Winboard Forum

by **Fabien Letouzey** » 24 Jun 2004, 11:25

Geschrieben von:/Posted by: Fabien Letouzey at 24 June 2004 12:25:49:

Hello,
Joachim Rang has proposed material settings for Fruit 1.5. I suggest that they are used by most as if Fruit were updated.
Two things are important:
1) if you do apply the changes please name the engine "Fruit 1.5 tralala" (the name requested by Joachim), or indicate the change in some other way if the name is too long (e.g. Fruit 1.5t). It is a "different" engine from Fruit 1.5 and for instance should have a independent rating.
2) Do not apply the settings if you want to rate only publicly-available engines (as I understand the UEL does), as I consider them as a personality that will only be used by few people.
That beeing said, these are the settings in PolyGlot syntax (to be added in the [Engine] section). Applying them in a UCI interface should be straightforward:
---
Pawn = 100
Pawn (Endgame) = 100
Knight = 350
Knight (Endgame) = 320
Bishop = 350
Bishop (Endgame) = 320
Rook = 550
Rook (Endgame) = 520
Queen = 1050
Queen (Endgame) = 1000
Bishop Pair = 15
Bishop Pair (Endgame) = 50
---
Thanks,
Fabien.
PS: These settings are not patented!

by **Robert Allgeuer** » 24 Jun 2004, 14:15

Geschrieben von:/Posted by: Robert Allgeuer at 24 June 2004 15:15:22:
Als Antwort auf:/In reply to: Fruit 1.5 tralala geschrieben von:/posted by: Fabien Letouzey at 24 June 2004 12:25:49:

Are there indications that these settings are stronger than the default ones?
Do you have intentions to use these settings in future versions of Fruit as default?
Thanks
Robert

by **Fabien Letouzey** » 24 Jun 2004, 14:25

Geschrieben von:/Posted by: Fabien Letouzey at 24 June 2004 15:25:13:
Als Antwort auf:/In reply to: Re: Fruit 1.5 tralala geschrieben von:/posted by: Robert Allgeuer at 24 June 2004 15:15:22:

Hi Robert,

Yes:
http://f11.parsimony.net/forum16635/messages/67693.htm
Yes.
Fabien.

by **Igor Gorelikov** » 24 Jun 2004, 14:53

Geschrieben von:/Posted by: Igor Gorelikov at 24 June 2004 15:53:04:
Als Antwort auf:/In reply to: Fruit 1.5 tralala geschrieben von:/posted by: Fabien Letouzey at 24 June 2004 12:25:49:

Which settings you prefer for the main cycle of New Infinfite Loop?
Igor

by **Fabien Letouzey** » 24 Jun 2004, 14:59

Geschrieben von:/Posted by: Fabien Letouzey at 24 June 2004 15:59:45:
Als Antwort auf:/In reply to: Re: Fruit 1.5 tralala geschrieben von:/posted by: Igor Gorelikov at 24 June 2004 15:53:04:

Hi Igor,

Tralala, you can consider it as a small update.
Thanks,
Fabien.

by **Igor Gorelikov** » 24 Jun 2004, 15:03

Geschrieben von:/Posted by: Igor Gorelikov at 24 June 2004 16:03:56:
Als Antwort auf:/In reply to: Re: Fruit 1.5 tralala geschrieben von:/posted by: Fabien Letouzey at 24 June 2004 15:59:45:

.

by **Dieter Bürßner** » 25 Jun 2004, 00:10

Geschrieben von:/Posted by: Dieter Bürßner at 25 June 2004 01:10:11:
Als Antwort auf:/In reply to: Fruit 1.5 tralala geschrieben von:/posted by: Fabien Letouzey at 24 June 2004 12:25:49:

Interesting. Rook is relatively worth less in the endgame than in earlier game phases. I thought, the contrary was the case ... Or are the tables tilted by the fact that pawn is set the same in all phases, and in reality is on average worth more in the endgame?
Regards,
Dieter

by **Uri Blass** » 25 Jun 2004, 03:11

Geschrieben von:/Posted by: Uri Blass at 25 June 2004 04:11:14:
Als Antwort auf:/In reply to: Re: Fruit 1.5 tralala geschrieben von:/posted by: Dieter Bürßner at 25 June 2004 01:10:11:

Rook against pawn is often tablebases draw when in the middle game advantage of rook against pawn is usually decisive so I am not sure if rook is worth more in the endgame relative to pawn.
Uri

by **Uri Blass** » 25 Jun 2004, 03:18

Geschrieben von:/Posted by: Uri Blass at 25 June 2004 04:18:01:
Als Antwort auf:/In reply to: Re: Fruit 1.5 tralala geschrieben von:/posted by: Uri Blass at 25 June 2004 04:11:14:

I can add that rook against 2 pawns may be even losing for the rook in some cases in the endgame.
When you say that rook is relatively worth more in the endgame you need to tell me to what you compare.
If you compare rook to the sum of bishop and knight than you can see that advantage of bishop and knight against rook and pawn worth more in the middle game based on the table.
Uri

by **Joachim Rang** » 25 Jun 2004, 09:23

Geschrieben von:/Posted by: Joachim Rang at 25 June 2004 10:23:10:
Als Antwort auf:/In reply to: Re: Fruit 1.5 tralala geschrieben von:/posted by: Dieter Bürßner at 25 June 2004 01:10:11:

Hi Dieter,
these figures aren't necessarily the best. It's just a set which was considerably stronger than the default ones and I didn't feel like spending the next month squeezing another 5 elopoints via tuning considering the fact that Fabien is implementing endgame knowledge and that after that another tuning is probably necessary. Fruit 1.5 has limited passer knowledge and only a (relatively small) bonus for passers dependent on the rank. So it often misunderestimates ;-)

the dangerousness of passers. With these reduced values for the endgame this problem is a bit eased (but not solved).
regards Joachim

by **Uri Blass** » 25 Jun 2004, 10:12

Geschrieben von:/Posted by: Uri Blass at 25 June 2004 11:12:11:
Als Antwort auf:/In reply to: Re: Fruit 1.5 tralala geschrieben von:/posted by: Joachim Rang at 25 June 2004 10:23:10:

I do not think that the bonus is relatively small and it is probably bad idea to increase the bonus.
There are cases when passed pawns do not help
see the following position when fruit has a significant difficulty to find the best move and bigger problem to see draw score(I gave up after some minutes with no draw score).
Note that movei does not perform better than Fruit here and cannot avoid Kxh4 at reasonable time.

movei00_8_245 - Movei00_8_247, Computer chess game 2004
8/8/8/4k3/1b3pKp/5BpP/8/8 w - - 0 1
Analysis by Fruit 1.5:
98.Kxh4
-+ (-2.46) Depth: 1/3 00:00:00
98.Kxh4 Bc5
-+ (-2.59) Depth: 2/3 00:00:00
98.Kxh4 Be7+ 99.Kg4 Bc5
-+ (-2.47) Depth: 3/5 00:00:00
98.Kxh4 Bd6 99.Kg5 Bc5
-+ (-2.38) Depth: 4/7 00:00:00
98.Kxh4 Bc5 99.Kg5 Be7+ 100.Kg4 Bc5
-+ (-2.47) Depth: 5/7 00:00:00
98.Kxh4 Kf5 99.Kh5 Bc5 100.Bd5 Bd4
-+ (-2.46) Depth: 6/10 00:00:00 2kN
98.Kxh4 Kf5 99.Kh5 Bc5 100.h4 Bd4 101.Bd5
-+ (-2.37) Depth: 7/12 00:00:00 5kN
98.Kxh4 Kf5 99.Kh5 Be7 100.h4 Ke5 101.Kg4 Bc5
-+ (-2.37) Depth: 8/13 00:00:00 9kN
98.Kxh4 Kf5 99.Kh5 Bc5 100.h4 Bd4 101.Bd5 Ke5 102.Bc4
-+ (-2.41) Depth: 9/14 00:00:00 15kN
98.Kxh4 Kf5 99.Kh5 Bc5 100.h4 Bd4 101.Bc6 Be5 102.Bd7+ Ke4 103.Kg5
-+ (-2.36) Depth: 10/16 00:00:00 27kN
98.Kxh4 Kf5 99.Kh5 Be7 100.h4 Bc5 101.Bc6 Bd4 102.Bd5 Ke5 103.Bc4
-+ (-2.41) Depth: 11/17 00:00:00 68kN
98.Kxh4 Kf5 99.Bc6 Be7+ 100.Kh5 Bc5 101.Bd7+ Ke5 102.Bc6 Kd6 103.Be4 Ke6 104.Kg4 Ke5 105.Bd3
-+ (-2.34) Depth: 12/20 00:00:00 142kN
98.Kxh4 Kf5 99.Bc6 Be7+ 100.Kh5 Bc5 101.Bd7+ Ke5 102.Bc6 Kd6 103.Be4 Bd4 104.Kg4 Ke5 105.Bc6 Bc5
-+ (-2.37) Depth: 13/20 00:00:01 219kN
98.Kxh4 Kf5 99.Bc6 Be1 100.Kh5 Bc3 101.Bd7+ Ke5 102.Bc6 Kd6 103.Be4 Bd4 104.Kg4 Ke5 105.Kf3
-+ (-2.28) Depth: 14/22 00:00:01 420kN
98.Kxh4 Kf5 99.Bc6 Be1 100.Kh5 Bc3 101.Bd7+ Ke5 102.Bc6 Kd6 103.Bf3 Bd4 104.Kg5 Ke5 105.h4 Bc5
-+ (-2.28) Depth: 15/25 00:00:02 732kN
98.Kxh4 Kf5 99.Bc6 Bc5 100.Kh5 Bd6 101.Kh4 Be7+ 102.Kh5 Bf6 103.Bd7+ Ke5 104.Bc6 Kd6 105.Be4 Bd4 106.Kg4 Ke5 107.Kf3
-+ (-2.28) Depth: 16/25 00:00:03 1049kN
98.Kxh4 Kf5 99.Bc6 Bc5 100.Bg2 Be7+ 101.Kh5 Bf6 102.Bd5 Bd4 103.Bc6 Be3 104.Bf3 Ke5 105.Kg4 Bd4 106.Kg5 Bc5 107.h4
-+ (-2.28) Depth: 17/26 00:00:05 1738kN
98.Kxh4 Kf5 99.Bc6 Bc5 100.Bg2 Be7+ 101.Kh5 Bf6 102.Bd5 Bd4 103.Bc6 Bc5 104.Kh4 Be3 105.Kh5 Bb6 106.h4 Bd4 107.Bd5 Ke5
-+ (-2.28) Depth: 18/27 00:00:07 2340kN
98.Kxh4 Kf5 99.Bc6 Bc5 100.Bg2 Be7+ 101.Kh5 Bf6 102.Bd5 Bd4 103.Bc6 Bc5 104.Bf3 Bb4 105.Kh4 Be7+ 106.Kh5 Bd6 107.Kh4 Ke5 108.Kg5 Bc5 109.h4
-+ (-2.28) Depth: 19/29 00:00:09 3140kN
98.Kxh4 Kf5 99.Bc6 Bc5 100.Bg2 Be7+ 101.Kh5 Bf6 102.Bd5 Bd4 103.Bc6 Bc5 104.Kh4 Be7+ 105.Kh5 Bd8 106.Bf3 Ba5 107.Bc6 Bb6 108.h4 Bd4 109.Bd5 Ke5
-+ (-2.28) Depth: 20/30 00:00:11 4011kN
98.Kxh4 Kf5 99.Bc6 Bc5 100.Bg2 Be7+ 101.Kh5 Bf6 102.Bd5 Bd4 103.Bc6 Bc5 104.Kh4 Be7+ 105.Kh5 Ba3 106.Bf3 Bb4 107.Kh4 Be7+ 108.Kh5 Bd6 109.Kh4 Ke5 110.Kg5 Bc5 111.h4
-+ (-2.28) Depth: 21/30 00:00:15 5534kN
98.Kxh4 Kd4 99.Kg4 Ke3 100.h4 Bf8 101.Bg2 Bh6 102.Bc6 Kf2 103.Kh5 g2 104.Kxh6 f3 105.Be8 g1Q 106.Kh7 Qb1+ 107.Kg7 Kg3 108.Bg6 Qb2+ 109.Kf7 Kxh4
-+ (-8.12) Depth: 22/40 00:01:08 30104kN
98.Bc6 Be7 99.Bf3 Bd6 100.Bc6 Kd4 101.Kf3 Kc3 102.Kg4 Bc7 103.Bb7 Kd4 104.Kf3 Kc4 105.Bc6 Be5 106.Kg4 Kd4 107.Bf3 Ke3 108.Bd5 Bf6
-+ (-3.10) Depth: 22/40 00:01:29 39655kN
98.Bc6 Be7 99.Ba8 Bd6 100.Bc6 Kd4 101.Kf3 Kc3 102.Kg4 Bc7 103.Bb7 Kc4 104.Kf3 Bd6 105.Ba6+ Kc5 106.Bb7 Be5 107.Kg4 Kd4 108.Bc6 Ke3 109.Bd5 Bf6
-+ (-3.10) Depth: 23/40 00:01:51 47365kN
98.Bc6 Be7 99.Ba8 Bd6 100.Bc6 Kd4 101.Kf3 Kc3 102.Kg4 Bc7 103.Bb7 Kd4 104.Kf3 Be5 105.Ba8 Bb8 106.Bc6 Bd6 107.Ba8 Be5 108.Kg4 Ke3 109.Bd5 Bf6
-+ (-3.10) Depth: 24/40 00:02:01 50451kN
98.Bc6 Be7 99.Ba8 Bd6 100.Bc6 Kd4 101.Kf3 Kc3 102.Kg4 Bc7 103.Bb7 Kd4 104.Kf3 Be5 105.Ba8 Kc5 106.Bb7 Kc4 107.Kg4 Kd3 108.Bd5 Ke3 109.Bc6 Bf6 110.Bd5
-+ (-3.10) Depth: 25/40 00:02:25 58464kN
98.Bc6 Be7 99.Ba8 Bd6 100.Bc6 Kd4 101.Kf3 Kc3 102.Kg4 Bc7 103.Bb7 Kd4 104.Kf3 Be5 105.Bc6 Kc5 106.Bb7 Kc4 107.Kg4 Kc3 108.Bc6 Kd3 109.Bf3 Ke3 110.Bd5 Bf6
-+ (-3.10) Depth: 26/40 00:03:01 71118kN
98.Bc6 Be7 99.Ba8 Bd6 100.Bc6 Kd4 101.Kf3 Kc3 102.Kg4 Bc7 103.Bb7 Kd4 104.Kf3 Ke5 105.Kg4 Bb8 106.Kf3 Ba7 107.Ba8 Bd4 108.Bc6 Kf5 109.Be4+ Kg5 110.Ke2 Bc5 111.Kf3 Bb4
-+ (-3.10) Depth: 27/40 00:03:44 87295kN
98.Bc6 Be7 99.Ba8 Bd6 100.Bc6 Kd4 101.Kf3 Kc3 102.Kg4 Bc7 103.Bb7 Kd4 104.Kf3 Ke5 105.Kg4 Bb8 106.Kf3 Bd6 107.Bc6 Kd4 108.Ke2 Bc7 109.Kf3 Be5 110.Kg4 Ke3 111.Bd5 Bf6
-+ (-3.10) Depth: 28/45 00:06:45 166173kN
98.Bc6 Be7 99.Ba8 Bd6 100.Bc6 Kd4 101.Kf3 Kc3 102.Kg4 Bc7 103.Kf3 Kd3 104.Ba8 Bd6 105.Bb7 Be5 106.Ba8 Kd4 107.Ke2 Kc4 108.Bc6 Bb8 109.Bg2 Kd4 110.Bb7 Ba7 111.Kf3 Ke5 112.Be4
-+ (-3.10) Depth: 29/49 00:10:21 260793kN
(Blass, Tel-Aviv 25.06.2004)
Uri

by **Joachim Rang** » 25 Jun 2004, 11:07

Geschrieben von:/Posted by: Joachim Rang at 25 June 2004 12:07:37:
Als Antwort auf:/In reply to: Re: Fruit 1.5 tralala geschrieben von:/posted by: Uri Blass at 25 June 2004 11:12:11:

thanks for this fine testposition. Actually I think it is not so good for tuning passer boni in general but for testing knowledge about bishop of opposite colour endings. Here one can test whether an engine has a good understanding of blocked and therefore useless passers or not.
Fruit X solves this at depth 19, 3 plies earlier in about 20 seconds which is already a bit better.
Although big passer boni are a speculative eval term and sometimes wrong I think they improve the overall playing strength since the emphasize the importance of passers in the endgame. A more detailed look which passers are dangerous and which not is better but one has to find abstract algorithms which justify the slow down they cause.
Currently Fruit gives a big bonus for passers which are free to advance, i.e. no square in front of them is occupied or attacked by an enemy piece. Fruit also increases and decreaes the value of a passer as a function of defender and attacker king distance to the square in front of the passer. What Fruit currently lacks is a definition of useless passers, for example such as in your testposition which are safely blocked and can't be advanced.
Overall I agree with Chritopher Theron who said a speculative passer eval has increased the strength of Tiger also he knows that in certain position the eval is totally crazy and wrong.
regards Joachim

by **Uri Blass** » 25 Jun 2004, 11:56

Geschrieben von:/Posted by: Uri Blass at 25 June 2004 12:56:08:
Als Antwort auf:/In reply to: Re: Fruit 1.5 tralala geschrieben von:/posted by: Joachim Rang at 25 June 2004 12:07:37:

As far as I know people suggested Dieter to increase the value of passed pawn and he found it to be counter productive.
I do some changes now that are hopefully productive.
Change 1 was to give a panelty also for isolated pawns that are also passed
pawns(I have no time to test to be sure if the change is productive but the nunn2 match between new and older version ended with 27.5-22.5 for the new version(1 minute/40 moves))
Change 2 that I will now try is to give a bigger bonus for passed pawns based on the stage of the game.
I am going to be careful and not change the bonus by more than 0.3 pawns(I will double the bonus in case it is smaller than 0.3 pawns).
Change 3 that I am going to test is probably some evaluation of backward pawns.
Uri

by **Uri Blass** » 25 Jun 2004, 13:59

Geschrieben von:/Posted by: Uri Blass at 25 June 2004 14:59:59:
Als Antwort auf:/In reply to: Re: Fruit 1.5 tralala geschrieben von:/posted by: Uri Blass at 25 June 2004 12:56:08:

I stopped the match after result of 12.5-7.5 against the version that gives a bigger bonus for passed pawns in the endgame.
looking at my code I already have a bonus in the endgame for advanced passed pawn(at least 6th rank) that threat to advance and maybe another bonus is too much.
Another explanation is that the score include score of passed pawn relative to the king and passed pawn that is supported by its king can get a big bonus
even in case that it is not advanced and increasing that bonus is not a good idea.
Uri

by **Dieter Bürßner** » 25 Jun 2004, 18:57

Geschrieben von:/Posted by: Dieter Bürßner at 25 June 2004 19:57:36:
Als Antwort auf:/In reply to: Re: Fruit 1.5 tralala geschrieben von:/posted by: Uri Blass at 25 June 2004 04:11:14:

Perhaps I am wrong. I had the impression, that a rook gets worth relatively more in the endgame. Say N (+perhaps 2 pawns) vs rook. And more pawns on both sides. I thought the N was chanceless in the endgame typically. While in the middle game the N might be not that much worse. My impression was, that with less pawns, more open lines, ... the rook gets worth more (and B or B-pair too).
Of course, there are cases were a rook vs. pawns is rather helpless (were for example a B could stop the pawns).
Regards,
Dieter

by **Uri Blass** » 25 Jun 2004, 19:17

Geschrieben von:/Posted by: Uri Blass at 25 June 2004 20:17:03:
Als Antwort auf:/In reply to: Re: Fruit 1.5 tralala geschrieben von:/posted by: Dieter Bürßner at 25 June 2004 19:57:36:

rook relative to knight and rook relative to pawn are different questions.
The table talks about rook relative to pawn.
Uri

by **Dieter Bürßner** » 25 Jun 2004, 19:26

Geschrieben von:/Posted by: Dieter Bürßner at 25 June 2004 20:26:52:
Als Antwort auf:/In reply to: Re: Fruit 1.5 tralala geschrieben von:/posted by: Uri Blass at 25 June 2004 20:17:03:

Not only.
Pawn = 100
Pawn (Endgame) = 100
Knight = 350
Knight (Endgame) = 320
...
Rook = 550
Rook (Endgame) = 520
pawn stays the same. R+N goes down by the same ammount. So R does not get stronger compared to N in the endgame. Which of course might be best as it is. I just would have expected it differently.
Regards,
Dieter

by **Uri Blass** » 25 Jun 2004, 20:57

Geschrieben von:/Posted by: Uri Blass at 25 June 2004 21:57:47:
Als Antwort auf:/In reply to: Re: Fruit 1.5 tralala geschrieben von:/posted by: Dieter Bürßner at 25 June 2004 20:26:52:

This conclusion is wrong because R get weaker relative to pawn in the endgame.
550/350

by **Dieter Bürßner** » 25 Jun 2004, 21:23

Geschrieben von:/Posted by: Dieter Bürßner at 25 June 2004 22:23:29:
Als Antwort auf:/In reply to: Re: Fruit 1.5 tralala geschrieben von:/posted by: Uri Blass at 25 June 2004 21:57:47:

Winboard Forum

Fruit 1.5 tralala

Fruit 1.5 tralala

Re: Fruit 1.5 tralala

Re: Fruit 1.5 tralala

Re: Fruit 1.5 tralala

Re: Fruit 1.5 tralala

OK

Re: Fruit 1.5 tralala

Re: Fruit 1.5 tralala

Re: Fruit 1.5 tralala

Re: Fruit 1.5 tralala

Re: Fruit 1.5 tralala

Re: Fruit 1.5 tralala

Re: Fruit 1.5 tralala

Re: Fruit 1.5 tralala

Re: Fruit 1.5 tralala

Re: Fruit 1.5 tralala

Re: Fruit 1.5 tralala

Re: Fruit 1.5 tralala

Re: Fruit 1.5 tralala

Who is online