Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / FYI: Critter 1.6
1 2 Previous Next  
- - By Iptables (**) Date 2012-06-15 07:43 Edited 2012-06-15 07:52
http://www.vlasak.biz/critter/

Thanks to Richard Vida

lg
Parent - - By Uly (Gold) Date 2012-06-15 07:55
:cool:
Parent - By Chaotic Chess (****) Date 2012-06-15 07:58
BIG thanks!!! Is there any further improvements with the previous Critter 1.5 38 version? (Leiden 2012 2nd place finish)

EDIT: oops was answered already :cool:
Parent - - By Richard Vida (**) Date 2012-06-15 07:58
Recent changes/additions:

- Automatic detection of SEE4.1 capable CPU. Both SSE4/non-SSE4 codepaths are now compiled in a single executable and chosen dynamically at runtime.

- Large pages suppport. This requires sufficient user privileges and right OS settings.

- Own book is automatically turned off when analysing.

- Engine now honors multi-pv even when the root position is a tablebase hit

- Session file now supports IDEA. SF writes are disabled at root when any moves were excluded.

- Session file writes are now protected by OS IPC mechanisms (semaphores/mutexes). Same session file can be accessed concurrenlty by multiple engine isntances without risking corruption (useful when running IDEA)

- Removed UCI option "SF strategy", now its value is hardwired to "depth"

- New UCI option "SF move limit" which disables learning after game has reached given number of moves. When set to 0 this option is ignored.

- New UCI option "SF material limit" (0..32). Disables learning when total amount of material is less than the given amount [Q=6, R=3, B&N=1]. When set to 0 this option is ignored.

- Gameplay related changes:
* Tweaked blocked-pawn recognizer
* Skewer detection in eval
* King safety tweaks
* Pseudo-contempt: slightly increased preference of moves increasing pressure on the opponents king
* slightly increased preference of pawn pushes in semi-blocked positions
* Recognizing more types of drawish endgames

- new console mode commands "sf probe", "sf delete", "sf store" for mainpulating session file entries

Richard
Parent - - By Uly (Gold) Date 2012-06-15 08:02
Thank you very much Richard!

> - new console mode commands "sf probe", "sf delete", "sf store" for mainpulating session file entries


Okay, this sounds very interesting, how does it work?
Parent - - By Richard Vida (**) Date 2012-06-15 20:40

>>> - new console mode commands "sf probe", "sf delete", "sf store" for mainpulating session file entries
> Okay, this sounds very interesting, how does it work?


Sorry, I'm very busy right now. I will post a brief "tutorial" soon.
Parent - - By Uly (Gold) Date 2012-06-16 01:16 Edited 2012-06-16 08:54
[EDIT] - Please ignore this message and the one of me below, I was just failing to disable LP in the OS.

Richard, when Critter tries to use Large pages and fails, it sends my whole system into memory swapping, I can no longer use Critter with another engine as that one will be memory swapped when Critter is loaded.

This was the same behavior that made me stop using Gull 1.2 :sad: [EDIT]: And yay, after fixing this it means I'm going to be able to use Gull now! :yell:
Parent - - By Barnard (Bronze) Date 2012-06-16 01:18
cant you ''tell'' Critter to dont use large pages?so the problem is solved
Parent - - By Harvey Williamson (*****) Date 2012-06-16 01:19
No, same problem with latest Houdini. It makes a fair engine match on 1 machine very difficult.
Parent - By Ingo (***) Date 2012-06-16 07:23
Just switch LP off in the testing computer OS. You want to switch it off anyhow ... absolutly no prob here - with no engine!

And still I have a "main" computer (the fastest one) where LP is on. I am using this for analysis or some single games from time to time. (LP doesn't work for automated testing in general. It is the OS which is messing up LP after a few load and unload operations, not the engine)

Bye
Ingo
Parent - - By Uly (Gold) Date 2012-06-16 01:22 Edited 2012-06-16 08:53
[EDIT] - This only happened because after disabling LPs on the OS I also needed to restart.

How? It tries to use them when loading, before the GUI sends parameters, so unless Critter has some kind of extra configuration file (ala rybka.config or Crafty), it always swaps the memory of all my other applications when loaded.

AFAIK, Houdini 1.5a also tries to use Large pages if possible, but it doesn't memory swap the OS.:smile:
Parent - By Harvey Williamson (*****) Date 2012-06-16 01:25
If you try an engine match between houdini and critter is it possible for them both to get large pages? I suspect only 1 engine will get them.
Parent - - By Barnard (Bronze) Date 2012-06-16 01:26
well,disable LP every time you must load Critter 1.6,and enable againt when you want to lod another engine

it is very time-wasting,but if you want to use Critter 1.6,is the only way you can use without swapping
Parent - - By Harvey Williamson (*****) Date 2012-06-16 01:29
This is a pain to do. You want LP on for analysis but probably off for an engine match on the same machine. I think all engines should have an off parameter.
Parent - - By Geomusic (*****) Date 2012-06-17 18:13
Uly maybe this will clear up why your (C) goes into swapping land....http://www.ibm.com/developerworks/linux/library/l-mem26/
Parent - By Uly (Gold) Date 2012-06-18 08:03
Switching to Linux?
Parent - Date 2012-06-16 01:35
Parent - - By Ray (****) Date 2012-06-16 07:30

> cant you ''tell'' Critter to dont use large pages?so the problem is solved


The problem is solved if large pages are disabled in Windows (which is the default). It can't possibly cause you any problems in that situation.
Parent - By Uly (Gold) Date 2012-06-16 08:37
Can you please tell me how to disable Large Pages on the OS? What I did was going to the Group Policy editor and removing "Lock Pages In Memory" permission from Administrator and Everyone, but the problems continue.

Unless, it only takes effect after restarting Windows :eek:, let me check...
Parent - - By Uly (Gold) Date 2012-06-16 08:50
Oh, awesome! All I needed was to restart Windows,and now the problem is gone!

I apologize for the time that may have been wasted by Richard or other people reading my messages, it was user fault (my fault.)

I even think I'm going to branch them out of this thread.
Parent - By Richard Vida (**) Date 2012-06-16 10:01
If the GUI supports passing command line arguments to the engine, LP can be turned off by putting "setoption name large pages value false" on command line. eg. "critter.exe setoption name large pages value false".
Parent - By Barnard (Bronze) Date 2012-06-16 18:56
thanks for the info :smile:
Parent - - By Richard Vida (**) Date 2012-06-16 10:25

>>> - new console mode commands "sf probe", "sf delete", "sf store" for mainpulating session file entries
> Okay, this sounds very interesting, how does it work?


Originally I used them for debugging purposes, but I thought it won't hurt if I let them stay.

Without parameters "sf probe" shows what is stored int the SF for the current position, "sf delete" deletes the current position from SF.
With a numeric parameter eg. "sf probe 3" the engine essentially performs what is called "perft" (enumerating all move paths from the current position to depth 3) and probes the SF for each encountered position.


setoption name session file value test.csf
position fen rnbqkbnr/pppp1ppp/8/4p3/4P3/5N2/PPPP1PPP/RNBQKB1R b KQkq
sf probe 3

depth score line
-------------------
20   +0.23 Na6 Ng1 Nb8 Nf3
12+  +0.17 Nc6 Nc3 Nf6 Bc4
13   +0.17 Nc6 Nc3 Nf6
20   +0.23 Nc6 Ng1 Nb8 Nf3
12   +0.17 Nc6 Bc4 Nf6 O-O
13   +0.17 Nc6 Bc4 Nf6
17   +0.18 Nc6 Bb5 Nf6 O-O
16   +0.23 Nc6 Bb5 Nd4 Nxd4
14   +0.25 Nc6 Bb5 Bc5 O-O
15   +0.32 Nc6 Bb5 Bd6 O-O
15   +0.27 Nc6 Bb5 Be7 O-O
13   +0.39 Nc6 Bb5 a6 Bxc6
18   +0.18 Nc6 Bb5 Nf6
13+  +0.10 Nc6 d4 exd4 Nxd4
11+  +0.10 Nc6 d4 exd4
18   +0.23 Nc6 Bb5
12+  +0.17 Nf6 Nc3 Nc6 Bc4
20   +0.23 Nf6 Ng1 Ng8 Nf3
12   +0.17 Nf6 Bc4 Nc6 O-O
17   +0.18 Nf6 Bb5 Nc6 O-O
20   +0.23 Nh6 Ng1 Ng8 Nf3
20   +0.23 Ne7 Ng1 Ng8 Nf3
20   +0.23 Ba3 Ng1 Bf8 Nf3
20   +0.23 Bb4 Ng1 Bf8 Nf3
20   +0.23 Bc5 Ng1 Bf8 Nf3
14   +0.25 Bc5 Bb5 Nc6 O-O
20   +0.23 Bd6 Ng1 Bf8 Nf3
15   +0.32 Bd6 Bb5 Nc6 O-O
20   +0.23 Be7 Ng1 Bf8 Nf3
15   +0.27 Be7 Bb5 Nc6 O-O
20   +0.23 Qh4 Ng1 Qd8 Nf3
20   +0.23 Qg5 Ng1 Qd8 Nf3
20   +0.23 Qf6 Ng1 Qd8 Nf3
20   +0.23 Qe7 Ng1 Qd8 Nf3
13   +0.39 a6 Bb5 Nc6 Bxc6
19   +0.23 Nc6
-------------------
36 entries found


"sf delete 3" - same as above but deletes every position it encounters (and the message at the end would be "36 entries deleted").

"sf store" - manual override of the stored score for the current position. It takes 3 parameters score, move and depth. Move is checked for legality of course.

eg.
sf store +0.14 Nc6 66

or
sf store -M13 a6 20
:)

Specifying a very large depth will prevent possible overwrites, but any line with this position will be cut off and not searched again.

One thing that may seem strange: scores are entered in pawn=1.00 metric, but the internal representation is in pawn=2.56 metric. So rounding errors are very probable. You "sf store" a score of +0.14, but "sf probe" shows you +0.13. Specifying score to 3 decimal places helps in some cases (+0.137 or such), but the rounding problems can not be entirely avoided.
Parent - - By Uly (Gold) Date 2012-06-16 11:00

> "sf store" - manual override of the stored score for the current position. It takes 3 parameters score, move and depth. Move is checked for legality of course.
>
> eg.
> sf store +0.14 Nc6 66
>
> or
> sf store -M13 a6 20
> :)


Oh, thanks, this sounds great Richard! I think Critter is the first engine that allows the user to modify its learning, this is very powerful because entire "Fortress" positions can now be tagged 0.00 and avoided, known lost positions can now be set to mate scores, positions user dislikes (like ones where engine wants to trade a good piece) can be pruned from the distance, and now I can tag 6men Tablebases correctly in one shot :smile:

Well done Richard, keep it up! :cool:
Parent - - By Pia (****) Date 2012-06-16 15:53

>its learning


So "Session File" is not just some type of log file but learning file!? Good news, thanks!
Parent - By Richard Vida (**) Date 2012-06-16 16:18

>> its learning


> So "Session File" is not just some type of log file but learning file!? Good news, thanks!


Yes, it is a learning file + equivalent of a persistent hash. It can save enormous amounts of analysis time when used properly.
Parent - - By Uly (Gold) Date 2012-06-17 08:23 Edited 2012-06-17 08:25
Yes, the best I've seen since it doesn't have several of the flaws of other implementations of other engines (I haven't tested Houdini Learning, but I know the user can't edit it :wink:.)

"Learning" is a feature used by books and other engines to tweak scores/move choice depending on game results, that's why a different name was required for the feature. "Preserve" is a word often used to protect analysis contents but those are discarded when the engine is unloaded. "Persistent Hash" was confusing new users because there are features in where the hash is made persistent by saving it whole to disk, yet, its entries are eventually overwritten like other hash entries. All those were bad names.

"Session File" creates a File that is loaded when you start your analysis Session with Critter that stores all important positions of analysis. You can use Interactive Analysis to show refutations and better moves to Critter, and backtrack to the root so Critter either shows your info as best or shows an alternative root move that it seems as better, thanks to Session File.

Now I can claim any engine without such a feature doesn't care much about users that use the engines for analyzing chess positions, in this day and age there is no reason for the engine to discard everything that is analyzed when it's unloaded, when it could be saving it to disk!
Parent - - By Banned for Life (Gold) Date 2012-06-17 08:34
Weren't you the guy who was responsible for having a similar feature removed from R4? :twisted:
Parent - By Uly (Gold) Date 2012-06-17 08:55
Here's the instant (2 years later) replay:

[...] Some of the engine parameters, namely persistent hash, won't be 100% polished. [...]

[...] If any urgent new problems arise, please make a new post here with bold letters and other bad internet etiquette.[...]

~Vas (emphasis mine.)

At least 3 Persistent hash items need urgent fixing:

1. The "Enabled" box crashes the engine (fixing so we can use PH)

2. Write depth is used as play depth (fixing so we can use PH for games)

3. Exclude moves makes Rybka learn that whatever moves were included in searchmoves are best (fixing so people can use PH together with IDEA) - My suggestion is that Rybka should only save at what depth the best move is best, and its score, but should not reach this depth immediately, she should still checks the alternative moves (this is Shredder learning behavior), this would also solve problems with propagation, as long as Rybka writes for current depth and lower (e.g. Rybka should know that the move that is best at depth 16 is also best at depth 15 and lower, without the user needing to switch to Multi-PV 2 and back)

(I guess there's no time for the other issues)

Save/Load Hash buttons are also broken, with no time for testing I suggest you get rid of them.


~Uly

Persistent Hash is intentionally removed.

~Vas

So yeah, it wasn't my fault, Vas asked for it! In my...huh...defense, I just saved people from having to see Rybka mating with Bf3+ in the opening position. And nobody but me was using Persistent Hash anyway! They found having to switch MutiPV lines constantly not worth the hassle.
Parent - - By Uly (Gold) Date 2012-06-21 07:24
Okay, I've tried it under Arena but I'm getting out of range errors?

-->1:sf store +0.18 Bb1 21
<--1:depth "21" out of range
-->1:sf store +0.18 Bb1 20
<--1:depth "20" out of range
-->1:sf store +0.18 Bb1 19
<--1:depth "19" out of range

What depth would be inside the range? :confused:
Parent - - By Richard Vida (**) Date 2012-06-21 09:43

> Okay, I've tried it under Arena but I'm getting out of range errors?
>
> -->1:sf store +0.18 Bb1 21
> <--1:depth "21" out of range
> -->1:sf store +0.18 Bb1 20
> <--1:depth "20" out of range
> -->1:sf store +0.18 Bb1 19
> <--1:depth "19" out of range
>
> What depth would be inside the range? :confused:


The upper bound is 64.
The lower bound supposed to be "SF Write depth", but I made a mistake and it is actually 2 * "SF Write Depth". Will be fixed soon. Until then, a possible workaround is to temporarily halve the write depth.
Parent - By Uly (Gold) Date 2012-06-21 10:14
Okay, thanks!
Parent - By Permanent Brain (*****) Date 2012-06-15 09:56
Thx alot! - Do you have an estimation of the strength increase?

We are flabbergasted. :lol:
Parent - By irulats (****) Date 2012-06-15 11:26
Thanks a million, Richard!
Parent - By keoki010 (Silver) Date 2012-06-15 17:54
Many, many, many thanks Richard!!!!!
Parent - - By dragon49 (****) Date 2012-06-15 23:03
Richard,

Thanks as always.

What elo improvement do you expect it to have over the previous version.
Parent - - By Richard Vida (**) Date 2012-06-17 12:23

> What elo improvement do you expect it to have over the previous version


I don't expect any elo improvement at all. In fact, considering all the extra stuff I packed in, I am happy if it would not show an elo regression. But subjectively, it analyses much better than the previous version in a wide variety of positions. Unfortunately this cannot be measured by engine-engine matches. I had to give up on the usual testing methodology (does not increase elo => throw that feature out) because I would have to reject most of the new changes.

Watching the games played with long time controls I think the most notable weakness is in passed pawn handling. Critter often underestimates their power especially if all heavy pieces are still on the board. It does not grasp the concept of tying up opponent's pieces with a promotion-threat. Even if the passed pawn can be safely captured this often comes with the cost of an extra tempo, which might be critical. Stockfish is much better in such positions, although it tends to overvalue passed pawns.

Richard

PS: ... a little off-topic trivia: is there an engine besides Critter1.6 which evaluates this position as drawn?
3k4/p7/Pp6/1Pp5/2P5/2P2K2/8/3B4 w - - 0 1
Parent - By dragon49 (****) Date 2012-06-17 15:05
Richard,

Thanks for the reply.

I'm going to play around with the test position.  I think I have another engine that will eventually evaluate it as drawn.
Parent - - By kgburcham (***) Date 2012-06-17 16:53
3k4/p7/Pp6/1Pp5/2P5/2P2K2/8/3B4 w - -


Critter 1.4 64-bit

43/67   0:01   +5.14   1.Ke4 Ke8 2.Kd5 Kd7 3.Bh5 Kc7 4.Bg4 Kb8 5.Kd6 Ka8 6.Bd1 Kb8 7.Kc6 Ka8 8.Kd7 Kb8 9.Ba4 Ka8 10.Bb3 Kb8 11.Kd8 Ka8 12.Bd1 Kb8 13.Bc2 Ka8 14.Kd7 (14.245.778) 9459
44/67   0:01   +5.14   1.Ke4 Ke8 2.Kd5 Kd7 3.Bh5 Kc7 4.Bg4 Kb8 5.Kd6 Ka8 6.Bd1 Kb8 7.Kc6 Ka8 8.Kd7 Kb8 9.Ba4 Ka8 10.Bb3 Kb8 11.Kd8 Ka8 12.Bd1 Kb8 13.Bc2 Ka8 14.Kd7 (21.621.066) 10859
45/68   0:02   +5.14   1.Ke4 Ke8 2.Kd5 Kd7 3.Bh5 Kc7 4.Bg4 Kb8 5.Kd6 Ka8 6.Bd1 Kb8 7.Kc6 Ka8 8.Kd7 Kb8 9.Ba4 Ka8 10.Bb3 Kb8 11.Kd8 Ka8 12.Bd1 Kb8 13.Bc2 Ka8 14.Kd7 (28.640.673) 11776
46/70   0:04   +5.14   1.Ke4 Ke8 2.Kd5 Kd7 3.Bh5 Kc7 4.Bg4 Kb8 5.Kd6 Ka8 6.Bd1 Kb8 7.Kc6 Ka8 8.Kd7 Kb8 9.Ba4 Ka8 10.Bb3 Kb8 11.Kd8 Ka8 12.Bd1 Kb8 13.Bc2 Ka8 14.Kd7 (57.792.188) 14092
47/74   0:06   +5.14   1.Ke4 Ke8 2.Kd5 Kd7 3.Bh5 Kc7 4.Bg4 Kb8 5.Kd6 Ka8 6.Bd1 Kb8 7.Kc6 Ka8 8.Kd7 Kb8 9.Ba4 Ka8 10.Bb3 Kb8 11.Kd8 Ka8 12.Bd1 Kb8 13.Bc2 Ka8 14.Kd7 (94.746.445) 14923
48/74   0:06   +5.14   1.Ke4 Ke8 2.Kd5 Kd7 3.Bh5 Kc7 4.Bg4 Kb8 5.Kd6 Ka8 6.Bd1 Kb8 7.Kc6 Ka8 8.Kd7 Kb8 9.Ba4 Ka8 10.Bb3 Kb8 11.Kd8 Ka8 12.Bd1 Kb8 13.Bc2 Ka8 14.Kd7 (95.478.722) 14855
49/79   0:20   +5.14   1.Ke4 Ke8 2.Kd5 Kd7 3.Bh5 Kc7 4.Bg4 Kb8 5.Kd6 Ka8 6.Bd1 Kb8 7.Kc6 Ka8 8.Kd7 Kb8 9.Ba4 Ka8 10.Bb3 Kb8 11.Kd8 Ka8 12.Bd1 Kb8 13.Bc2 Ka8 14.Kd7 (358.996.530) 17717
50/79   0:21   +5.14   1.Ke4 Ke8 2.Kd5 Kd7 3.Bh5 Kc7 4.Bg4 Kb8 5.Kd6 Ka8 6.Bd1 Kb8 7.Kc6 Ka8 8.Kd7 Kb8 9.Ba4 Ka8 10.Bb3 Kb8 11.Kd8 Ka8 12.Bd1 Kb8 13.Bc2 Ka8 14.Kd7 (372.919.898) 17624
51/79   0:21   +5.14   1.Ke4 Ke8 2.Kd5 Kd7 3.Bh5 Kc7 4.Bg4 Kb8 5.Kd6 Ka8 6.Bd1 Kb8 7.Kc6 Ka8 8.Kd7 Kb8 9.Ba4 Ka8 10.Bb3 Kb8 11.Kd8 Ka8 12.Bd1 Kb8 13.Bc2 Ka8 14.Kd7 (375.122.292) 17557
52/88   1:05   +5.14   1.Ke4 Ke8 2.Kd5 Kd7 3.Bh5 Kc7 4.Bg4 Kb8 5.Kd6 Ka8 6.Bd1 Kb8 7.Kc6 Ka8 8.Kd7 Kb8 9.Ba4 Ka8 10.Bb3 Kb8 11.Kd8 Ka8 12.Bd1 Kb8 13.Bc2 Ka8 14.Kd7 (1.253.300.682) 19256
53/88   1:07   +5.14   1.Ke4 Ke8 2.Kd5 Kd7 3.Bh5 Kc7 4.Bg4 Kb8 5.Kd6 Ka8 6.Bd1 Kb8 7.Kc6 Ka8 8.Kd7 Kb8 9.Ba4 Ka8 10.Bb3 Kb8 11.Kd8 Ka8 12.Bd1 Kb8 13.Bc2 Ka8 14.Kd7 (1.296.347.396) 19177
54/88   1:09   +5.14   1.Ke4 Ke8 2.Kd5 Kd7 3.Bh5 Kc7 4.Bg4 Kb8 5.Kd6 Ka8 6.Bd1 Kb8 7.Kc6 Ka8 8.Kd7 Kb8 9.Ba4 Ka8 10.Bb3 Kb8 11.Kd8 Ka8 12.Bd1 Kb8 13.Bc2 Ka8 14.Kd7 (1.322.491.020) 19047

Critter 1.6 64-bit

47/68   0:01   +0.05   1.Ke4 Kc7 2.Ke5 Kc8 3.Bf3 Kb8 4.Kd6 Kc8 5.Bg4+ Kb8 6.Ke5 Kc7 7.Kd5 Kb8 8.Kd6 Ka8 9.Kc6 Kb8 10.Kd5 Ka8 11.Ke5 Kb8 12.Bf3 Kc8 13.Bb7+ Kc7 14.Kd5 (24.813.564) 15926
48/70   0:02   +0.05   1.Ke4 Kc7 2.Ke5 Kc8 3.Bf3 Kb8 4.Kd6 Kc8 5.Bg4+ Kb8 6.Ke5 Kc7 7.Kd5 Kb8 8.Kd6 Ka8 9.Kc6 Kb8 10.Kd5 Ka8 11.Ke5 Kb8 12.Bf3 Kc8 13.Bb7+ Kc7 14.Kd5 (49.445.516) 16637
49/70   0:03   +0.05   1.Ke4 Kc7 2.Ke5 Kc8 3.Bf3 Kb8 4.Kd6 Kc8 5.Bg4+ Kb8 6.Ke5 Kc7 7.Kd5 Kb8 8.Kd6 Ka8 9.Bf3+ Kb8 10.Ke5 Kc8 11.Bb7+ Kc7 12.Kd5 Kb8 13.Ke6 Kc7 14.Ke5 (61.170.289) 16897
50/71   0:04   +0.05   1.Ke4 Kc7 2.Ke5 Kc8 3.Bf3 Kb8 4.Kd6 Kc8 5.Bg4+ Kb8 6.Ke5 Kc7 7.Kd5 Kb8 8.Kd6 Ka8 9.Bf3+ Kb8 10.Ke5 Kc8 11.Bb7+ Kc7 12.Kd5 Kb8 13.Ke6 Kc7 14.Ke5 (78.972.866) 17023
51/73   0:06   +0.05   1.Ke4 Kc7 2.Ke5 Kc8 3.Bf3 Kb8 4.Kd6 Kc8 5.Bg4+ Kb8 6.Ke5 Kc7 7.Kd5 Kb8 8.Kd6 Ka8 9.Bf3+ Kb8 10.Ke5 Kc8 11.Bb7+ Kc7 12.Kd5 Kb8 13.Ke6 Kc7 14.Ke5 (115.618.818) 17817
52/74   0:07   +0.05   1.Ke4 Kc7 2.Ke5 Kc8 3.Bf3 Kb8 4.Kd6 Kc8 5.Bg4+ Kb8 6.Ke5 Kc7 7.Kd5 Kb8 8.Kd6 Ka8 9.Bf3+ Kb8 10.Ke5 Kc8 11.Bb7+ Kc7 12.Kd5 Kb8 13.Ke6 Kc7 14.Ke5 (138.819.745) 17970
53/76   0:23   +0.05   1.Ke4 Kc7 2.Ke5 Kc8 3.Bf3 Kb8 4.Kd6 Kc8 5.Bg4+ Kb8 6.Ke5 Kc7 7.Kd5 Kb8 8.Kd6 Ka8 9.Bf3+ Kb8 10.Ke5 Kc8 11.Bb7+ Kc7 12.Kd5 Kb8 13.Ke6 Kc7 14.Ke5 (444.820.645) 18718
54/78   0:24   +0.05   1.Ke4 Kc7 2.Ke5 Kc8 3.Bf3 Kb8 4.Kd6 Kc8 5.Bg4+ Kb8 6.Ke5 Kc7 7.Kd5 Kb8 8.Kd6 Ka8 9.Bf3+ Kb8 10.Ke5 Kc8 11.Bb7+ Kc7 12.Kd5 Kb8 13.Ke6 Kc7 14.Ke5 (461.047.667) 18712
55/80   0:25   +0.05   1.Ke4 Kc7 2.Ke5 Kc8 3.Bf3 Kb8 4.Kd6 Kc8 5.Bg4+ Kb8 6.Ke5 Kc7 7.Kd5 Kb8 8.Kd6 Ka8 9.Bf3+ Kb8 10.Ke5 Kc8 11.Bb7+ Kc7 12.Kd5 Kb8 13.Ke6 Kc7 14.Ke5 (484.332.270) 18741
56/82   0:57   +0.05   1.Ke4 Kc7 2.Ke5 Kc8 3.Bf3 Kb8 4.Kd6 Kc8 5.Bg4+ Kb8 6.Ke5 Kc7 7.Kd5 Kb8 8.Kd6 Ka8 9.Bf3+ Kb8 10.Ke5 Kc8 11.Bb7+ Kc7 12.Kd5 Kb8 13.Ke6 Kc7 14.Ke5 (1.098.949.992) 19144
57/84   0:57   +0.05   1.Ke4 Kc7 2.Ke5 Kc8 3.Bf3 Kb8 4.Kd6 Kc8 5.Bg4+ Kb8 6.Ke5 Kc7 7.Kd5 Kb8 8.Kd6 Ka8 9.Bf3+ Kb8 10.Ke5 Kc8 11.Bb7+ Kc7 12.Kd5 Kb8 13.Ke6 Kc7 14.Ke5 (1.099.703.766) 19136
58/86   1:24   +0.05   1.Ke4 Kc7 2.Ke5 Kc8 3.Bf3 Kb8 4.Kd6 Kc8 5.Bg4+ Kb8 6.Ke5 Kc7 7.Kd5 Kb8 8.Kd6 Ka8 9.Bf3+ Kb8 10.Ke5 Kc8 11.Bb7+ Kc7 12.Kd5 Kb8 13.Ke6 Kc7 14.Ke5 (1.672.469.674) 19686
Parent - By Geomusic (*****) Date 2012-06-17 18:17
No fail high nice!
Parent - By Indrajit (***) Date 2012-06-17 17:39
Richard would you be planning to incorporate a UCI Limit Strength option in future?
Parent - By Geomusic (*****) Date 2012-06-17 18:40
Deka still log on? Couldn't someone use his attained error rates for different ratings and plug them into a chess program to make a program more human like? When it starts up it would ask you for the ELO then it would systematically play 10 hardwired positions very quickly within its memory and self-adjust its pruning until it arrives at the desired error rate.
Parent - - By dragon49 (****) Date 2012-06-17 21:08
I expected Zappa to see the truth, but after 6 hours, it still thinks white is winning:

3k4/p7/Pp6/1Pp5/2P5/2P2K2/8/3B4 w - - 0 1


Analysis by Zappa Mexico II 64:

1.Ke4 Ke8 2.Kd5 Kd7 3.Bg4+ Kc7 4.Bh3 Kb8 5.Kd6 Ka8 6.Bf5 Kb8 7.Kc6 Ka8 8.Kd6
  +-  (3.43)   Depth: 20/31   00:00:00  2245kN
1.Ke4 Ke8 2.Kd5 Kd7 3.Bg4+ Kc7 4.Bf5 Kb8 5.Kd6 Ka8 6.Bd7 Kb8 7.Kc6 Ka8 8.Kd5 Kb8 9.Kd6 Ka8 10.Bc6+ Kb8 11.Bd5 Kc8 12.Be4 Kb8 13.Kc6
  +-  (3.43)   Depth: 20/31   00:00:00  2527kN
1.Ke4 Ke8 2.Kd5 Kd7 3.Bg4+ Kc7 4.Bf3 Kd7 5.Bg4+
  +-  (3.43)   Depth: 21/32   00:00:00  5532kN
1.Be2 Kd7 2.Ke4 Ke7 3.Kd5 Kd7 4.Bf3 Kc7 5.Bg4 Kb8 6.Kd6 Ka8 7.Bd7 Kb8 8.Kc6 Ka8 9.Bf5 Kb8 10.Be4 Kc8 11.Bf5+ Kb8
  +-  (3.44)   Depth: 21/32   00:00:00  6320kN
1.Be2 Kd7 2.Ke4 Ke7 3.Kd5 Kd7 4.Bf3 Kc7 5.Bg4 Kb8 6.Kd6 Ka8 7.Kc6 Kb8 8.Bf3 Kc8 9.Kd5 Kd7 10.Be4 Kc7 11.Bf5 Kb8 12.Be4 Ka8 13.Bf3 Kb8 14.Kd6 Kc8 15.Bg4+ Kb8 16.Kc6 Ka8 17.Kd6 Kb8
  +-  (3.44)   Depth: 21/32   00:00:01  7258kN, tb=2
1.Be2 Kd7 2.Ke4 Ke7 3.Kd5 Kd7 4.Bf3 Kc7 5.Bg4 Kb8 6.Kd6 Ka8 7.Kc6 Kb8 8.Kd7 Ka8
  +-  (3.44)   Depth: 22/32   00:00:01  8368kN, tb=26
1.Be2 Kd7 2.Ke4 Ke7 3.Kd5 Kd7 4.Bf3 Kc7 5.Ke6 Kb8 6.Kd6 Kc8 7.Bb7+ Kb8 8.Bd5 Kc8 9.Be6+ Kb8 10.Kc6 Ka8 11.Bd5 Kb8 12.Kd6
  +-  (3.44)   Depth: 22/33   00:00:01  11345kN, tb=45
1.Be2 Kd7 2.Ke4 Ke7 3.Kd5 Kd7 4.Bf3 Kd8 5.Kd6 Kc8 6.Bb7+ Kb8 7.Bd5 Kc8
  +-  (3.43)   Depth: 23/34   00:03:06  1922mN, tb=75
1.Bc2 Kd7 2.Ke4 Kc7 3.Kf4 Kb8 4.Bf5 Ka8 5.Ke5 Kb8 6.Kf4
  +-  (3.44)   Depth: 23/35   00:03:58  2510mN, tb=75
1.Bc2 Kd7 2.Ke4 Kc7 3.Kf4 Kb8 4.Kf5 Kc7 5.Ke4 Kb8 6.Kd5 Kc7 7.Be4 Kd7 8.Bh7 Kc7 9.Be4
  +-  (3.44)   Depth: 23/35   00:03:59  2512mN, tb=173
1.Bc2 Kd7 2.Ke4 Kc7 3.Kf4
  +-  (3.44)   Depth: 24/35   00:04:00  2530mN, tb=173
1.Bc2 Kd7 2.Ke4 Kc7 3.Kf4 Kc8 4.Ke5 Kb8 5.Ke4 Ka8 6.Kf4 Kb8 7.Ke5 Kc7 8.Bf5 Kb8 9.Kf4 Ka8 10.Bc2
  +-  (3.44)   Depth: 24/37   00:04:03  2560mN, tb=173
1.Bc2 Kd7 2.Ke4 Kc7 3.Kf4 Kd6 4.Bb3 Ke7 5.Ke5 Kd8 6.Bc2 Kc8 7.Kf5 Kb8 8.Kg4 Kc7 9.Kf4
  +-  (3.44)   Depth: 25/37   00:04:06  2586mN, tb=183
1.Bc2 Kd7 2.Ke4 Kc7 3.Kf4 Kd7 4.Ke4
  +-  (3.44)   Depth: 25/37   00:04:07  2594mN, tb=183
1.Bc2 Kd7 2.Ke4 Kc7 3.Kf4 Kb8 4.Kf5 Kc7 5.Kg4 Kd6 6.Kf5 Ke7 7.Kf4 Ke8 8.Kg4 Kd7 9.Bf5+ Kd8 10.Bc2 Kd7
  +-  (3.44)   Depth: 26/37   00:04:14  2662mN, tb=523
1.Bc2 Kd7 2.Ke4 Kc7 3.Kf4 Kb8 4.Ke5 Kc7 5.Bf5 Kb8 6.Kd6 Ka8 7.Be4+ Kb8 8.Bc6 Kc8 9.Ba8 Kb8 10.Bg2 Kc8 11.Bh3+ Kb8 12.Bf5
  +-  (3.44)   Depth: 26/37   00:04:34  2850mN, tb=2158
1.Bc2 Kd7 2.Kf4 Kc7 3.Ke5 Kd7 4.Ba4 Ke7 5.Bb3 Kd8 6.Kd6 Kc8 7.Bc2 Kb8 8.Be4 Kc8 9.Kd5 Kb8 10.Bf5 Kc7 11.Bc2 Kb8 12.Bf5
  +-  (3.44)   Depth: 27/45   00:47:09  25394mN, tb=2158
1.Bc2 Kd7 2.Kf4 Kc7 3.Ke5 Kd7 4.Ba4 Kc7 5.Bc2
  +-  (3.44)   Depth: 27/45   00:50:22  25855mN, tb=2502
1.Bc2 Kd7 2.Kf4 Kc7 3.Ke5 Kd7 4.Bf5+ Kc7 5.Be4 Kb8 6.Kd6 Kc8 7.Bb7+ Kb8 8.Bd5 Kc8 9.Bb7+
  +-  (3.44)   Depth: 28/45   01:07:24  27823mN, tb=2551
1.Bc2 Kd7 2.Kf4 Kc7 3.Ke5 Kd7 4.Bf5+ Kc7 5.Be4 Kb8 6.Bb7 Kc7 7.Be4
  +-  (3.44)   Depth: 28/45   01:08:02  27869mN, tb=3099
1.Bc2 Kd7 2.Kf4 Kc7 3.Ke5 Kd7 4.Bf5+ Kc7 5.Be4 Kb8 6.Bb7 Kc7 7.Bc6 Kd8
  +-  (3.44)   Depth: 29/45   01:08:36  27932mN, tb=4262
1.Bc2 Kd7 2.Kf4 Kc7 3.Ke5 Kd7 4.Bf5+ Kc7 5.Be4 Kb8 6.Bb7 Kc7 7.Be4
  +-  (3.44)   Depth: 29/45   01:09:51  28029mN, tb=4499
1.Bc2 Kd7 2.Kf4 Kc7 3.Ke5 Kd7 4.Bf5+ Kc7 5.Be4 Kb8 6.Bb7 Kc7 7.Be4
  +-  (3.44)   Depth: 30/47   01:10:28  28223mN, tb=8171
1.Bc2 Kd7 2.Kf4 Kc7 3.Ke5 Kd7 4.Bf5+ Kc7 5.Be4 Kb8
  +-  (3.44)   Depth: 30/47   01:14:27  28922mN, tb=13778
1.Bc2 Kd7 2.Kf4 Kc7 3.Ke5 Kd7 4.Bf5+ Kc7 5.Be4 Kb8 6.Bb7 Kc7 7.Kf6 Kd6 8.Bc8 Kc7 9.Bb7
  +-  (3.44)   Depth: 31/47   01:24:45  29749mN, tb=15565
1.Bc2 Kd7 2.Kf4 Kc7 3.Ke5 Kd7 4.Bf5+ Kc7 5.Be4 Kb8
  +-  (3.44)   Depth: 31/48   01:30:49  31405mN, tb=26310
1.Bc2 Kd7 2.Kf4 Kc7 3.Ke5 Kd7 4.Bf5+ Kc7 5.Be4 Kb8 6.Bb7 Kc7 7.Bd5 Kd8 8.Be6
  +-  (3.44)   Depth: 32/48   01:34:15  32113mN, tb=27712
Parent - - By Uly (Gold) Date 2012-06-18 08:07
I used my stock pile of programs with such knowledge of drawn positions, all of them failed it :smile:

The position in itself is not important, what matters is that an engine may wrongly think it's winning and aim for such a position, while Critter 1.6 will avoid it and go for a really winning position instead.
Parent - - By dragon49 (****) Date 2012-06-18 14:56
I'm surprised as I have Never had Zappa fail to solve something.
Parent - - By InspectorGadget (*****) Date 2012-06-19 19:56

> I'm surprised as I have Never had Zappa fail to solve something.


Really? I have see it fail at some positions that were posted on the forum.  :cool:
Parent - - By dragon49 (****) Date 2012-06-20 00:29
Can you please reply with a link to some of the posts.
Parent - By InspectorGadget (*****) Date 2012-06-21 17:30
It was a couple of years back. It might take a long time to find them. I will check if I saved them in my hard drive. I am not kidding thee.
Parent - By siah (***) Date 2012-06-19 07:34
Even shredder shows winning eval in this position.
Parent - By RFK (Gold) Date 2012-06-17 18:16
Wow, nice! Great work. Spot on, Sir! :cool:
Up Topic The Rybka Lounge / Computer Chess / FYI: Critter 1.6
1 2 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill