Hej, a sanny thread
Simon Krahnke
no living Sanny thread? I felt the need to do a game on his ugly site,
anyway.
Here it is, Rybka 2.2 32bit with 1 ply search depth against Getclub easy
level:
[Date "2010.01.01"]
[White "Rybka 2.2 32 bit"]
[Black "Getclub easy"]
1. Nc3 {(Nc3) +0.12/1 0} d5
2. Nf3 {(Nf3) +0.18/1 0} Nc6 { out of book here, really, on second move }
3. d3 {(d3) +0.08/1 0} Nf6
4. Bf4 {(Bf4) +0.13/1 0} Bf5
5. Nb5 {(Nb5) +0.12/1 0} e5 { why would black attempt this exchange? }
6. Nxe5 {(Nxe5) +0.31/1 0} Nxe5
7. Bxe5 {(Bxe5) +0.35/1} Bd6
8. Bxd6 {(Bxd6) +0.43/1 0} cxd6 { isolated doubled pawn! }
9. Nd4 {(Nd4) +0.19/1 0} Bg4 { why move once, when you can move twice? }
10. h3 {(h3) +0.40/1 0} Bd7 { what was the job of that bishop on the kingside? }
11. Qd2 {(Qd2) +0.17/1 0} O-O
12. Qb4 {(Qb4) +0.21/1 0} Qc7
13. g4 {(g4) +0.22/1 0} Rac8 { oh, this power! }
14. Bg2 {(Bg2) +0.34/1 0} Rfe8
15. O-O {(0-0) +0.40/1 0} Re5 { yippie! }
16. f4 {(f4) +0.39/1 0} Re3
17. g5 {(g5) +0.39/1 0} a5
18. Qd2 {(Qd2) +1.83/1 0} Rg3 { "but i want to lose this rook!" }
19. gxf6 {(gxf6) +1.77/1 0} Bxh3 { this put-your-pieces-near-the-opponent's-king-strategy again? }
20. Kh2 {(Kh2) +2.14/1 0} Bxg2
21. Kxg3 {(Kxg3) +2.24/1 0} Bxf1
22. Rxf1 {(Rxf1) +2.42/1 0} gxf6 { two isolated doubled pawns }
23. Rg1 {(Rg1) +2.74/1 0} Qc5 { the action is on the other side, girl! }
24. Qe3 {(Qe3) +2.79/1 0} Rb8 { let's protect this unattacked backward pawn }
25. Kh4+ {(Kh4+) +3.12/1 0} Kh8 { king's best place: h8! }
26. Qg3 {(Qg3) +319.00/1 0} h6 { this is the place, where Getclub normally gives up }
27. Qg7# {(Qg7+) +319.00/1 0} { THE END }
1-0
Just let Rybka have white, and it's a quick win in 27 moves. Quick for
Rybka, it took less than a second, while Getclub took around 40 minutes.
And Rybka actually won, Getclub actually lost. It didn't give up at the
last moment. It's mate!
Getclub just walked into it, and maybe Sanny fixed this issue. But then
again, when Getclub is stupid enough to walk into a mate, why would it
find it on its last move?
I won't post a URL, someone might complain, and the list of moves is
fucked up, again.
mfg, simon .... 4 at night, all quiet!
Sanny
> again, when Getclub is stupid enough to walk into a mate, why would it
> find it on its last move?
The work is in progress. Lets hope my computer gets repaired this
week.
> I won't post a URL, someone might complain, and the list of moves is
> fucked up, again.
This problem will be sorted out as soon as my computer is repaired.
I have a few ideas for GetClub. That may make GetClub 3-4 times
stronger than now.
But It will take 1-2 weeks to bring the changes.
Thanks, for playing game. I will see that the problem of false moves
recording is sorted out as soon as possible.
Bye
Sanny
--
Play Chess at: http://www.GetClub.com/Chess.html
Why Play at GetClub.com Chess.
1. Its free to play. And view Animated Games. [NO DOWNLOADS]
2. Your games are recorded.along with time taken by you.
3. Your games are analyzed by higher Levels
4. You win Prizes every month. Also get a RATING
5. Earn $1.00 for playing 30 games every month. [Win]
6. Only those games in which you win are counted for $1.00 every
month.
Start a game: http://www.GetClub.com/Chess.html
Simon Krahnke
>> Getclub just walked into it, and maybe Sanny fixed this issue. But then
>> again, when Getclub is stupid enough to walk into a mate, why would it
>> find it on its last move?
>
> The work is in progress. Lets hope my computer gets repaired this
> week.
Your computer is broken? Sorry to read that.
> I have a few ideas for GetClub. That may make GetClub 3-4 times
> stronger than now.
Maybe then, Vas' work on the search algorithm will be worth something
against Getclub.
mfg, simon .... l
Martin Brown
It is surprising here that GetClub could not find
17. ... Nh5 which moves the threatened piece and puts it back on the
path to equality. 18. Bxd5 is punished by a decent kingside attack.
What black cannot afford to do is allow Qd2 and then move the rook into
immediate peril, but that is how the game proceeds. It is all over bar
the shouting at this stage.
> 18. Qd2 {(Qd2) +1.83/1 0} Rg3 { "but i want to lose this rook!" }
> 19. gxf6 {(gxf6) +1.77/1 0} Bxh3 { this put-your-pieces-near-the-opponent's-king-strategy again? }
> 20. Kh2 {(Kh2) +2.14/1 0} Bxg2
> 21. Kxg3 {(Kxg3) +2.24/1 0} Bxf1
> 22. Rxf1 {(Rxf1) +2.42/1 0} gxf6 { two isolated doubled pawns }
> 23. Rg1 {(Rg1) +2.74/1 0} Qc5 { the action is on the other side, girl! }
> 24. Qe3 {(Qe3) +2.79/1 0} Rb8 { let's protect this unattacked backward pawn }
> 25. Kh4+ {(Kh4+) +3.12/1 0} Kh8 { king's best place: h8! }
This is quite incredible. It chose the forced mate line against the only
alternative Kf8 which holds out much longer. It is more like a helpmate.
> 26. Qg3 {(Qg3) +319.00/1 0} h6 { this is the place, where Getclub normally gives up }
Perhaps a "bug" in the code prevents it spuriously giving up if every
possible outcome is mate #1.
> 27. Qg7# {(Qg7+) +319.00/1 0} { THE END }
> 1-0
>
> Just let Rybka have white, and it's a quick win in 27 moves. Quick for
> Rybka, it took less than a second, while Getclub took around 40 minutes.
Is that less than a second in aggregate timing?
>
> And Rybka actually won, Getclub actually lost. It didn't give up at the
> last moment. It's mate!
Seems like it didn't see the mate coming which is hard to understand.
Happy New Year!
Regards,
Martin Brown
Simon Krahnke
> It is surprising here that GetClub could not find
> 17. ... Nh5 which moves the threatened piece and puts it back on the
> path to equality. 18. Bxd5 is punished by a decent kingside attack.
>
> What black cannot afford to do is allow Qd2 and then move the rook into
> immediate peril, but that is how the game proceeds. It is all over bar
> the shouting at this stage.
Would you try a game against Rybka at 1 ply per move? I would like to
know, how strong that really is.
>> 18. Qd2 {(Qd2) +1.83/1 0} Rg3 { "but i want to lose this rook!" }
>> 19. gxf6 {(gxf6) +1.77/1 0} Bxh3 { this put-your-pieces-near-the-opponent's-king-strategy again? }
>> 20. Kh2 {(Kh2) +2.14/1 0} Bxg2
>> 21. Kxg3 {(Kxg3) +2.24/1 0} Bxf1
>> 22. Rxf1 {(Rxf1) +2.42/1 0} gxf6 { two isolated doubled pawns }
>> 23. Rg1 {(Rg1) +2.74/1 0} Qc5 { the action is on the other side, girl! }
>> 24. Qe3 {(Qe3) +2.79/1 0} Rb8 { let's protect this unattacked backward pawn }
>> 25. Kh4+ {(Kh4+) +3.12/1 0} Kh8 { king's best place: h8! }
>
> This is quite incredible. It chose the forced mate line against the only
> alternative Kf8 which holds out much longer. It is more like a helpmate.
Yeah, Sanny, if you ask for a wrong move, here it is.
>> 26. Qg3 {(Qg3) +319.00/1 0} h6 { this is the place, where Getclub normally gives up }
>
> Perhaps a "bug" in the code prevents it spuriously giving up if every
> possible outcome is mate #1.
Maybe some day I will try to look at the code again.
>> 27. Qg7# {(Qg7+) +319.00/1 0} { THE END }
>> 1-0
>>
>> Just let Rybka have white, and it's a quick win in 27 moves. Quick for
>> Rybka, it took less than a second, while Getclub took around 40 minutes.
>
> Is that less than a second in aggregate timing?
I think so. It's what Arena shows. All the moves were shown as done in
0s, and it adds them up as 0s. The responses were all below 100ms, that
it, instantly.
>> And Rybka actually won, Getclub actually lost. It didn't give up at the
>> last moment. It's mate!
>
> Seems like it didn't see the mate coming which is hard to understand.
Selective Search. But even that makes one wonder, what it though about
on the last two moves.
Maybe at 25. ... Kh8 it thought about Kf8 for a while, found it to be
inferior and then just chose the other move.
mfg, simon .... l
JVMe...@aol.com
>
> > Seems like it didn't see the mate coming which is hard to understand.
>
> Selective Search. But even that makes one wonder, what it though about
> on the last two moves.
>
> Maybe at 25. ... Kh8 it thought about Kf8 for a while, found it to be
> inferior and then just chose the other move.
Inferior to a mate in 2??? There is no reason for this at all, except
for a horrific bug.
Maybe some aggressive pruning could cause the engine to hold off
seeing the mate in 2 until depth 7 or 8, but that still shouldn't take
more than one or two seconds to see. You don't show the times for each
of GetCrap's moves, but since you said that it took 40 minutes, and
made only 27 moves, I think it's pretty safe to assume that at least
15 seconds was taken on this terrible move. Looks like GC hasn't
improved, in terms of total playing strength (i.e. the ability to
score points), in all these years if it can't even avoid a mate in 2.
jm
Simon Krahnke
> On Jan 1, 8:53�pm, Simon Krahnke <overl...@gmx.li> wrote:
>
>>> Seems like it didn't see the mate coming which is hard to understand.
>>
>> Selective Search. But even that makes one wonder, what it though about
>> on the last two moves.
>>
>> Maybe at 25. ... Kh8 it thought about Kf8 for a while, found it to be
>> inferior and then just chose the other move.
>
> Inferior to a mate in 2??? There is no reason for this at all, except
> for a horrific bug.
My suggestion is, that it didn't look at Kh8 enough to see it was a mate
in 2 and chose it because Kf8 is negative.
> Maybe some aggressive pruning could cause the engine to hold off
> seeing the mate in 2 until depth 7 or 8, but that still shouldn't take
> more than one or two seconds to see.
Selective Search leaves some moves totally unevaluated, you can easily
miss mates that way.
> You don't show the times for each of GetCrap's moves,
I don't have them, Arenas times include my operator time, and the list
of moves on site is fucked up.
> but since you said that it took 40 minutes,
That's my estimate based on Arena's total time for black and a guess on
my operator time.
> and made only 27 moves, I think it's pretty safe to assume that at
> least 15 seconds was taken on this terrible move.
Getclubs list mentions "32. Kg8-h8 (76)", I guess that means it spent 76
seconds on this move.
> Looks like GC hasn't improved, in terms of total playing strength
> (i.e. the ability to score points), in all these years if it can't
> even avoid a mate in 2.
That obvious from its unability to win any games.
mfg, simon .... l
JVMe...@aol.com
> * <JVMerl...@aol.com> (07:22) schrieb:
>
> > On Jan 1, 8:53 pm, Simon Krahnke <overl...@gmx.li> wrote:
>
> >>> Seems like it didn't see the mate coming which is hard to understand.
>
> >> Selective Search. But even that makes one wonder, what it though about
> >> on the last two moves.
>
> >> Maybe at 25. ... Kh8 it thought about Kf8 for a while, found it to be
> >> inferior and then just chose the other move.
>
> > Inferior to a mate in 2??? There is no reason for this at all, except
> > for a horrific bug.
>
> My suggestion is, that it didn't look at Kh8 enough to see it was a mate
> in 2 and chose it because Kf8 is negative.
>
> > Maybe some aggressive pruning could cause the engine to hold off
> > seeing the mate in 2 until depth 7 or 8, but that still shouldn't take
> > more than one or two seconds to see.
>
> Selective Search leaves some moves totally unevaluated, you can easily
> miss mates that way.
True, you can miss LONGER mates. But for a mate in 2 that move would
only remain unevaluated for early search depths, as I mentioned. With
only two legal moves, this should take no time at all.
My engine is rated around 1950, and it avoids Kh8 in a quarter of a
second at depth 8 (as I said, it has some aggressive pruning that puts
off seeing the mate for a few search depths). Let it think for 76
seconds (on my P4-2.4GHz) and it's already evaluating depth 14.
I doubt there is a single decent engine (and by decent, I mean rated
at least 1600-1700) that would play Kh8 even if it had only a few
seconds to think. Of course, GetClub is not decent, so it played it.
jm
Sanny
> at least 1600-1700) that would play Kh8 even if it had only a few
> seconds to think. Of course, GetClub is not decent, so it played it.
This problem was sorted out today.
Now GetClub will see such mates.
Just play a new game and see how good GetClub plays now.
Bye
Sanny
JVMe...@aol.com
> > I doubt there is a single decent engine (and by decent, I mean rated
> > at least 1600-1700) that would play Kh8 even if it had only a few
> > seconds to think. Of course, GetClub is not decent, so it played it.
>
> This problem was sorted out today.
>
> Now GetClub will see such mates.
>
> Just play a new game and see how good GetClub plays now.
I doubt it. I'll let other people who are much more masochistic than I
am play on your site.
jm
Martin Brown
> * <newspam|||@nezumi.demon.co.uk> (13:04) schrieb:
>
>> It is surprising here that GetClub could not find
>> 17. ... Nh5 which moves the threatened piece and puts it back on the
>> path to equality. 18. Bxd5 is punished by a decent kingside attack.
>>
>> What black cannot afford to do is allow Qd2 and then move the rook into
>> immediate peril, but that is how the game proceeds. It is all over bar
>> the shouting at this stage.
>
> Would you try a game against Rybka at 1 ply per move? I would like to
> know, how strong that really is.
I had a try using Rybka 2.2 in Arena at fixed 1 ply, 2ply and 3 ply. I
think the nominally fixed depth search is very nominal...
I could beat it reliably at 1 ply by setting traps, and hold my own at 2
ply but I usually did something stupid and lost to it at 3 ply. That was
in a few quick games. I didn't get on with Arena very well though. It
kept spontaneously minimising on me, and sometimes made a move for me
that I wasn't intending.
I'd hazard a guess that Rybka 2.2 plays around 1600-1800 ELO at 1 ply
with occasional tactical brilliances when singular extensions get
triggered in forced lines.
The fastest I managed to beat it was in 24 moves. Coincidentally about
the same time that Sannys program lasts against Rybka on a 1 ply search.
Here are a couple of representative games with Rybka 2.2 at fixed ply 1,
and ply 2 respectively.
You can see from the score that Rybka 2.2 didn't see the attack on its
king until far too late at ply 1. Its undoing was 18. ... Nxf3+ and it
still thinks it is winning after playing 19. ... Kc8 (one move too late
to work) when it actually mate #6.
I didn't like its choice of opening very much.
[Event "Test Fixed ply 1"]
[Date "2010.01.05"]
[White "Martin"]
[Black "Rybka 2.2 32 bit"]
[Result "1-0"]
[Opening "Scandinavian"]
[Time "18:01:15"]
[Variation "Portuguese, 4.Nf3 Qxd5 5.Be2 Nc6"]
[WhiteElo "2100"]
[Termination "normal"]
[PlyCount "47"]
[WhiteType "human"]
[BlackType "program"]
1. e4 Nf6 {(Ng8f6) +0.02/1 0} 2. Nc3 Nc6 {(Nb8c6) +0.08/1 0} 3. Nf3 d5
{(d7d5) +0.08/1} 4. exd5 Nxd5 {(Nf6xd5) +0.08/1 0} 5. Nxd5 Qxd5 {(Qd8xd5)
+0.24/1 0} 6. d4 Bg4 {(Bc8g4) +0.28/1 0} 7. c3 O-O-O {(OOO) +0.36/1 0} 8.
Be2 e6 {(e7e6) +0.25/1 0} 9. O-O Bd6 {(Bf8d6) +0.14/1 0} 10. Re1 Kb8
{(Kc8b8) +0.17/1 0} 11. a4 a6 {(a7a6) +0.21/1 0} 12. b4 h5 {(h7h5) +0.16/1
0} 13. c4 Qf5 {(Qd5f5) +0.24/1 0} 14. b5 Nb4 {(Nc6b4) -0.05/1 0} 15. Qb3
axb5 {(a6xb5) +0.67/1 0} 16. axb5 Nc2 {(Nb4c2) +0.65/1 0} 17. Be3 Nxe1
{(Nc2xe1) +0.52/1 0} 18. Qa4 Nxf3+ {(Ne1xf3+) +1.63/1 0} 19. Bxf3 Kc8
{(Kb8c8) +2.74/1} 20. Bc6 bxc6 {(b7xc6) -3.97/1 0} 21. bxc6 Bxh2+
{(Bd6xh2+) -7.33/1 0} 22. Kxh2 Qb5 {(Qf5b5 Qa4a8+ Qb5b8 Qa8a6+ Qb8b7
Qa6xb7+) -M3/1 0} 23. cxb5 Rxd4 {(Rd8xd4 Qa4a8+) -M1/1 0} 24. Qa8# 1-0
It used under a second on this entire game!
[Event "Test Fixed Ply 2"]
[Date "2010.01.05"]
[White "Martin"]
[Black "Rybka 2.2 32 bit"]
[Result "1/2-1/2"]
[Opening "Scandinavian"]
[Time "17:39:50"]
[Variation "Portuguese, 4.Nf3 Qxd5 5.Be2 Nc6"]
[WhiteElo "2100"]
[Termination "normal"]
[PlyCount "97"]
[WhiteType "human"]
[BlackType "program"]
1. e4 Nf6 {(Ng8f6) +0.03/2 0} 2. Nc3 Nc6 {(Nb8c6) +0.03/2} 3. Nf3 d5
{(d7d5) +0.06/2 0} 4. exd5 Nxd5 {(Nf6xd5) +0.09/2 0} 5. Nxd5 Qxd5 {(Qd8xd5)
+0.19/2 0} 6. d4 Bg4 {(Bc8g4) +0.40/2 0} 7. c3 O-O-O {(OOO) +0.40/2 0} 8.
Be2 e6 {(e7e6) +0.23/2 0} 9. O-O Kb8 {(Kc8b8) +0.10/2 0} 10. a4 Bd6
{(Bf8d6) +0.17/2 0} 11. b4 h5 {(h7h5) +0.18/2 0} 12. b5 Na5 {(Nc6a5)
+0.10/2 0} 13. Rb1 Bf5 {(Bg4f5) +0.30/2 0} 14. Bd3 Nc4 {(Na5c4) +0.27/2 0}
15. Qe2 Bxd3 {(Bf5xd3) +0.25/2 0} 16. Qxd3 Rhe8 {(Rh8e8) +0.21/2 0} 17. Rd1
e5 {(e6e5) +0.24/2 0} 18. dxe5 Qxd3 {(Qd5xd3) +0.25/2 0} 19. Rxd3 Nxe5
{(Nc4xe5) +0.50/2} 20. Nxe5 Bxe5 {(Bd6xe5) +0.54/2 0} 21. Rxd8+ Rxd8
{(Re8xd8 Bc1b2 Rd8d2 g2g3 Kb8c8) +0.75/5 0} 22. Bb2 b6 {(b7b6 g2g3) +0.52/2
0} 23. h4 Kb7 {(Kb8b7) +0.52/2} 24. g3 Rd2 {(Rd8d2) +0.80/2} 25. Ba1 Ra2
{(Rd2a2) +0.92/2 0} 26. Re1 f6 {(f7f6) +0.99/2 0} 27. f4 Bd6 {(Be5d6)
+1.12/2 0} 28. f5 Bf4 29. gxf4 {(g3xf4 Ra2xa4 Re1f1 c7c6 b5xc6+) +0.88/2 0}
Rxa4 {(Ra2xa4 Re1f1 c7c6 b5xc6+) -0.94/2 0} 30. Rf1 c6 {(c7c6 b5xc6+)
-0.85/2 0} 31. bxc6+ Kxc6 {(Kb7xc6 Kg1g2 Ra4a5 c3c4 Ra5a4) -0.85/2} 32. Bb2
Ra5 {(Ra4a5) -0.85/2 0} 33. Ra1 Rxf5 {(Ra5xf5) -0.42/2} 34. Rxa7 Rxf4
{(Rf5xf4) -0.50/2 0} 35. Rxg7 Rxh4 {(Rf4xh4) -0.70/2 0} 36. Rf7 Rf4
{(Rh4f4) -0.44/2 0} 37. Rf8 Kc5 {(Kc6c5) -0.46/2 0} 38. c4 Kxc4 {(Kc5xc4)
-0.08/2 0} 39. Rxf6 Rxf6 {(Rf4xf6) +0.10/2 0} 40. Bxf6 Kb3 {(Kc4b3 Bf6a1
Kb3c4 Ba1f6) 0.00/2 0} 41. Kh2 Ka2 {(Kb3a2 Bf6h4) 0.00/2 0} 42. Kh3 Ka3
{(Ka2a3) 0.00/2} 43. Kh4 Ka2 {(Ka3a2 Kh4h3 b6b5 Bf6c3 h5h4 Bc3e5) 0.00/2 0}
44. Kxh5 Kb3 {(Ka2b3) 0.00/2 0} 45. Be7 b5 {(b6b5) 0.00/2} 46. Kg5 Ka2
{(Kb3a2) 0.00/2} 47. Kf5 Kb3 {(Ka2b3) 0.00/2 0} 48. Ke5 b4 {(b5b4 Ke5f5
Kb3c2 Be7xb4) 0.00/2 0} 49. Bxb4 {Insufficient material} 1/2-1/2
It used about 2s on this latter game and it was noticeably harder to
trick into a corner.
BTW is there a way in Arena to switch to the other convention of white
winning is positive and black winning is negative?
I have tried to play Rybka under CB at nominal 10s+0s for the computer
and tournament rules for me. This gives the engine a full strength 4-8
ply search and extensions to around 15 ply on forced tactical variations
(avoiding them is a good idea). So far I haven't been able to beat this
combination and it is nowhere near running out of time (3s used out of
10 for 40 moves). I have managed the odd draw. I will try turning ponder
off next to see if that will give me enough of an edge to win.
Regards,
Martin Brown