A Chess forum. ChessBanter

If this is your first visit, be sure to check out the FAQ by clicking the link above. You may have to register before you can post: click the register link above to proceed. To start viewing messages, select the forum that you want to visit from the selection below.

Go Back   Home » ChessBanter forum » Chess Newsgroups » rec.games.chess.analysis (Chess Analysis)
Site Map Home Register Authors List Search Today's Posts Mark Forums Read Web Partners

Wonderful game by help bot with Knight/ Rook sacrifices.



 
 
Thread Tools Display Modes
  #1  
Old August 25th 09, 08:53 AM posted to rec.games.chess.analysis,rec.games.chess.misc,alt.chess,rec.games.chess.computer
Martin Brown
external usenet poster
 
Posts: 1,015
Default Wonderful game by help bot with Knight/ Rook sacrifices.

Taylor Kingston wrote:
On Aug 21, 1:50 am, Sanny wrote:
In this game GetClub got an extra Knight. Then It was having an extra
Rook for one pawn.

I was thinking now GetClub will win but Help Bot managed to get a win
despite being a Rook down.

Was there any way for Beginner to win this game with its extra Rook?
What was the reasion GetClub lost even when it was a Rook up?


34...Rxb3 35.Qxb3 Rxg5+ 36.Kf1 Qd5 37.Re3 Qh1+ 38.Ke2 Rg1 39.Qb8+ Kh7
40.Qxc7?

A serious mistake that throws away the win. 40.Qb2! is the only good
move.


I think 40. Qb3! is marginally better. It covers both white rooks and
closes down blacks options faster. Black then has the choice of move the
king, make a few token checks or losing a queen for a rook.

But it had to be one of those two moves to win. There were a few other
drawn looking moves too.

If then 40...Re1+ 41.Kd2 Rd1+ (or 41...Rf1 42.Rc1 Rxf2+ 43.Kc3
Qh4 44.Qb1+) 42.Kc3 Rc1+ 43.Kb4 and the king is safe. The text allows
Black to draw by repetition.

40...Rb1 41.Rec3??




A game typical of GCs recent play: it got a won game, only to throw
it away as it did against SAT-W7. Its limited horizon was apparent.


It seems to do strange things when a position has a potential mate for
the opponent if the wrong move is played. Almost as if it plays into it
in the style of a helpmate!

Regards,
Martin Brown
  #2  
Old August 25th 09, 03:12 PM posted to rec.games.chess.analysis,rec.games.chess.misc,alt.chess,rec.games.chess.computer
Taylor Kingston[_2_]
external usenet poster
 
Posts: 3,256
Default Wonderful game by help bot with Knight/ Rook sacrifices.

On Aug 25, 3:53*am, Martin Brown
wrote:
Taylor Kingston wrote:

34...Rxb3 35.Qxb3 Rxg5+ 36.Kf1 Qd5 37.Re3 Qh1+ 38.Ke2 Rg1 39.Qb8+ Kh7
40.Qxc7?


* A serious mistake that throws away the win. 40.Qb2! is the only good
move.


I think 40. Qb3! is marginally better.


Could well be, Martin. An unfortunate quirk Fritz8 has is that it
sometimes will evaluate a potential move at 0.00 if it's part of a
line with the potential for threefold repetition, even if the
repetition can be avoided before the third time. In its analysis of
the position after 39...Qh7, that's what it did with 40.Qb3, and this
not being a game I wanted to spend a lot of time on, I just glanced at
the 0.00 and took it at its word when I wrote "40.Qb2! is the only
good move." Later I went back and actually played 40.Qb3, and lo! It
immediately came up with non-repeating lines and changed its
evaluation to winning for White.

Anyone know if this bug has been eliminated in later versions of
Fritz?

  #3  
Old August 25th 09, 08:12 PM posted to rec.games.chess.analysis,rec.games.chess.misc,alt.chess,rec.games.chess.computer
help bot
external usenet poster
 
Posts: 9,302
Default Wonderful game by help bot with Knight/ Rook sacrifices.

On Aug 25, 10:12 am, Taylor Kingston
wrote:

A serious mistake that throws away the win. 40.Qb2! is the only good
move.


I think 40. Qb3! is marginally better.


Could well be, Martin. An unfortunate quirk Fritz8 has is that it
sometimes will evaluate a potential move at 0.00 if it's part of a
line with the potential for threefold repetition, even if the
repetition can be avoided before the third time. In its analysis of
the position after 39...Qh7, that's what it did with 40.Qb3, and this
not being a game I wanted to spend a lot of time on, I just glanced at
the 0.00 and took it at its word when I wrote "40.Qb2! is the only
good move." Later I went back and actually played 40.Qb3, and lo! It
immediately came up with non-repeating lines and changed its
evaluation to winning for White.

Anyone know if this bug has been eliminated in later versions of
Fritz?



I don't have Fritz8, but with Rybka running in the
free Arena GUI, it quite often will misevaluate any
position where there is a possibility of repeating a
position it has already evaluated. Of course, if
one analyzes an entire game, that will include
/every position/ except the final one.

Rybka (I may be wrong in that this could instead
relate to Arena, or the way they communicate
with one another) will very often insist that a given
position's score is 0.0, when there are very simple
ways to avoid threefold repetitions. I, too, would
call this a bug, because users do not want mis-
evaluations displayed.

One also notes the fact of TK's denial-- his
refusal to accept responsibility for even a small
analytical error in a very complicated ending,
preferring instead to transfer the blame to Fritz
for what he himself wrote here in rgc (i.e. "Qb2
is the only good move").

After the fiasco in this game, I intend to try and
keep things a bit simpler-- even if that means
just trading pieces to win in the King-and-pawn
endings, where GetClub is just plain awful. But
you never know-- Sanny is always introducing
new "improvements", which afford one many
opportunities to win by simply prolonging the
game to give the GetClub engine more oppor-
tunities to self-destruct.

I can still recall a game in which DeepThought
had deliberately sacrificed a pawn against Gary
Kasparov, then "changed its mind" next turn
and went on to lose like a carrot. They blamed
the horizon effect, yet when humans do this
sort of thing, it is blamed on them being weak
players, for weak players are those who give
things away without good reason. That too,
was a self-destruct, but at a more respected,
grandmaster level.

As I have pointed out to Sanny time and time
again, the vast majority of its games are losses
for the GetClub engine, so it only makes sense
to work on improving its ability to defer, if not
prevent itself from, being checkmated. This
"tough defense" may not work against strong
players, but it may very well work against the
majority of human opponents-- those who are
not operating commercial chess engines,
anyway.


-- help bot


  #4  
Old August 26th 09, 12:02 AM posted to rec.games.chess.analysis,rec.games.chess.computer
Simon Krahnke
external usenet poster
 
Posts: 625
Default Wonderful game by help bot with Knight/ Rook sacrifices.

* Taylor Kingston (16:12) schrieb:

Could well be, Martin. An unfortunate quirk Fritz8 has is that it
sometimes will evaluate a potential move at 0.00 if it's part of a
line with the potential for threefold repetition, even if the
repetition can be avoided before the third time.


That's a complicated issue, of which there are several different
possibilities:

All three positions on the board: trivial.

Two positions on the board, one inside search: Simple, draw is the
result of best play.

Three positions inside search: Again, result of best play. You don't
really need to search for the final repetition, if first sequence is
part of best play, the second will be, too.

One position on the board, one inside search: The hard case, the part of
the sequence of moves leading to the repetition that's on the board
might not be best play.

An engine optimized for play, may to choose to ignore the last
complication and always look for the first repetition of the current
position and return draw.

mfg, simon .... a programmer's view
  #5  
Old August 26th 09, 09:11 AM posted to rec.games.chess.analysis,rec.games.chess.misc,alt.chess,rec.games.chess.computer
Martin Brown
external usenet poster
 
Posts: 1,015
Default Wonderful game by help bot with Knight/ Rook sacrifices.

Taylor Kingston wrote:
On Aug 25, 3:53 am, Martin Brown
wrote:
Taylor Kingston wrote:

34...Rxb3 35.Qxb3 Rxg5+ 36.Kf1 Qd5 37.Re3 Qh1+ 38.Ke2 Rg1 39.Qb8+ Kh7
40.Qxc7?
A serious mistake that throws away the win. 40.Qb2! is the only good
move.

I think 40. Qb3! is marginally better.


Could well be, Martin. An unfortunate quirk Fritz8 has is that it
sometimes will evaluate a potential move at 0.00 if it's part of a
line with the potential for threefold repetition, even if the
repetition can be avoided before the third time.


I vaguely remember that. It sometimes allows a way to beat Fritz8. I
checked and my copy running under the latest GUI still gives a phantom
0.00 draw evaluation for Qb3. So it is an engine rather than GUI quirk.

I'd recommend Shredder10 - it is a lot more fun to spar against and its
analysis is more incisive and less staid than Fritz. Sometimes it even
agrees with published theory after about 24 hours runtime.

Unique amongst the engines I have to hand it sees Qb3 as OK here.

Anyone care to try Rybka3 on it? Or later versions of any other engine?

Even though the game is rubbish this position is a good test of engines.

In its analysis of
the position after 39...Qh7, that's what it did with 40.Qb3, and this
not being a game I wanted to spend a lot of time on, I just glanced at
the 0.00 and took it at its word when I wrote "40.Qb2! is the only
good move." Later I went back and actually played 40.Qb3, and lo! It
immediately came up with non-repeating lines and changed its
evaluation to winning for White.

Anyone know if this bug has been eliminated in later versions of
Fritz?


Dunno. Fritz8 is the highest version that I have to hand.
I tried running a few other engines on it as a test.

Turing falls for the material grab Qxc7 at ply 7 in 7mins
(and would take hours for the next ply)

Comet B68 finds Qb2 more by good luck than good judgement.
Qxc7 is its second choice and it thinks Qb3 is drawn and ranked 27/38!!
(there are a lot of obviously losing moves above it)
It looks to be an extremely materialistic and flat evaluation.

Fritz 5.32 gets Qb2 but takes a ply 11 search to see Qb4 is nbg.
Qb3 evaluates as drawn.

Junior 7 same as Fritz 5.32
Crafty 19.19 is the same too
Crafty 20.14 same too (but faster on multicpu)
HIARCS 10 Qb4 seen drawn at ply 10, Qb3 phantom draw.
Rybka2.32 Qb4 seen drawn at ply 8, Qb3 phantom draw.

So Shredder10 is unique so far in correctly putting Qb3 at the top of
the pile. OK we can haggle about whether Qb2 or Qb3 is the best move.
Are there any other engines that can find Qb3?

The other thing I have noticed with all the engines is that when
backward annotating a game or blunderchecking they very very rarely put
a crazy evaluation onto a position that makes no sense at all. If you
analyse the position directly or even redo the analysis immediately with
the same settings a different result will be found.

Regards,
Martin Brown
  #6  
Old August 26th 09, 03:08 PM posted to rec.games.chess.analysis,rec.games.chess.misc,alt.chess,rec.games.chess.computer
Taylor Kingston[_2_]
external usenet poster
 
Posts: 3,256
Default Wonderful game by help bot with Knight/ Rook sacrifices.

On Aug 26, 4:11*am, Martin Brown
wrote:
Taylor Kingston wrote:
On Aug 25, 3:53 am, Martin Brown
wrote:
Taylor Kingston wrote:


34...Rxb3 35.Qxb3 Rxg5+ 36.Kf1 Qd5 37.Re3 Qh1+ 38.Ke2 Rg1 39.Qb8+ Kh7
40.Qxc7?
* A serious mistake that throws away the win. 40.Qb2! is the only good
move.
I think 40. Qb3! is marginally better.


* Could well be, Martin. An unfortunate quirk Fritz8 has is that it
sometimes will evaluate a potential move at 0.00 if it's part of a
line with the potential for threefold repetition, even if the
repetition can be avoided before the third time.


I vaguely remember that. It sometimes allows a way to beat Fritz8. I
checked and my copy running under the latest GUI still gives a phantom
0.00 draw evaluation for Qb3. So it is an engine rather than GUI quirk.

I'd recommend Shredder10 - it is a lot more fun to spar against and its
analysis is more incisive and less staid than Fritz. Sometimes it even
agrees with published theory after about 24 hours runtime.

Unique amongst the engines I have to hand it sees Qb3 as OK here.

Anyone care to try Rybka3 on it? Or later versions of any other engine?

Even though the game is rubbish this position is a good test of engines.

In its analysis of
the position after 39...Qh7, that's what it did with 40.Qb3, and this
not being a game I wanted to spend a lot of time on, I just glanced at
the 0.00 and took it at its word when I wrote "40.Qb2! is the only
good move." Later I went back and actually played 40.Qb3, and lo! It
immediately came up with non-repeating lines and changed its
evaluation to winning for White.


* Anyone know if this bug has been eliminated in later versions of
Fritz?


Dunno. Fritz8 is the highest version that I have to hand.
I tried running a few other engines on it as a test.

Turing falls for the material grab Qxc7 at ply 7 in 7mins
(and would take hours for the next ply)

Comet B68 finds Qb2 more by good luck than good judgement.
Qxc7 is its second choice and it thinks Qb3 is drawn and ranked 27/38!!
(there are a lot of obviously losing moves above it)
It looks to be an extremely materialistic and flat evaluation.

Fritz 5.32 gets Qb2 but takes a ply 11 search to see Qb4 is nbg.
Qb3 evaluates as drawn.

Junior 7 same as Fritz 5.32
Crafty 19.19 is the same too
Crafty 20.14 same too (but faster on multicpu)
HIARCS 10 Qb4 seen drawn at ply 10, Qb3 phantom draw.
Rybka2.32 Qb4 seen drawn at ply 8, Qb3 phantom draw.

So Shredder10 is unique so far in correctly putting Qb3 at the top of
the pile. OK we can haggle about whether Qb2 or Qb3 is the best move.
Are there any other engines that can find Qb3?

The other thing I have noticed with all the engines is that when
backward annotating a game or blunderchecking they very very rarely put
a crazy evaluation onto a position that makes no sense at all. If you
analyse the position directly or even redo the analysis immediately with
the same settings a different result will be found.

Regards,
Martin Brown


Hmmmm, sounds like this phantom-draw bug is pretty common among
chess engines. Strange.
  #7  
Old August 26th 09, 09:50 PM posted to rec.games.chess.analysis,rec.games.chess.misc,alt.chess,rec.games.chess.computer
Thomas T. Veldhouse[_2_]
external usenet poster
 
Posts: 165
Default Rybka vs GetClub

In rec.games.chess.computer Sanny wrote:
Now Beginner is very strong. Today I had a game with Rybka The game
was equal till 17th move. But then the rybka-server got disconnected.


Rybka probably claims a win on time and stops playing. You really should know
better by now than to state such a stupid comment like you did above, where
you claim that you are sure if the Rybka server hadn't disconnected it would
have been a drawn game; that is simply ludicrous and you know it.

--
Thomas T. Veldhouse

Religion is a crutch, but that's okay... humanity is a cripple.
  #8  
Old August 27th 09, 12:32 AM posted to rec.games.chess.analysis,rec.games.chess.misc,alt.chess,rec.games.chess.computer
none
external usenet poster
 
Posts: 3,073
Default Rybka vs GetClub

On Aug 26, 4:50*pm, "Thomas T. Veldhouse" wrote:
In rec.games.chess.computer Sanny wrote:

Now Beginner is very strong. Today I had a game with Rybka The game
was equal till 17th move. But then the rybka-server got disconnected.


Rybka probably claims a win on time and stops playing. *You really should know
better by now than to state such a stupid comment like you did above, where
you claim that you are sure if the Rybka server hadn't disconnected it would
have been a drawn game; that is simply ludicrous and you know it.

--
Thomas T. Veldhouse

* Religion is a crutch, but that's okay... humanity is a cripple.


The opiate of the masses.
  #9  
Old August 27th 09, 09:16 AM posted to rec.games.chess.analysis,rec.games.chess.misc,alt.chess,rec.games.chess.computer
Martin Brown
external usenet poster
 
Posts: 1,015
Default Wonderful game by help bot with Knight/ Rook sacrifices.

Taylor Kingston wrote:
On Aug 26, 4:11 am, Martin Brown
wrote:
Taylor Kingston wrote:


Anyone know if this bug has been eliminated in later versions of
Fritz?

Dunno. Fritz8 is the highest version that I have to hand.
I tried running a few other engines on it as a test.

Turing falls for the material grab Qxc7 at ply 7 in 7mins
(and would take hours for the next ply)

Comet B68 finds Qb2 more by good luck than good judgement.
Qxc7 is its second choice and it thinks Qb3 is drawn and ranked 27/38!!
(there are a lot of obviously losing moves above it)
It looks to be an extremely materialistic and flat evaluation.

Fritz 5.32 gets Qb2 but takes a ply 11 search to see Qb4 is nbg.
Qb3 evaluates as drawn.

Junior 7 same as Fritz 5.32
Crafty 19.19 is the same too
Crafty 20.14 same too (but faster on multicpu)
HIARCS 10 Qb4 seen drawn at ply 10, Qb3 phantom draw.
Rybka2.32 Qb4 seen drawn at ply 8, Qb3 phantom draw.

So Shredder10 is unique so far in correctly putting Qb3 at the top of
the pile. OK we can haggle about whether Qb2 or Qb3 is the best move.
Are there any other engines that can find Qb3?

The other thing I have noticed with all the engines is that when
backward annotating a game or blunderchecking they very very rarely put
a crazy evaluation onto a position that makes no sense at all. If you
analyse the position directly or even redo the analysis immediately with
the same settings a different result will be found.


Hmmmm, sounds like this phantom-draw bug is pretty common among
chess engines. Strange.


Seems to be endemic. So far only Shredder10 has managed it correctly.

1Q6/2p2ppk/2p1p3/8/P1RP4/4R3/4KP2/6rq w - - 0 40

Anyone care to post results for other well known engines?

It would be interesting to find a set of test positions that can be used
to classify engines by their mistakes.

Regards,
Martin Brown
 




Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

vB code is On
Smilies are On
[IMG] code is Off
HTML code is Off
Forum Jump

Similar Threads
Thread Thread Starter Forum Replies Last Post
My conventional end game with Ivan, game on hold now , my move too SAT W-7 rec.games.chess.computer (Computer Chess) 2 November 28th 08 04:20 AM
Pinning the knight to the king/queen in the opening John Salerno rec.games.chess.misc (Chess General) 4 October 2nd 08 09:01 PM
Counting knight moves foot rec.games.chess.misc (Chess General) 42 October 13th 07 01:12 PM
Interview with CJA Award Winning Historian in The Chess Journalist The Historian rec.games.chess.politics (Chess Politics) 215 November 16th 06 08:34 PM


All times are GMT +1. The time now is 05:13 AM.


Powered by vBulletin® Version 3.6.4
Copyright ©2000 - 2017, Jelsoft Enterprises Ltd.Content Relevant URLs by vBSEO 2.4.0
Copyright 2004-2017 ChessBanter.
The comments are property of their posters.