (*) [us] Date 2007-09-18 22:25
Thanks Ed.  It was nice having you ignore any of my valid points and instead wave your hands in the air about your system's superiority.  Every question I ask ignored, only to be buried under three more questions from you.  It's your style of debate and you can't change so I guess I should give up trying to debate you.
Parent - - By GothicChessInventor (**) [us] Date 2007-09-19 01:44
You have no valid points. You're an argumentative idiot.
Parent - By neoliminal (*) [us] Date 2007-09-19 01:46
I'll repeat:

Put a Queen + Rook + King vs. King + Rook.  How much longer does that take on a 10x8 vs 8x8?

Even an argumentative idiot like you can clearly answer that question, even if I have to repost it.
Parent - - By Permanent Brain (*****) Date 2007-09-19 05:30 Edited 2007-09-22 12:24
Please Ed, don't waste your time. Just click his name, click ignore, and forget him. He is a troll, not worth listening too. Put him on the ignore list and he is simply gone... Enjoy this board and interesting conversations about Gothic Chess, chess in general or whatever [moderated]
Parent - By GothicChessInventor (**) [us] Date 2007-09-19 17:06
Permanent Brain,

I have to agree with you there. But just in case you are interested:

All statistics for up to 5 pieces for the 80 square board are available here:

http://www.gothicchess.com/db_5_stats.txt

The question that other person posted was ridiculous since it's such a lopsided advantage, one side being up a full queen. The more lopsided, the less information, (and possibly the less intelligent the person is if they think the position is interesting.)

Compare that to King + Archbishop vs. King + Knight + Bishop with a longest win of 202 moves.

http://www.gothicchess.com/endings/kakbn/game.htm

Now that's interesting information, it tells you the opposing sides are very nearly matched.
Parent - - By Octopus (**) [de] Date 2007-09-19 09:54
SMIRF uses following average piece values
(without to be overlayed positional impacts):

Q =  9.6005
C =  8.9090
A =  6.8916
R =  5.7112
B =  3.6637
N =  3.0556
P =  1.0000

Reinhard
Parent - - By neoliminal (*) [us] Date 2007-09-19 19:41
Can I post this to Wikipedia?
Parent - - By Octopus (**) [de] Date 2007-09-19 19:57 Edited 2007-09-19 20:02
Well, I have had published those values on my now inactive web pages.
It looked like following and you might enter those numbers as my values.

P.S.: I must admit, that I am about to refine my model, and in the coming
model the values will be depending on a given board emptyness then.
Parent - - By neoliminal (*) [us] Date 2007-09-20 00:26 Edited 2007-09-20 00:31
Would it be acceptable to simply use the memo values, as the values on the wikipedia are for human reference.  It's nice to see that someone else thinks sliders increase in value on a larger board. ;-)
Parent - By Octopus (**) [de] Date 2007-09-20 07:00 Edited 2007-09-20 07:20
a) I do not recommend to publish that memo values alone, because to avoid any impression, that they might be that precise.

b) As I have stated in another answer on Ed's arguments, big pieces' values are not growing unlimitly with the board size, instead they are approaching the value from a borderless plane.
Parent - - By Octopus (**) [de] Date 2008-05-03 14:44
After one year I have modified my approach a little bit, made it more precise. See its value table among others at:
http://www.10x8.net/Compu/schachveri1_e.html

N,B,R,A,C,Q for 8x8:
3.0000, 3.4119, 5.1515, 6.7824, 8.7032, 9.0001

N,B,R,A,C,Q for 10x8:
3.0556, 3.6305, 5.5709, 7.0176, 9.1204, 9.6005
Parent - - By h.g.muller (****) [nl] Date 2008-05-06 07:37
Considering that a position with all pieces on the board, except lacking an Archbishop for one side, and a Bishop + Knight + Pawn for the other side (such as would occur in play after the first trade) is lost in 80-90% of the cases, one can wonder what you mean by "more precise", or in fact, what your piece values describe at all.

Aren't piece values supposed to describe how strong the piece is in play, so that you can use them as a guide in trading material, as an indication if your prospects of winning the game would go up and down, making the trade? Don't we consider a Bishop less valuable than a Rook because, after giving the exchange, the winning chances for the side with the Bishop dwindle?

So why do you set the value of the Archbishop lower than that of other material, that would not nearly be enough in compensation to prevent you from losing the game (with 90% probability) if you actually traded the Archbishop for it? Even after trading an Archbishop for Rook + Knight, the side with R+N will consistensy lose (unless he has extreme positional compensation of course, like 2 Pawns worth of King safety), as can be easily checked through playtesting by anyone who cares to know.
Parent - - By Octopus (**) [de] Date 2008-05-06 22:05
Play with 9 Knights against 4 Archbishops at the starting base arrays, and see what will happen ...
Parent - - By lkaufman (*****) Date 2008-05-07 02:36
This is a strange remark; piece values are averages of situations from real games, how is this pairing of any relevance? We don't value the queen in normal chess based on games with 3 queens vs. eight minor pieces!
Parent - - By Octopus (**) [de] Date 2008-05-07 06:14
Because H.G.M.'s piece values are derived from matches between different armies, I name such an extreme one, with a huge pawn unit advantage in his piece value system for the Archbishops' side, for to demonstrate, that this approach has to be corrected from the bad elephantiasis payload, as I have discussed in chessvariants.org comments.
Parent - - By h.g.muller (****) [nl] Date 2008-05-07 06:45 Edited 2008-05-07 07:14
You did not answer the question, Reinhard. That was not about the piece values I determined, but about your piece values. So let me repeat it:

What do these piece values mean, and how are we supposed to use them?

Apparently not by judging if an A  vs R+N trade, or an A vs B+B trade is good, in the opening! Because these piece values predict your material value is going up by them, while in fact they are as losing as trading a Rook for a Bishop. Are you saying that these 'piece values' cannot be used to calculate the desirability of trades of unequal material, because after the trade the players then have different armies?

If so, it seems these piece values are totally meaningless and useless quantities, as any set of random numbers would predict that trading of equal material is neutral...
Parent - - By Octopus (**) [de] Date 2008-05-07 09:45
Piece values have to be used what they are thought for: to evaluate a board situation before applying detail evaluation corrections.

You seem not to agree with my Archbishop average value, and I answered, that your calculation for your value estimation has been based on games of different armies. And there, so I think, an additional term has to be applied in the detail evaluation, the elephantiasis correction. To prove its existence and a need for that approach I described an extreme example 9*N vs. 4*A, which shows, that the superior pieces are not that strong as they seem to be merely based on their average exchange values.

There are indeed positions, where it will make sense to exchange an Archbishop e.g. against N+B, or as in that extreme example against N+N.
Parent - - By h.g.muller (****) [nl] Date 2008-05-07 10:09 Edited 2008-05-07 10:15
There are positions where it is good to sacrifice a Queen for a Pawn. That can't be used as an excuse for presenting a set of piece values where Q=0.75.

What you further say doesn't answer the question, but just rephrases it. We know very well what the classical piece values N=B=3, R=5 etc. are "thought for": they are thought for providing the part of the evaluation dependent on material present (as opposed to where exactly it stands on the board). And they can be used to predict who is winning the game (in a statistical sense), by simply adding them. So that he who has Q, where his opponent has only R+N, is in bad shape if he doesn't have an enormous amount of positional compensation.

But apparently that is not what your piece values are thought to do: adding them to evaluate an A vs R+N position (all other material being equal)  would produce 7 vs 8.5, where the side that is heavily winning has 7. Apparently you blame that on "detail evaluation corrections", that somehow compensate about abnother 3 Pawns worth of value independent on position (i.e. only based on material present). I would say that having such terms in your evaluation subverts the concept of piece values to the point where the values you give are completely meaningless, misleading numbers. If you have terms in your evaluation that make a "detailed evaluation" of a position with A vs R+N (say rn1bqkbcnr/pppppppp/10/10/10/10/PPPPPPPP/2ABQKBCNR w Kkq - 0 1) evaluate as favorable for white, you cannot say that the piece value of A=7 and that of N=3 and R=5.5. These would be at most partial piece values then, a large part of the true piece values being hidden in your "detail corrections". For the classical piece values in normal chess, there aren't any such "detail corrections".

So show us in detail how you would evaluate the material component of the position given above!
Parent - By h.g.muller (****) [nl] Date 2008-05-07 15:59
Of course I meant to say at the end of the second paragraph: He who has R+N against his opponent's Q is in bad shape, as R+N=8, and Q=9.
Parent - - By Octopus (**) [de] Date 2008-05-07 17:09
That position has a great tactical instability: I give two evaluation lines of SMIRF MS-174d-0,
depending on which size is about to move:

02:20.7 (14.01=) +3.910 1.Rxj8 (=/=) Ci6 2.Rj3 Nh6 3.e4 Ci7 4.Qe3 e5 5.d3 Qe6 6.Nh3 Nc6 7.Cg3 Be7 8.Cg6+ hxg6
00:50.7 (13.01=) +3.871 1.Rxj8 (=/=) Ci6 2.Rj3 Nh6 3.e4 Ci7 4.Nh3 f6 5.Rj8 Nc6 6.Cg3 Qg6 7.Cxg6+ Cxg6 8.Bf3 Cg3+
00:17.5 (12.01=) +3.818 1.Rxj8 (=/=) Ci6 2.Rj3 Nc6 3.e4 Cd6 4.Cg3 e5 5.Nh3 Bh4 6.Cc3 g5
00:07.2 (11.01=) +3.746 1.Rxj8 (=/=) Ci6 2.Rj3 Nc6 3.e4 Ch4 4.g3 Cg6 5.Nh3 Nh6 6.Ad3 Cxg3+ 7.hxg3 Nd4
00:02.5 (10.01=) +3.754 1.Rxj8 (=/=) Ci6 2.Rj3 Nc6 3.e4 Ch4 4.Cg3 Nh6 5.Bh5 Nd4
00:01.2 (09.01=) +3.693 1.Rxj8 (=/=) Ci6 2.Rj3 Nc6 3.e4 Cg6 4.Bh5 Cf4 5.Rj8
00:00.4 (08.01=) +3.543 1.Rxj8 (=/=) Ci6 2.Rj5 Nh6 3.e4 Nxj5 4.Bxi6 hxi6 5.Qe3 Nc6
00:00.2 (07.01=) +3.543 1.Rxj8 (=/=) Ci6 2.Rj5 Nh6 3.e4 Nxj5 4.Bxi6 hxi6 5.Qe3 Nc6
00:00.1 (05.01=) +3.543 1.Rxj8 (=/=) Ci6 2.Rj5 Nh6 3.e4 Nxj5 4.Bxi6 hxi6 5.Qe3 Nc6
00:00.0 (02.01=) +3.848 1.Rxj8 (=/=) Nc6 2.Rxi8 Cxi8 3.Nh3

01:26.4 (13.01=) +7.186 1...Rxj1 (=/=) 2.Ci3 Rj6 3.e4 Nc6 4.Cb3 f6 5.Cxb7 Rb8 6.Cc5 Nh6 7.f4 Qi4
00:43.4 (12.01=) +7.172 1...Rxj1 (=/=) 2.Ci3 Rj6 3.Cb3 b6 4.Nh3 f6 5.Cg3 Nh6 6.e4 Cg6 7.f4 Bc4+
00:20.8 (11.01=) +7.088 1...Rxj1 (=/=) 2.Ci3 Ci6 3.Cxj1 Ch4 4.Ad3 Cxh2+ 5.Bxh2 Nc6
00:07.9 (10.01=) +7.156 1...Rxj1 (=/=) 2.e4 Rxi1 3.Cxi1 Nh6 4.f4 Nc6 5.d3 Nd4
00:03.0 (09.01=) +6.736 1...Rxj1 (=/=) 2.Ci3 Ci6 3.Cxj1 Ch4 4.Ad3 Cxh2+ 5.Bxh2 Nc6
00:01.9 (08.01=) +6.912 1...Rxj1 (=/=) 2.Ci3 Ci6 3.Cxj1 Ch4 4.Ad3 Cxh2+ 5.Bxh2 Nc6
00:01.0 (07.01=) +6.924 1...Rxj1 (=/=) 2.Ad3 Ci6 3.e4 Rxi1 4.Bxi6 hxi6 5.Cxi1
00:00.0 (07.01+) +3.500 1...Rxj1 (=/=) 2.Ad3 Ci6 3.Axh7+ Bxh7 4.e4 Ch8 5.Nh3 Rxh1 6.f4
Parent - By h.g.muller (****) [nl] Date 2008-05-07 17:21
Your Rooks can make captures by jumping over Pawns!? :-O

Are you sure you have not put Smirf in Xiangqi mode, so that it thinks they are Cannons? ;-)
Parent - - By h.g.muller (****) [nl] Date 2008-05-07 17:53
Ah, I see. The FEN was invalid, and your GUI did not catch the error, and left some squares empty.
Try this one in stead, then:

rn1bqkbcnr/pppppppppp/10/10/10/10/PPPPPPPPPP/2ABQKBCNR w Kkq - 0 1
Parent - - By Octopus (**) [de] Date 2008-05-08 17:41 Edited 2008-05-08 17:47
Why should SMIRF catch an error within a legal FEN?
(A "/" mostly might be accepted as skipping to next row.)

[Event "SmirfGUI Different Armies Game"]
[Site "MAC-PC-RS"]
[Date "2008.05.08"]
[Time "19:07:18"]
[Round "25 min / game + 10 sec / move"]
[White "SMIRF MS-174c-DN (Derek Nalls values)"]
[Black "SMIRF MS-174d-0 (without El. corr.)"]
[Result "*"]
[Annotator "RS"]
[SetUp "1"]
[FEN "rn1bqkbcnr/pppppppppp/10/10/10/10/PPPPPPPPPP/2ABQKBCNR w Kkq - 0 1"]

1. Nh3 f6 {(11.01=) +1.418} 2. e4 Nh6 {(11.01) +1.367} 3. f4 Nc6 {(10.01=)
+1.422} 4. d3 e5 {(10.00) +1.447} 5. f5 d5 {(09.35) +1.301} 6. Ci3 Be7 {(09.20)
+1.463} 7. Bf2 dxe4 {(09.36) +1.412} 8. dxe4 Bf7 {(10.03=) +1.516} 9. c3 Qd7
{(10.01) +1.625} 10. Bb3 O-O-O {(10.01) +1.596} *

I still cannot see any dominating from White's side.
Parent - - By h.g.muller (****) [nl] Date 2008-05-08 18:02 Edited 2008-05-08 18:11

> Why should SMIRF catch an error within a legal FEN?


Because otherwise you would waste time on treating it as if it was specifying the position that was meant, perhaps?

I do not consider FENs that don't have the same number of squares on each rank legal.
And if you are in favor of unique representations, you should certainly not allow that...

But let's not get into that.

What I see is 10 moves made by Smirf. What I don't see is how you can conclude anything
from that about domination or not, as none of the sides is checkmated yet, and after the moves
the position for black is just as bad as the initial one. And in no way the disastroius material imbalance
was resolved; they just traded a Pawn.

The point is that if you play this position between equally strong players, white wins 70% or so,
and black must be very lucky to even salvage a draw. So the question is, why does Smirf say
about +1.5 for such an extremely precarious position?

Is this due to wrong material evaluation, or does it see some mysterious compensation in the
way the pieces are placed on the board? If so, how would it evaluate material alone?

My conjecture is that because you use a very wrong piece value for the Archbishop, and that
whatever "detail evaluation" you do apparently is not able to compensate for that.
Parent - - By Octopus (**) [de] Date 2008-05-08 20:12
Well, Harm, you give a higher value to the Archbishop than I do.
We do this because of different reasons. I have published a model
to estimate such values. You have made some experiments using
different armies. I suppose, that the programs involved with that,
have had implemented your value model. I am not convinced of
the validity of your approach, because there are two obstacles to
accept it: there is the Elephantiasis effect, and there is a possibility
for a self fulfilling prophecy, if not experimenting with engines
having different value models implemented.

May be others will accept your model, but still except of me. It is
merely that very Archbishop, which establishes the main difference.
There will be a time, where this puzzle will be solved ...
Parent - - By h.g.muller (****) [nl] Date 2008-05-08 21:16
You are dodging the question, Reinhard. Did you see my piece values mentioned anywhere in that question? So why are you babbling about them all the time?

I asked you to show us how you would evaluate the material in the given position, as it is apparently not by simply adding the piece values you give above, but seems to require some mysterious elephantiasis correction. Which apparently is not a positional term, as it depends only on material present. So how does your prescription for evaluating material work out in this explicit example? Is the +1.5 for black all material, as I suspect?

One thing is clear, though: YOU SUPPOSE WRONG! The programs I used to learn that in the position above the Rook + Knight are crushed by the Archbishop did NOT implement my piece values at all. As I explained elsewhere, the fact that R+N are crushed by A in this position is almost totally independent of the piece value implemented for the Archbishop. If you tell the engines A =7, (and R=5, N=3), it crushes the Rook plus the Knight. If you tell them A=8, it crushes the Rook + Knight. If you tell them A=9, it crushes the Rook + Knight. No matter what you implement for its value (within a reasonable range, of course; you must not make it negative or something silly like that), an Archbishop crushes a Rook+Knight. It is as simple as that.

But if you say that you intentionally program your engine to consider the position for black better anyway, because you do not want to know that it is a heavily lost position, and without any playtesting can ASSUME black is better, based on some square-counting argument tangential to playing real Chess, then evrything is clear, and we can terminate this discussion. Because people reading this do not have to believe either of us, and can simply playtest the position with any engine of their choice (e.g. Fairy-Max, which has user-adjustable piece values, or Smirf, or TSCP Gothic, which has also low Archbishop value), and watch with their own eyes how the Archbishop crushes the Rook+Knight. And then they will all see that you haven't the slightest idea what you are talking about, when you talk about piece values... That it is all totally unfounded guesses based on fancy, but totally wrong numerology...

That is the nice thing about science: there is an OBJECTIVE REALITY that is accessible to everyone. And in that objective reality, an Archbishop crushes a Rook + Knight, in the above position, and any position similar to it (i.e. quiet, non-tactical positions, with little or no positional compensation for either side). And he who will continue denying what everyone can check for himself will be ridiculed in the same way we ridicule people who maintain the Earth is flat...
Parent - - By George Tsavdaris (****) Date 2008-05-08 21:29
Sorry that i'm replying without following this conversation (since it's too big and have no time right now):
I guess you both disagree about Archbishop is not so strong and you say it's too powerful while Reinhard says as always:-) that it is not, and both sides giving all kind of arguments and possible experiments and procedures to measure that.

My proposal is simple: Why not create a Smirf with your(HMG) values, and let it play against current Smirf with current values?
200-300 fast blitz games(form random starting positions) would be enough.

In my opinion would be enough to show that you(HMG) are right. :-)
Parent - By h.g.muller (****) [nl] Date 2008-05-09 06:38
Well, for one, I cannot change Smirf's piece values, as Smirf is not open source, and its piece values are not configurable at run time (AFAIK). Furthermore, on alll versions of Smirf I tried this so far, playing blitz games only produces one result: a forfeit on time (unrelated to if it is winning or losing). I don't know if that has been fixed, and if Smirf can play 1 min (or even 5 min) games now.

But even if Smirf could play such games, testing piece values by pitting to engines against each other is a notoriously unreliable method, that might need far more games than 300. The problem is that, unless the value of the Archbishop you program into the engine is very wrong (like smaller than a Rook or larger than a Queen), it still will exchange the Archbishop for an inferior piece combination in only ~10% of the games if the value is only off by 2 Pawns or so. Because you would need complex trades for this, capturing multiple pieces for one. And the opportunity to make such trades does not occur very often. Especially if they have to occur by chance, as the opponent does think these trades are bad, and therefore does not expect us to make them, and therefor has no incentive to offer them to us (as a trap).

So in most games, (80-90% of them), A will be traded for A, which is always a neutral trade to both engines, no matter how they value A. And in such games, the programmed value of A will not have exerted any influence on the game result at all; both sides have been enjoying the Archbishop equally long. So of the 300 Blitz games, you might have 270 games that are totally irrelevant for the question which piece values are better, and only 30 games where the side witth the inferior piece values would have a 70% certain loss because of the bad trade it voluntarily made. So it scores on average 9 points out of those games, in stead of the expected 15 between equal opponents. That is a systematic bias against the bad values of only 6 points. But you would have to detect those 6 points amongst the statistical noise caused by 300 games. While it would already have been highly unreliable if you would have had to detect it amongst the noise produced by 30 games. (30 games is in general not enouugh to make a reliable conclusion, although with a bias as strong as 70-30 you might get away with it).

So what you propose might only work if you play 3,000 games, (giving you the equivalent confidence of a 30-game test if the trades occur in 10% of the games), or if you would actually select only the games where A is traded for different material, wait until you have 100 of those, and then apply the statistical analisys to that set. Then at least you eliminate the random noise in the result of the irrelevant games, which were masking the result of interest.

It would be far easier to play positions like the one I give above, where the imbalance is already present in the initial setup, and thus in 100% of the games. Then you need only ~100 games in total, as every game counts.
Parent - - By Octopus (**) [de] Date 2008-05-09 05:02
Harm, so excuse me for the wrong assumption of using only one value model within your experiments. It has not been clear to me, how such a diversification has had or has had no influence in the outcome of your experiments.

Supposed, your value model would be right, there would be theoretical explanations for that, too. I am waiting for such arguments.
Parent - By h.g.muller (****) [nl] Date 2008-05-09 06:15
Well, you will be waiting forever, as they will not be coming. From me, at least. I have no need for such a theory. The only "theory" I believe in, is that if you typically lose after giving an Archbishop to capture a Rook+Knight, (all else being equal), the Archbishop has a higher value than a Rook+Knight. Especially so if that higher value also correctly predicts that the chances are approximately equal after trading an Archbishop for a Chancellor, or trading Archbishop+Pawn for a Queen.

You have been obscuring the issue with this elephantiasis correction, (which might be needed to make better predictons of ridiculous and unreachable positions with 9 Knights, where the additive piece-value model fails in calculating the strength of a complete army), but it is really very simple:

Either the piece values you give are meaningless quantities, because they cannot be used to calculate anything when people do not have exactly equal material, and a large "elephantiasis correction" is needed to see that the (1) QCARBBNPPPPPPPPPP army is stronger than a (2) QCRRBBNNPPPPPPPPPP army.

Or: even with this correction your system of material evaluation, using your piece values, predict that army (1) is weaker than army (2), while in fact army (1) beats the cr*p out of army (2). In which case your piece values are useless as well.

In both cases the piece values you give are no good by themselves. And that conclusion is completely independent from what other people give as piece values, or, in fact, if any other people give piece values at all. A system of material evaluation that is not able to get a positive score for wnning piece combinations, and a negative score for losing ones, is a useless system.
Parent - - By lkaufman (*****) Date 2007-09-23 03:21
     These values look strange to me. Although I'm not an expert on this type of chess, I do know that the Archbishop is considerably more valuable than its two component pieces, just as the queen is in normal chess. In both cases this is partly because the combined piece lacks the restriction to one color that the bishop suffers. Here the Archbishop gets only a 0.17 bonus, which is way too low. Also the queen bonus is way too low, as in normal chess it is at least a pawn. Furthermore the rook can't be worth nearly two knights! I think this table is not even close to accurate.
Parent - - By Octopus (**) [de] Date 2007-09-23 15:25 Edited 2007-09-23 15:31
Simply make a better model and include arguments for it. Maybe it would be convincing then. My approach seems to be working fine in SMIRF, where are no extra filters for good or bad (to prevented) trades. Having an evaluation model, where such exceptions have to be implemented in chess engines, it is disqualifying itself as a global model. So prove your arguments by supplying a 10x8 engine with your (unrestrictedly working) value model. SMIRF welcomes every new opponent program!

P.S.: I always have to add following note: if trades are made, not only the average piece exchange values (as specified in my model) are vanishing from the board, also the modified positional influences of involved pieces have to be seen. SMIRF is strictly unifying both aspects when calculating trades. Therefore such piece values de facto do not exist seperatedly from board scenes.
Parent - - By lkaufman (*****) Date 2007-09-23 17:15
I am not a programmer myself, just an expert on chess evaluation functions. Naturally if there are other positional values involved, the nominal piece values may be very misleading. If you would like for me to propose a set of piece values (and the logic behind them) for you to test I would be glad to do so, but they would be ¨raw¨values that might need to be corrected if your positional values for each piece do not average zero.
Parent - By Octopus (**) [de] Date 2007-09-23 17:19
Simply send me an email with your preferred values, then I would send you an appropriately modified SMIRF for testing purposes.
Parent - - By GothicChessInventor (**) [us] Date 2007-09-25 04:54
Hi Larry,

The hardest thing for me to try and figure out is the value of Archbishop + Knight versus the Queen. Which is stronger, and by how much? In 8x8 chess, the Queen is worth 3 minors. In Gothic Chess, the Archbishop + Knight is three minors, but on just 2 squares, so it is worth more. But... knights are weaker on the 10x8 board.

So while it is safe to say Archbishop + Knight > Queen on an 8x8 board, I think on a 10x8 board the Queen has more power because of the dilution of the Knight.

If you look at the 5-piece tablebase stats for Gothic Chess on the 80 square board you see:

Archbishop + Knight to move:
Longest win = 33 moves; Wins 36.8% of the time, draws 57.7% of the time; Average win = 12.2 moves

Queen to move:
Longest win = 48 moves; Wins 51.5% of the time, draws 45.8% of the time; Average win = 14.3 moves

This seems to support Q > A + N, but by how much is the real question.
Parent - - By lkaufman (*****) Date 2007-09-25 17:24
     First, I would say that endgame tablebase statistics have little to do with the average value of a piece in an entire game. I would say the a queen vs. archbishop is pretty much the same as a rook vs. a knight, as the added bishop power should help both pieces about the same amount. Since even in Gothic chess two knights are surely better than a rook in general (perhaps not in the endgame), I would say it follows that Archbishop plus knight should be superior to the queen by about the same amount. The way to verify this would be with database statistics from actual games, but I don´t know if such databases exist for Gothic and if they do whether the average level of play is high enough to tell us much. Keep in mind that rooks are much better in the endgame than earlier, and therefore pieces with rook power included are also weaker in general than any endgame based studies would suggest.
     In sum, everyone seems to be undervaluing the archbishop.
Parent - - By GothicChessInventor (**) [us] Date 2007-09-25 19:49
Here are two snapshots from a game that might help. The left diagram below is white to move.



There is a way for white to pick up either a Queen and Pawn or a Chancellor and Pawn for an Archbishop + Knight. From the left diagram:

1. Nxd5 cxd5 2. Ng5 (intermezzo) Ad8 3. Ab5 produces the diagram above and on the right.

The Archbishop has a strange type of skewer where the Bishop array hits the Chancellor and the Knight aura can capture the Queen.

The question is: should this combination have been initiated?
Parent - By lkaufman (*****) Date 2007-09-25 20:46
Speaking only theoretically, not analyzing past your diagram, I would say that it is wrong to exchange archbishop and knight for chancellor and pawn.
Up Topic Rybka Support & Discussion / Rybka Discussion / Masters of the Chess: on Chess Capablanca (10 x 8) 40 pieces
1 2 3 4 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill