Author: Robert Hyatt
Date: 14:24:41 10/03/05
Go up one level in this thread
On October 03, 2005 at 16:04:51, Will Singleton wrote:
>On October 03, 2005 at 11:43:22, Robert Hyatt wrote:
>
>>On October 02, 2005 at 13:49:37, Will Singleton wrote:
>>
>>>White wins easily by promoting and exchanging queens, but most comps can't see
>>>it.
>>>
>>>[d]8/5p1P/8/8/6K1/5P2/1k6/q7 w - -
>>>
>>>Amateur failed to exchange queens, and drew the game. I'm not quite sure why,
>>>seems like a simple search should be enough to handle this. Oddly, if the
>>>exchange is forced, the win is seen immediately. This happens with other progs
>>>as well. I thought I had taken steps to eliminate root-position dependencies in
>>>the search, but maybe something else is happening.
>>>
>>>Will
>>
>>
>>Here is what you need:
>>
>>Your program probably understands that if you remove the black queen, white is
>>winning because of the "outside passed pawn"??? (ignore the fact that the pawn
>>promotes instantly for now). If so, you need one more critical piece of
>>knowledge.
>>
>>Your program needsd to realize that if things are not obvious, no pawn to
>>promote, no distant passer on the board, but its king is closer to the remaining
>>pawns than the enemy king, the game is _still_ won.
>>
>>Set up a position like white/black pawns locked up at a4/a5, white passer at h2,
>>both kings near the center of the board. The easiest way to win is to push the
>>h-pawn. Black has to go over and stop it, and once it is gone, white is too
>>close to the remaining black a-pawn and the game is over.
>>
>>If you don't catch this case, you will play for a distant passer, but then
>>refuse to give it up, and once your king gets to the distant passer to defend
>>it, you can't force the promotion, and when you wait too long, as your king
>>finally gives up the passer, black eats it and then follows you across the board
>>and hems in your king on the a-file or else black gets to the a-file and the
>>game is a draw. But if you just push and get the black king too far away, you
>>win easily...
>>
>>Crafty plays this move instantly with no endgame tables of any kind, and shows a
>>big +score for white immediately, due to the above...
>>
>>
>
>Thanks for the info, it will be helpful in other situations. This one is
>different. After h8=Q a1=Q, we get this position:
>
>[d]7Q/5p2/8/8/6K1/5P2/k7/q7 w - -
>
>From here, Crafty takes 15 ply (over a minute on my comp) to see the exchange,
>then another ply to get a winning score. This is the problem I was referring
>to, albeit clumsily.
>
>Will
Here is crafty's output on my box, no endgame tables of any kind:
3 0.07 1.84 1. Qxa1+ Kxa1 2. Kf5 Kb1 3. Ke5
3-> 0.08 1.84 1. Qxa1+ Kxa1 2. Kf5 Kb1 3. Ke5
4 0.09 +1 1. Qxa1+!!
4 0.10 2.82 1. Qxa1+ Kxa1 2. Kf5 Kb1 3. Kf6 Kc2
4. Kxf7
4-> 0.12 2.82 1. Qxa1+ Kxa1 2. Kf5 Kb1 3. Kf6 Kc2
4. Kxf7
5 0.14 2.82 1. Qxa1+ Kxa1 2. Kf5 Kb1 3. Kf6 Kc2
4. Kxf7
5-> 0.16 2.82 1. Qxa1+ Kxa1 2. Kf5 Kb1 3. Kf6 Kc2
4. Kxf7
6 0.18 2.82 1. Qxa1+ Kxa1 2. Kf5 Kb1 3. Kf6 Kc2
4. f4 Kd3 5. Kxf7
6-> 0.20 2.82 1. Qxa1+ Kxa1 2. Kf5 Kb1 3. Kf6 Kc2
4. f4 Kd3 5. Kxf7
7 0.22 2.82 1. Qxa1+ Kxa1 2. Kf5 Kb1 3. Kf6 Kc2
4. f4 Kd3 5. Kxf7
7-> 0.24 2.82 1. Qxa1+ Kxa1 2. Kf5 Kb1 3. Kf6 Kc2
4. f4 Kd3 5. Kxf7
8 0.26 2.72 1. Qxa1+ Kxa1 2. Kf5 Kb1 3. Kf6 Kc2
4. f4 Kd3 5. Kxf7 Ke4
8-> 0.31 2.72 1. Qxa1+ Kxa1 2. Kf5 Kb1 3. Kf6 Kc2
4. f4 Kd3 5. Kxf7 Ke4
9 0.32 -1 1. Qxa1+
9 0.34 1.66 1. Qxa1+ Kxa1 2. Kf5 Kb1 3. f4 Kc2
4. Ke5 Kd3 5. f5 Ke3 6. f6
9-> 0.48 1.66 1. Qxa1+ Kxa1 2. Kf5 Kb1 3. f4 Kc2
4. Ke5 Kd3 5. f5 Ke3 6. f6
10 0.51 1.30 1. Qxa1+ Kxa1 2. Kg3 Kb2 3. Kf2 Kb3
4. f4 Kc3 5. Ke1 Kd3 6. f5 Ke3
10-> 2.90 1.30 1. Qxa1+ Kxa1 2. Kg3 Kb2 3. Kf2 Kb3
And it never changes its mind...
It is certainly possible that you have a version with an evaluation bug. I have
been going over the thing line by line, and have completely re-vamped its
"organization" before we start a complete "re-do" of many features, particularly
king safety. But it should always trade queens, so long as the endgame code is
being utilized correctly, as above....
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.