It seems it cannot promote to bishop. At least in this position it wants to promote to queen with a score of 0.
[d]8/5k2/4p3/4Pp1p/5P1P/3Rn2K/6p1/8 b - - bm g1=B
I heard that Komodo prune minor promotions with exception of the knight promotion when it is giving a check. Anyways, I didn't know it was pruning at root.
It seems it cannot promote to bishop. At least in this position it wants to promote to queen with a score of 0.
[d]8/5k2/4p3/4Pp1p/5P1P/3Rn2K/6p1/8 b - - bm g1=B
I heard that Komodo prune minor promotions with exception of the knight promotion when it is giving a check. Anyways, I didn't know it was pruning at root.
Komodo considers all underpromotions at the root I thought too. Perhaps I rigged the generator to ignore everything but the knight, I will have to check.
But you are right about knight under-promotions in general, they are always considered if they give check and this covers the majority of cases (but not all of them) where it's best to promote to knight.
Don wrote:Komodo considers all underpromotions at the root I thought too. Perhaps I rigged the generator to ignore everything but the knight, I will have to check.
But you are right about knight under-promotions in general, they are always considered if they give check and this covers the majority of cases (but not all of them) where it's best to promote to knight.
Don, is there any reason for not generating underpromotions all the time?
The performance hit must be absolutely marginal (< 0.1 %), and even if in practical play underpromotions are very rare, people kind of expect that a 2800+ engine handles them correctly.
Don wrote:Komodo considers all underpromotions at the root I thought too. Perhaps I rigged the generator to ignore everything but the knight, I will have to check.
But you are right about knight under-promotions in general, they are always considered if they give check and this covers the majority of cases (but not all of them) where it's best to promote to knight.
Don, is there any reason for not generating underpromotions all the time?
The performance hit must be absolutely marginal (< 0.1 %), and even if in practical play underpromotions are very rare, people kind of expect that a 2800+ engine handles them correctly.
In promotion rich positions looking at several extra moves that are almost certainly not best can slow the program down. Have you found that not to be the case?
The total slowdown amortized over all positions is probably very small.
I probably should take another look at this. I think I could at least include under-promotions at deep levels without any substantial impact.
Does Houdini include all underpromotions in the main search as well as in the quies or do you have restrictions?