bhlangonijr wrote:Thanks for the quick reply Robert, but what I really want to know is, what does Houdini's static evaluation give for the original position I have posted?
The static eval for the initial position (8/p1p5/6pp/PPP2k2/8/4PK2/8/8 w - - 0 43) is +0.09.
One could add pattern recognition for the Q-side pawn formation to decide statically that it's won, but in practice it's dangerous to do so. There are many positions in which Black would be able to generate counterplay and/or stop White's passed pawn with the King by combining offensive and defensive threats.
Remember the classical Réti position?
jacobbl wrote:I was wondering if there are any good testsuits for pawn endings? I'm thinking if you have a set of positions and a score saying if it is a win, draw or lose, it would be interesting to check it against the evaluation function. Hoping one would find high scores for the winning positions and low scores for the losing positions.
This is a surprisingly difficult problem. Most pawn dominated positions that I have are either locked pawn (therefore deep search) positions or endgame positions easily resolved by endgame table base.
I do not know if this helps but here is a short list of pawn endings*
*Taken from the book Pawn endings by A. Cetkov and fundamental Chess Endings by K. Muller & F. Lamprecht
tano-urayoan wrote:[d] k7/2p1pp2/2P3p1/4P1P1/5P2/p7/Kp3P2/8 w - - 0 0
What about this position?
or this [d] 8/5pp1/7p/5P1P/2k3P1/2p5/2K2P2/8 w - - 0 0
static good for black, after 0.05 seconds, white winning.
[d] 8/1p4p1/2k2p1p/5P2/4P1PP/3K4/8/8 w - - 0 0
static = equal, search quickly says +3
All of these show that static knowledge will have a hard time with these kinds of positions as there are too many special cases, while a good search + good evaluation can resolve them quickly...
tano-urayoan wrote:According to the original sources yes . Maybe not all positions are correctly evaluated. You want me to post the solutions from the books here?
The since we discuss only eval (and not search), solution is irrelevant. Only identical evals for each rotation/permutation are important.
Thanks for the list of positions. Do you know which of these positions are winning for white, winning for black or a draw. Then one could compare this to the evaluation, and hopefully get good scores on the winning positions, and close to 0 scores on the draw positions.
tano-urayoan wrote:According to the original sources yes . Maybe not all positions are correctly evaluated. You want me to post the solutions from the books here?
The since we discuss only eval (and not search), solution is irrelevant. Only identical evals for each rotation/permutation are important.
One can get an evaluation by running a deep search, presumably to find whether it's win, loss or draw. Then the only issue is whether the evaluation without a search is reasonable.
It's difficult to get an evaluation of the root position without a search unless you have the source code for the program of interest or the program has a feature to support this. So perhaps a 1 ply search is a reasonable compromise?
Dann Corbit wrote:The since we discuss only eval (and not search), solution is irrelevant. Only identical evals for each rotation/permutation are important.
Sorry but then I do not understand this comment:
Dann Corbit wrote: Do these have a list of solutions?
. Maybe not all positions are correctly evaluated. You want me to post the solutions from the books here?