Author: Robert Hyatt
Date: 03:46:40 01/11/00
Go up one level in this thread
On January 11, 2000 at 03:45:04, Ed Schröder wrote:
>>Posted by Robert Hyatt on January 10, 2000 at 23:38:17:
>>
>>>>Put black's f7 pawn on g7 and black is suddenly doing lots better.
>>>
>>>And can save a few bucks as in my case. I think it is doable to write
>>>code for this kind of situations only that in the 18 years I am working
>>>now on Rebel I haven't seen such a case of a trapped bishop happening.
>>>These guys surely know how to debug a chess engine.
>>>
>>>Ed
>>>
>>>
>>>>bruce
>>
>>
>>There are three issues:
>>
>>(1) sometimes you simply have to take your lumps and go home. This might be
>>considered to be one of those kinds of positions where they are so rare, they
>>aren't worth fixing.
>
>Problem: this one can't be solved be search.
>
>>(2) you can also avoid this in a different way, namely the way I do this in
>>my eval, by (a) penalizing blocked pawns and (b) rewarding potential pawn
>>levers to make sure the position doesn't become so hopelessly blocked as the
>>one you saw. CptnBlueBear will murder you with this if you aren't careful, as
>>it is _exactly_ what he tries to do. And as a GM, it is very difficult to
>>keep him from doing it without accepting a weakness or two that you really
>>would prefer not to.
>
>Understood.
>
>Problem-2: Like Crafty Rebel has knowledge on this case only that its current
>value (penalty) is in the 0.10 - 0.15 range. For this particular case it should
>be at least 1.00 - 2.00 right?
>
>>(3) you can fix this in code. I did it once for Roman. But it was _very_ slow,
>>for the good that it provided, and he later agreed that trying to avoid the
>>blocked position just because it was blocked, rather than worrying about a
>>hopelessly bad bishop, probably was more effective in most cases.
>
>Disadvantage: too often moving into blocked pawns is the only right move
>and you need to play the right moves all the time against these guys. One
>bad move and you are lost, this game is the ultimate example I would say.
>After 21..e5?? the game was over.
Maybe, maybe not. IE perhaps the best time to block things is when you are
afraid of losing. IE in some positions where you are in danger of getting
pawn-rolled on the king-side. blocking the queen-side means you can defend
on the king-side without worrying about stuff over on the queen-side. But
maybe it is also reasonable to simply keep the game open. The danger is that
sometimes, to keep it open, you end up in a lost endgame due to a pawn weakness
that is too much to overcome.
I view it like this:
If you allow blocked positions, you might see a match score like +5, -10, =15
(from the computer's perspective.) If you try hard to not allow them, you might
see +12, -12, =6 (or some such result). IE you win more than you lose, but you
lose more than you used to draw.
>
>>The thing that worries me is that there are _hundreds_ more of these kinds of
>>special cases that are left to discover. I have discovered many in my code. I
>>am afraid I have only scratched the surface, yet I have over 50,000 blitz games
>>vs GM players to go on. :(
>
>Hundreds of these kind of exceptions huh?
>
>I tend to agree, here are two of them:
>
>[d]r4rk1/p1q2ppp/1pn2n2/2p1p3/3PP1b1/P1P2N2/B4PPP/R1BQR1K1 w - - 0 14
>
>[White "Kasparov, Gary"]
>[Black "Kramnik, Vladimir"]
>
>14. dxc5! { Natural 14.d5 will be serious positional mistake- bishop on a2
>could be closed for a long time.}
>
>I bet most programs (Rebel included play 14.d5). 14.d5 looks okay and for
>many good reasons still it's called a serious positional mistake. That hurts!
>
Common one at times. Getting a protected passer, but blocking things up. I
value the passed pawn more than the resulting blocked position...
>
>Case 2:
>
>[d]2rq2k1/4bppp/p1rp4/1p1NpP2/4P3/2PQ4/PP4PP/3R1R1K w - -
>
>Robert Fischer's famous: Ra1!!
>
>I assume almost every program will play f6? here leaving white with
>nothing. I wonder if any program will find this fine positional move.
>
>Ed
Crafty never considers f6 here. Here is my output. :)
8 1.84 0.39 1. Qg3 f6 2. Qg4 Qd7 3. Kg1 a5 4. Rfe1
Rc4
8 3.64 0.41 1. Kg1 Bg5 2. Rf3 Qd7 3. Nb4 Rb6 4.
Nd5 Rbc6
8-> 5.57 0.41 1. Kg1 Bg5 2. Rf3 Qd7 3. Nb4 Rb6 4.
Nd5 Rbc6
9 6.73 0.41 1. Kg1 Bg5 2. Rf3 Qd7 3. Rg3 Bh4 4.
Rh3 Qa7+ 5. Re3 Qc5
9 14.37 0.42 1. Qe2 f6 2. Kg1 Qd7 3. Rfe1 Bd8 4.
Nb4 Qa7+ 5. Kf1 Rb6
9-> 14.68 0.42 1. Qe2 f6 2. Kg1 Qd7 3. Rfe1 Bd8 4.
Nb4 Qa7+ 5. Kf1 Rb6
10 18.12 0.39 1. Qe2 Bg5 2. Kg1 f6 3. Qg4 Rc4 4.
Rd3 h5 5. Qe2 Qd7
10 34.31 0.44 1. Ra1 Rc4 2. Kg1 a5 3. b3 R4c5 4.
a4 bxa4 5. Rxa4 f6 6. Raa1
10-> 42.36 0.44 1. Ra1 Rc4 2. Kg1 a5 3. b3 R4c5 4.
a4 bxa4 5. Rxa4 f6 6. Raa1
11 1:55 0.52 1. Ra1 Rb8 2. Nb4 Rcb6 3. a4 a5 4.
Nd5 R6b7 5. b4 axb4 6. axb5 Rxb5 7.
cxb4
11-> 2:02 0.52 1. Ra1 Rb8 2. Nb4 Rcb6 3. a4 a5 4.
Nd5 R6b7 5. b4 axb4 6. axb5 Rxb5 7.
cxb4
This from my notebook. It sticks with Ra1 thru 12 plies anyway, where I
stopped it. I will have to study it a bit to see why it actually likes this.
Actually I won't. Seems to create a passed pawn again. it is a sort of "hidden
majority" if white reacts quickly.
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.