hgm wrote:My suspicion is that most of the Elo difference is caused by the Bishop pair. I assume that its value is implemented in the material tables; as it could be implemented there by simply initializing the tables differently, it would be silly to implement it by a seperate eval rule, which would take execution time. But of course any self-respecting program that does not use materia tables, still implements the Bishop pair through a dedicated rule.
So it is really a bit of a meaningless comparison. You cannot switch of the subtle effects of the material table without removing very basic knowledge that almost very engine has.
I think that the bishop pair bonus does not worth most of the 68 elo difference that is reported by Osipov and my guess is that the gain is mainly because of non optimal value of pieces that you start with them(for example overevaluating pawns so you need bad trade tables for most cases that you trade a bishop for 3 pawns).
Well, the Bishop pair is worth 50 cP, and opportunities to trade a Bishop for a Knight are plentiful.
But in case you have both material tables (or piece-square tables) and piece base values, the way you divide the piece value over those is arbitrary. So you can set it up such that the base values are total crap (e.g. all ero), and then you would probably lose even more than 68 Elo by switching off the material tables.
As I read somewhere that the Rybka material tables are unreliable (in case of multiple Queens or underpromotions), I suppose that an effort was made to make the base values the best possible approximation of the value incuding the material-tables, so that the values in the latter can remain as small as posible, and it does not hurt too much when they are applied wrongly.
hgm wrote:My suspicion is that most of the Elo difference is caused by the Bishop pair. I assume that its value is implemented in the material tables; as it could be implemented there by simply initializing the tables differently, it would be silly to implement it by a seperate eval rule, which would take execution time. But of course any self-respecting program that does not use materia tables, still implements the Bishop pair through a dedicated rule.
So it is really a bit of a meaningless comparison. You cannot switch of the subtle effects of the material table without removing very basic knowledge that almost very engine has.
I don't think so. Many "bishop pair" scores (ours included) are dynamic and depend on the mobility of both bishops. I just increase the mobility score for each bishop if I have a pair, so that I won't try to keep a pair when one or both is "bad".
hgm wrote:Well, the Bishop pair is worth 50 cP, and opportunities to trade a Bishop for a Knight are plentiful.
But in case you have both material tables (or piece-square tables) and piece base values, the way you divide the piece value over those is arbitrary. So you can set it up such that the base values are total crap (e.g. all ero), and then you would probably lose even more than 68 Elo by switching off the material tables.
As I read somewhere that the Rybka material tables are unreliable (in case of multiple Queens or underpromotions), I suppose that an effort was made to make the base values the best possible approximation of the value incuding the material-tables, so that the values in the latter can remain as small as posible, and it does not hurt too much when they are applied wrongly.
The bishop pair is not worth 50cp in all positions. If the bishop has no mobility, it does not do a thing with regard to the other bishop and the pair is worthless. In fact, getting rid of the pair is often a good plan when one of the pair is "bad" and is going to remain so for a long time. I've not done a static bishop pair score in at least a couple of years now. It is certainly better than nothing, but it causes you to enter horribly cramped positions, just to maintain a pair of bishops that are almost worthless, at times.
That is positional evaluation, and has nothing to do with piece values or material tables. The latter by definition have nothing to do with positioning of the pieces, the best they can do is describe the average or typical case. They ar meant to decide questions like "without knowing anything else, when given the choice between a position with two Bishops or with B+N, which should I pick to maximize my chances".
Last edited by hgm on Mon Apr 27, 2009 9:43 pm, edited 1 time in total.
hgm wrote:That is positional evaluation, and has nothing to do with piece values or material tables. The latter by definition have nothing to do with poitioning of the pieces, the best they can do is describe the average or typical case. They ar meant to decide questions like "without knowing anything else, when given the choice between a position with two Bishops or with B+N, which should I pick to maximize my chances".
I realize that. And that is why "bishop pair" is not folded into those pc/sq tables. Those are per-piece, and simply increasing the value of one bishop won't cut it for a bishop-pair bonus.
BTW positional evaluation _does_ have something to do with material values. That's the point, in fact, to be able to recognize when one piece is worth more than another and should be removed, regardless of the old 1-3-3-5-9 weights.
bob wrote:I realize that. And that is why "bishop pair" is not folded into those pc/sq tables. Those are per-piece, and simply increasing the value of one bishop won't cut it for a bishop-pair bonus.
Yes, but the material tables can include it. They are exactly intended to describe non-linear effects of the materil composition. But they are blind to positioning of the pieces.
In Xiangqi the material tables are even more important than in Chess: there are pieces that have no value at all unless the opponent has offensive material: te pieces that are forbidden to cross the river. They can only be used in defense, and derive their value solely from what the opponent has to attack. You can be 4 such piece ahead, and it can still be a draw by rule (insuff. mating material).
hgm wrote:My suspicion is that most of the Elo difference is caused by the Bishop pair. I assume that its value is implemented in the material tables; as it could be implemented there by simply initializing the tables differently, it would be silly to implement it by a seperate eval rule, which would take execution time. But of course any self-respecting program that does not use materia tables, still implements the Bishop pair through a dedicated rule.
So it is really a bit of a meaningless comparison. You cannot switch of the subtle effects of the material table without removing very basic knowledge that almost very engine has.
I don't think so. Many "bishop pair" scores (ours included) are dynamic and depend on the mobility of both bishops. I just increase the mobility score for each bishop if I have a pair, so that I won't try to keep a pair when one or both is "bad".
"Bishop pair"(not matter how bad is) is a "potential", and it is worth about 1/2 pawn (or more). I can say from my own experience that it is very hard to convert pawn more into win when play against bishop pair. One or even two bishop could be bad, but sacrificing one (or sometimes two) pawn to open position is good way to keep balance.
bob wrote:I realize that. And that is why "bishop pair" is not folded into those pc/sq tables. Those are per-piece, and simply increasing the value of one bishop won't cut it for a bishop-pair bonus.
Yes, but the material tables can include it. They are exactly intended to describe non-linear effects of the materil composition. But they are blind to positioning of the pieces.
In Xiangqi the material tables are even more important than in Chess: there are pieces that have no value at all unless the opponent has offensive material: te pieces that are forbidden to cross the river. They can only be used in defense, and derive their value solely from what the opponent has to attack. You can be 4 such piece ahead, and it can still be a draw by rule (insuff. mating material).
How can the material balance tables include something that does not apply in all positions? The bishop pair score is one of those ideas. Putting it in the pc/sq table is problematic. Because it should not be included in any position with two bishops for the same side. There are other constraints that have to be met before a realistic bishop pair score can be factored in. A bishop pair with a bad bishop is worthless and should actually get a negative score, not a bonus for two bishops.
A bishop and knight vs a rook and pawn is a more natural case. The two minors are almost always better except in certain types of endgames where the extra pawn becomes decisive.
hgm wrote:My suspicion is that most of the Elo difference is caused by the Bishop pair. I assume that its value is implemented in the material tables; as it could be implemented there by simply initializing the tables differently, it would be silly to implement it by a seperate eval rule, which would take execution time. But of course any self-respecting program that does not use materia tables, still implements the Bishop pair through a dedicated rule.
So it is really a bit of a meaningless comparison. You cannot switch of the subtle effects of the material table without removing very basic knowledge that almost very engine has.
I don't think so. Many "bishop pair" scores (ours included) are dynamic and depend on the mobility of both bishops. I just increase the mobility score for each bishop if I have a pair, so that I won't try to keep a pair when one or both is "bad".
"Bishop pair"(not matter how bad is) is a "potential", and it is worth about 1/2 pawn (or more). I can say from my own experience that it is very hard to convert pawn more into win when play against bishop pair. One or even two bishop could be bad, but sacrificing one (or sometimes two) pawn to open position is good way to keep balance.
You put the bishop pair score up there that high and play against a good GM and I will show you why it is wrong. Been there several years ago and had long conversations with a GM friend (Roman Dzhindi) about this very issue. He is the one that convinced me that just a plain pair score is wrong. And he showed me several games against Crafty where he demonstrated this clearly.
As in physics, potential energy is worthless. It is only when it is transformed to kenetic energy that anything happens. I don't want to wreck my pawn structure, so that I can keep a pair of bishops, one of which is locked in behind my pawns and will never contribute anything to the game.
