now the impact--gantlet Fruit 1.5, 600 games (Blitz)

[Heinz van Kempen]

Geschrieben von:/Posted by: Heinz van Kempen at 15 June 2004 11:38:48:

Hi ,
"a bombshell is coming". I do not know if you remember the words from Dann Corbit when announcing the first Fruit beta. After the first tests I thought "okay, not really a bombshell, but a fine and solid engine, running smoothly in UCI mode". I did not understand that it was a prophecy.
Here are the results from a 600 games Nunn gauntlet over all 20 positions on Athlon 2600+, 4 min. + 2 sec.:



Fruit 1.5   - Jonny 2.63b             22.5 - 17.5
Fruit 1.5   - Slow Chess 2.93         21.5 - 18.5
Fruit 1.5   - King of Kings 2.56      25.5 - 14.5
Fruit 1.5   - Patriot 0.172-Light     19.0 - 21.0
Fruit 1.5   - Movei 00_8_232          27.0 - 13.0
Fruit 1.5   - Pharaon 2.62            25.0 - 15.0
Fruit 1.5   - Abrok 5.0               24.5 - 15.5
Fruit 1.5   - Dragon 4.5 CF           24.0 - 16.0
Fruit 1.5   - Snitch 0.3.66           27.5 - 12.5
Fruit 1.5   - Gnuchess 5.07           34.0 - 6.0
Fruit 1.5   - Ufim 5.00               23.5 - 16.5
Fruit 1.5   - Eeyore 1.37             34.5 - 5.5
Fruit 1.5   - Chiron 0.3.8            34.0 - 6.0
Fruit 1.5   - PostModernist 1010a     29.0 - 11.0
Fruit 1.5   - Rybka 1.6.0.1x          33.5 - 6.5




So Fruit only lost one single match by a very narrow margin against Patriot free version, what is no shame at all. To explain the results a bit, I have to say that I took engines from all my recent tournaments for better rating calculation for all. Ufim is very much improved (123 ELO points stronger in my rating list at this moment), Slow Chess, King of Kings, Jonny and PostModernist also improved (see Nunn Top 2). Snitch and Rybka are still betas and doing well in Nunn G. There strength at this stage is close to Resp or Chezzz. The latest Chiron and Eeyore are also not bad.
Rating comparison for Fruit:
Fruit 1.0 2494
Fruit 1.5 2656 + 162
When Fruit 1.0 was released? I think an improvement like this in only a few months at already such a high level is rather unique. The beta Fruit X 05/09 roughly one month ago is rated 2550.
Fabien wrote something about small speed improvements and better time management that he added in only one day and first thought it would be only "cosmetic". He surely can explain this shameless improvement better.
Fruit 1.5 gauntlet download at:
http://www.husvankempen.de/nunn/
Best Regards
Heinz

[Uri Blass]

Geschrieben von:/Posted by: Uri Blass at 15 June 2004 14:11:58:
Als Antwort auf:/In reply to: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Heinz van Kempen at 15 June 2004 11:38:48:

Hi ,
"a bombshell is coming". I do not know if you remember the words from Dann Corbit when announcing the first Fruit beta. After the first tests I thought "okay, not really a bombshell, but a fine and solid engine, running smoothly in UCI mode". I did not understand that it was a prophecy.
Here are the results from a 600 games Nunn gauntlet over all 20 positions on Athlon 2600+, 4 min. + 2 sec.:




So Fruit only lost one single match by a very narrow margin against Patriot free version, what is no shame at all. To explain the results a bit, I have to say that I took engines from all my recent tournaments for better rating calculation for all. Ufim is very much improved (123 ELO points stronger in my rating list at this moment), Slow Chess, King of Kings, Jonny and PostModernist also improved (see Nunn Top 2). Snitch and Rybka are still betas and doing well in Nunn G. There strength at this stage is close to Resp or Chezzz. The latest Chiron and Eeyore are also not bad.
Rating comparison for Fruit:
Fruit 1.0 2494
Fruit 1.5 2656 + 162
When Fruit 1.0 was released? I think an improvement like this in only a few months at already such a high level is rather unique. The beta Fruit X 05/09 roughly one month ago is rated 2550.
Fabien wrote something about small speed improvements and better time management that he added in only one day and first thought it would be only "cosmetic". He surely can explain this shameless improvement better.
Fruit 1.5 gauntlet download at:
http://www.husvankempen.de/nunn/
Best Regards
Heinz

>
>Fruit 1.5   - Jonny 2.63b             22.5 - 17.5
>Fruit 1.5   - Slow Chess 2.93         21.5 - 18.5
>Fruit 1.5   - King of Kings 2.56      25.5 - 14.5
>Fruit 1.5   - Patriot 0.172-Light     19.0 - 21.0
>Fruit 1.5   - Movei 00_8_232          27.0 - 13.0
>Fruit 1.5   - Pharaon 2.62            25.0 - 15.0
>Fruit 1.5   - Abrok 5.0               24.5 - 15.5
>Fruit 1.5   - Dragon 4.5 CF           24.0 - 16.0
>Fruit 1.5   - Snitch 0.3.66           27.5 - 12.5
>Fruit 1.5   - Gnuchess 5.07           34.0 - 6.0
>Fruit 1.5   - Ufim 5.00               23.5 - 16.5
>Fruit 1.5   - Eeyore 1.37             34.5 - 5.5
>Fruit 1.5   - Chiron 0.3.8            34.0 - 6.0
>Fruit 1.5   - PostModernist 1010a     29.0 - 11.0
>Fruit 1.5   - Rybka 1.6.0.1x          33.5 - 6.5
>
>
I cannot believe that 100 elo point improvement is only thanks to small speed improvements and better time management.
I believe that 100 elo improvement is equivalent to being more than twice faster
but I may be wrong for the nunn position.
It may be interesting if you also test with unequal time control to find out rating for some programs when they get twice the time of the opponent(it is possible to do it under Fritz).
It means that they get 10+6 time control when the opponent gets 5+3 time control(you can use Fruit X05/09 with 10+6 time control against the same oponnents to find if it is better or worse than Fruit1.5.
Uri

[Uri Blass]

Geschrieben von:/Posted by: Uri Blass at 15 June 2004 14:14:24:
Als Antwort auf:/In reply to: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Heinz van Kempen at 15 June 2004 11:38:48:

Hi ,
"a bombshell is coming". I do not know if you remember the words from Dann Corbit when announcing the first Fruit beta. After the first tests I thought "okay, not really a bombshell, but a fine and solid engine, running smoothly in UCI mode". I did not understand that it was a prophecy.
Here are the results from a 600 games Nunn gauntlet over all 20 positions on Athlon 2600+, 4 min. + 2 sec.:




So Fruit only lost one single match by a very narrow margin against Patriot free version, what is no shame at all. To explain the results a bit, I have to say that I took engines from all my recent tournaments for better rating calculation for all. Ufim is very much improved (123 ELO points stronger in my rating list at this moment), Slow Chess, King of Kings, Jonny and PostModernist also improved (see Nunn Top 2). Snitch and Rybka are still betas and doing well in Nunn G. There strength at this stage is close to Resp or Chezzz. The latest Chiron and Eeyore are also not bad.
Rating comparison for Fruit:
Fruit 1.0 2494
Fruit 1.5 2656 + 162
When Fruit 1.0 was released? I think an improvement like this in only a few months at already such a high level is rather unique. The beta Fruit X 05/09 roughly one month ago is rated 2550.
Fabien wrote something about small speed improvements and better time management that he added in only one day and first thought it would be only "cosmetic". He surely can explain this shameless improvement better.
Fruit 1.5 gauntlet download at:
http://www.husvankempen.de/nunn/
Best Regards
Heinz

>
>Fruit 1.5   - Jonny 2.63b             22.5 - 17.5
>Fruit 1.5   - Slow Chess 2.93         21.5 - 18.5
>Fruit 1.5   - King of Kings 2.56      25.5 - 14.5
>Fruit 1.5   - Patriot 0.172-Light     19.0 - 21.0
>Fruit 1.5   - Movei 00_8_232          27.0 - 13.0
>Fruit 1.5   - Pharaon 2.62            25.0 - 15.0
>Fruit 1.5   - Abrok 5.0               24.5 - 15.5
>Fruit 1.5   - Dragon 4.5 CF           24.0 - 16.0
>Fruit 1.5   - Snitch 0.3.66           27.5 - 12.5
>Fruit 1.5   - Gnuchess 5.07           34.0 - 6.0
>Fruit 1.5   - Ufim 5.00               23.5 - 16.5
>Fruit 1.5   - Eeyore 1.37             34.5 - 5.5
>Fruit 1.5   - Chiron 0.3.8            34.0 - 6.0
>Fruit 1.5   - PostModernist 1010a     29.0 - 11.0
>Fruit 1.5   - Rybka 1.6.0.1x          33.5 - 6.5
>
>
correction I see that you used 4+2 time control so you can test 8+4 against 4+2
I thought for some reasons that you used 5+3.
Uri

[Bryan Hofmann]

Geschrieben von:/Posted by: Bryan Hofmann at 15 June 2004 14:28:03:
Als Antwort auf:/In reply to: Re: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Uri Blass at 15 June 2004 14:11:58:

Hi ,
"a bombshell is coming". I do not know if you remember the words from Dann Corbit when announcing the first Fruit beta. After the first tests I thought "okay, not really a bombshell, but a fine and solid engine, running smoothly in UCI mode". I did not understand that it was a prophecy.
Here are the results from a 600 games Nunn gauntlet over all 20 positions on Athlon 2600+, 4 min. + 2 sec.:




So Fruit only lost one single match by a very narrow margin against Patriot free version, what is no shame at all. To explain the results a bit, I have to say that I took engines from all my recent tournaments for better rating calculation for all. Ufim is very much improved (123 ELO points stronger in my rating list at this moment), Slow Chess, King of Kings, Jonny and PostModernist also improved (see Nunn Top 2). Snitch and Rybka are still betas and doing well in Nunn G. There strength at this stage is close to Resp or Chezzz. The latest Chiron and Eeyore are also not bad.
Rating comparison for Fruit:
Fruit 1.0 2494
Fruit 1.5 2656 + 162
When Fruit 1.0 was released? I think an improvement like this in only a few months at already such a high level is rather unique. The beta Fruit X 05/09 roughly one month ago is rated 2550.
Fabien wrote something about small speed improvements and better time management that he added in only one day and first thought it would be only "cosmetic". He surely can explain this shameless improvement better.
Fruit 1.5 gauntlet download at:
http://www.husvankempen.de/nunn/
Best Regards
Heinz

I cannot believe that 100 elo point improvement is only thanks to small speed improvements and better time management.
I believe that 100 elo improvement is equivalent to being more than twice faster
but I may be wrong for the nunn position.
It may be interesting if you also test with unequal time control to find out rating for some programs when they get twice the time of the opponent(it is possible to do it under Fritz).
It means that they get 10+6 time control when the opponent gets 5+3 time control(you can use Fruit X05/09 with 10+6 time control against the same oponnents to find if it is better or worse than Fruit1.5.
Uri

>>
>>Fruit 1.5   - Jonny 2.63b             22.5 - 17.5
>>Fruit 1.5   - Slow Chess 2.93         21.5 - 18.5
>>Fruit 1.5   - King of Kings 2.56      25.5 - 14.5
>>Fruit 1.5   - Patriot 0.172-Light     19.0 - 21.0
>>Fruit 1.5   - Movei 00_8_232          27.0 - 13.0
>>Fruit 1.5   - Pharaon 2.62            25.0 - 15.0
>>Fruit 1.5   - Abrok 5.0               24.5 - 15.5
>>Fruit 1.5   - Dragon 4.5 CF           24.0 - 16.0
>>Fruit 1.5   - Snitch 0.3.66           27.5 - 12.5
>>Fruit 1.5   - Gnuchess 5.07           34.0 - 6.0
>>Fruit 1.5   - Ufim 5.00               23.5 - 16.5
>>Fruit 1.5   - Eeyore 1.37             34.5 - 5.5
>>Fruit 1.5   - Chiron 0.3.8            34.0 - 6.0
>>Fruit 1.5   - PostModernist 1010a     29.0 - 11.0
>>Fruit 1.5   - Rybka 1.6.0.1x          33.5 - 6.5
>>
>>

All the info you need to find out is here http://wbec-ridderkerk.nl/html/details/Fruit.html . You can even look at the source code to see how the program was improved.

[Fabien Letouzey]

Geschrieben von:/Posted by: Fabien Letouzey at 15 June 2004 14:41:52:
Als Antwort auf:/In reply to: Re: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Uri Blass at 15 June 2004 14:11:58:


Hi Uri,

I cannot believe that 100 elo point improvement is only thanks to
small speed improvements and better time management.

You might be right, but I think the other changes (e.g. restricting
the definition of backward pawns) only had a minor effect.
I could produce a development version with time-management options, so
we can evaluate the impact of this change alone (of course not on
Heinz's computers). I might want to do it anyway because if the
improvement is large, I will consider additional heuristics.
I can also dig through my change log files to see if I forgot about a
potentially important change.
Joachim Rang probably knows more than me about this, but he will come
back from a journey only on Thursday ...
Fabien.

[Uri Blass]

Geschrieben von:/Posted by: Uri Blass at 15 June 2004 14:52:47:
Als Antwort auf:/In reply to: Re: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Bryan Hofmann at 15 June 2004 14:28:03:

Hi ,
"a bombshell is coming". I do not know if you remember the words from Dann Corbit when announcing the first Fruit beta. After the first tests I thought "okay, not really a bombshell, but a fine and solid engine, running smoothly in UCI mode". I did not understand that it was a prophecy.
Here are the results from a 600 games Nunn gauntlet over all 20 positions on Athlon 2600+, 4 min. + 2 sec.:




So Fruit only lost one single match by a very narrow margin against Patriot free version, what is no shame at all. To explain the results a bit, I have to say that I took engines from all my recent tournaments for better rating calculation for all. Ufim is very much improved (123 ELO points stronger in my rating list at this moment), Slow Chess, King of Kings, Jonny and PostModernist also improved (see Nunn Top 2). Snitch and Rybka are still betas and doing well in Nunn G. There strength at this stage is close to Resp or Chezzz. The latest Chiron and Eeyore are also not bad.
Rating comparison for Fruit:
Fruit 1.0 2494
Fruit 1.5 2656 + 162
When Fruit 1.0 was released? I think an improvement like this in only a few months at already such a high level is rather unique. The beta Fruit X 05/09 roughly one month ago is rated 2550.
Fabien wrote something about small speed improvements and better time management that he added in only one day and first thought it would be only "cosmetic". He surely can explain this shameless improvement better.
Fruit 1.5 gauntlet download at:
http://www.husvankempen.de/nunn/
Best Regards
Heinz

I cannot believe that 100 elo point improvement is only thanks to small speed improvements and better time management.

All the info you need to find out is here http://wbec-ridderkerk.nl/html/details/Fruit.html . You can even look at the source code to see how the program was improved.

>>>
>>>Fruit 1.5   - Jonny 2.63b             22.5 - 17.5
>>>Fruit 1.5   - Slow Chess 2.93         21.5 - 18.5
>>>Fruit 1.5   - King of Kings 2.56      25.5 - 14.5
>>>Fruit 1.5   - Patriot 0.172-Light     19.0 - 21.0
>>>Fruit 1.5   - Movei 00_8_232          27.0 - 13.0
>>>Fruit 1.5   - Pharaon 2.62            25.0 - 15.0
>>>Fruit 1.5   - Abrok 5.0               24.5 - 15.5
>>>Fruit 1.5   - Dragon 4.5 CF           24.0 - 16.0
>>>Fruit 1.5   - Snitch 0.3.66           27.5 - 12.5
>>>Fruit 1.5   - Gnuchess 5.07           34.0 - 6.0
>>>Fruit 1.5   - Ufim 5.00               23.5 - 16.5
>>>Fruit 1.5   - Eeyore 1.37             34.5 - 5.5
>>>Fruit 1.5   - Chiron 0.3.8            34.0 - 6.0
>>>Fruit 1.5   - PostModernist 1010a     29.0 - 11.0
>>>Fruit 1.5   - Rybka 1.6.0.1x          33.5 - 6.5
>>>
>>>
Fruit1.0 was improved by better evaluation and it is not news but the main surprise is the improvement from Fruit X05/09 to Fruit1.5
I got the impression from fabian's words that the improvement from
Fruit X05/09 to Fruit1.5 is only thanks to better time management and better compiler.
I understand that fruit did not know to use more time in case of fail low in X05/09 and maybe there were more improvement in time management(I did not read the relevant part of fruit source to find out) but I did not think that it should give more than 30 elo improvement.
together with better compiler 40 elo improvement make sense but 100 elo seems too good to be correct.
It is possible that I underestimated the time management improvement and the meaning of it is the same of being twice faster.
Uri

[Heinz van Kempen]

Geschrieben von:/Posted by: Heinz van Kempen at 15 June 2004 15:00:15:
Als Antwort auf:/In reply to: Re: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Uri Blass at 15 June 2004 14:11:58:

I cannot believe that 100 elo point improvement is only thanks to small speed improvements and better time management.
I believe that 100 elo improvement is equivalent to being more than twice faster
but I may be wrong for the nunn position.
It may be interesting if you also test with unequal time control to find out rating for some programs when they get twice the time of the opponent(it is possible to do it under Fritz).
It means that they get 10+6 time control when the opponent gets 5+3 time control(you can use Fruit X05/09 with 10+6 time control against the same oponnents to find if it is better or worse than Fruit1.5.
Uri

Hello Uri,
you can be sure that I would run all kinds of tests if I had more CPU time.
After finishing Nunn Top 2 in a few days there will be a few tournaments for the weaker engines and also a knockout tournament with 64 best engines and a time control 40 minutes plus 20 seconds each move, where Nunn positions are chosen accidentally. A short qualify will be run beforehand.
I will post the pairings for the qualify in another posting.
Best Regards
Heinz

[Heinz van Kempen]

Geschrieben von:/Posted by: Heinz van Kempen at 15 June 2004 15:10:07:
Als Antwort auf:/In reply to: Re: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Uri Blass at 15 June 2004 14:52:47:

Fruit1.0 was improved by better evaluation and it is not news but the main surprise is the improvement from Fruit X05/09 to Fruit1.5
I got the impression from fabian's words that the improvement from
Fruit X05/09 to Fruit1.5 is only thanks to better time management and better compiler.
I understand that fruit did not know to use more time in case of fail low in X05/09 and maybe there were more improvement in time management(I did not read the relevant part of fruit source to find out) but I did not think that it should give more than 30 elo improvement.
together with better compiler 40 elo improvement make sense but 100 elo seems too good to be correct.
It is possible that I underestimated the time management improvement and the meaning of it is the same of being twice faster.
Uri

Hello Uri,
as you might have noticed other results posted underline such an improvement like the results from UEL or those from Torsten Schoop only two days ago with much more time. These are really impressive.
Best Regards
Heinz

[Fabien Letouzey]

Geschrieben von:/Posted by: Fabien Letouzey at 15 June 2004 15:32:43:
Als Antwort auf:/In reply to: Re: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Uri Blass at 15 June 2004 14:52:47:


Hi,

Fruit1.0 was improved by better evaluation and it is not news but the
main surprise is the improvement from Fruit X05/09 to Fruit1.5
I got the impression from fabian's words that the improvement from
Fruit X05/09 to Fruit1.5 is only thanks to better time management and
better compiler.
I understand that fruit did not know to use more time in case of fail
low in X05/09 and maybe there were more improvement in time
management(I did not read the relevant part of fruit source to find
out) but I did not think that it should give more than 30 elo
improvement.
together with better compiler 40 elo improvement make sense but 100
elo seems too good to be correct.
It is possible that I underestimated the time management improvement
and the meaning of it is the same of being twice faster.
Uri

There was an intermediate version Fruit X 05/21 that played in some
tournaments (e.g. WBEC division qualifications).
Fruit X 05/21 and Fruit 1.5 are very similar in strength.
As far as I can remember, the only differences between Fruit X 05/09
and Fruit X 05/21 were:
1) time management
2) 20% speed improvement (on Linux)
I will check in my log files.
I mentionned the compiler as a way of thanking Slater Wold. I
obtained some speed improvement by changing low-level data structures.
Only Windows users can measure the actual difference in NPS. Joachim
Rang has the development executables and Dann probably does as well.
We can measure that if you want. You can't compare with Fruit 1.0
though because mobility slowed it down quite a lot.
I don't use aspiration search so I have no fail low. But Fruit 1.0
always stops the search when the time limit is reached, even if it
could "know" (with appropriate code) that the move is very bad.
Bear in mind that Fruit 1.0 has a hard time limit (no time management
by my standards). Fixing this is part of what I call "time management"
Fabien.

[Uri Blass]

Geschrieben von:/Posted by: Uri Blass at 15 June 2004 15:35:43:
Als Antwort auf:/In reply to: Re: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Fabien Letouzey at 15 June 2004 14:41:52:

Hi Uri,

I cannot believe that 100 elo point improvement is only thanks to
small speed improvements and better time management.

You might be right, but I think the other changes (e.g. restricting
the definition of backward pawns) only had a minor effect.
I could produce a development version with time-management options, so
we can evaluate the impact of this change alone (of course not on
Heinz's computers). I might want to do it anyway because if the
improvement is large, I will consider additional heuristics.
I can also dig through my change log files to see if I forgot about a
potentially important change.
Joachim Rang probably knows more than me about this, but he will come
back from a journey only on Thursday ...
Fabien.

What is your definition of backward pawns?
I still do not have backward pawns evaluation in movei and it is one of the things that I need to add.
I also have no evaluation for weak pawns.

I guess that my pawn evaluation is relatively stupid.
Here is short descreption
In all stages of the game:
1)static panelty of 0.1 pawn for every isolated pawn that is not passed pawn
2)static panelty of 0.2 pawns for every double pawn(2 pawns in the same file get panelty of 0.2 and 3 pawns on the same file a panelty of 0.4 pawns)
3)bonus for passed pawns based on the rank of the pawn
I use values 4 8 16 20 30 30 as basic values(100 is an estimate for the value of a pawn and I have piece square table for that).
4)These values are multiplied by 2 if the passed pawn is protected by a pawn and they are divided by 2 if the passed pawn is blocked by an enemy piece.
5)Additional bonus for advanced passed pawns if the opponent does not control the square that they want to go
6)evaluating the passed pawn relative to the kings.
In endgames:
1)evaluate pawns relative to the kings(for example if all the pawns are on files a-b-c and one of the king is in file h than the king get a panelty based on the distance between file c and file h that is 5).

2)panelty for king that it's rank is behind the most advanced passed pawn(when the panelty is bigger in case of advanced passed pawn).
I think that it may be better if I replace it by panelty for king that is not in the square of the pawn.
My evaluation does not consider side to move there and I guess that it also should be improved.
3)bonus of 0.1 pawns for having passed pawn in the biggest or smallest file.
(If there are no pawns in file a then white can get a bonus for having passed pawns in file b).
In pawn endgames:some evaluation to detect unstoppable passed pawns in part of the cases to give them a big bonus.
Uri

[Fabien Letouzey]

Geschrieben von:/Posted by: Fabien Letouzey at 15 June 2004 15:45:41:
Als Antwort auf:/In reply to: Re: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Uri Blass at 15 June 2004 15:35:43:

What is your definition of backward pawns?

In Fruit a pawn is backward if:
1) it is not isolated
2) friendly neighbour pawns are more advanced
3) it cannot get to the same rank as one of them in one move (e.g. the square is controlled by an opponent pawn)
Fruit 1.0 only used 1) and 2) and that visibly resulted in stupid moves in many games.
Other than that pawn evaluation was not modified.
Thanks for sharing your work on pawn evaluation. Working on it is now
one of my priorities now (pawn-structure knowledge is necessary for
the quiet playing style I am aiming at) and I think we will end up
having similar features.
Fabien.

[Tord Romstad]

Geschrieben von:/Posted by: Tord Romstad at 15 June 2004 16:08:26:
Als Antwort auf:/In reply to: Re: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Uri Blass at 15 June 2004 15:35:43:

What is your definition of backward pawns?
I guess that my pawn evaluation is relatively stupid.
Here is short descreption
In all stages of the game:
1)static panelty of 0.1 pawn for every isolated pawn that is not passed pawn
2)static panelty of 0.2 pawns for every double pawn(2 pawns in the same file get panelty of 0.2 and 3 pawns on the same file a panelty of 0.4 pawns)
3)bonus for passed pawns based on the rank of the pawn
I use values 4 8 16 20 30 30 as basic values(100 is an estimate for the value of a pawn and I have piece square table for that).
4)These values are multiplied by 2 if the passed pawn is protected by a pawn and they are divided by 2 if the passed pawn is blocked by an enemy piece.
5)Additional bonus for advanced passed pawns if the opponent does not control the square that they want to go
6)evaluating the passed pawn relative to the kings.
In endgames:
1)evaluate pawns relative to the kings(for example if all the pawns are on files a-b-c and one of the king is in file h than the king get a panelty based on the distance between file c and file h that is 5).

2)panelty for king that it's rank is behind the most advanced passed pawn(when the panelty is bigger in case of advanced passed pawn).
I think that it may be better if I replace it by panelty for king that is not in the square of the pawn.
My evaluation does not consider side to move there and I guess that it also should be improved.
3)bonus of 0.1 pawns for having passed pawn in the biggest or smallest file.
(If there are no pawns in file a then white can get a bonus for having passed pawns in file b).
In pawn endgames:some evaluation to detect unstoppable passed pawns in part of the cases to give them a big bonus.

My definition is identical to Fabien's.
I have a penalty which depends on the file. Central isolated pawns are
considered more serious than isolated pawns on the wings. The values
range from 4 units (rook pawns) to 20 units (e and d pawns). The penalty
is decreased if the pawn is blocked by an enemy pawn, or if the pawn
is free to advance (as judged by my SEE) and there is an enemy pawn a
knight's move in front of the pawn. My evaluation of backward pawns are
quite similar.
Isolated and backward pawns are also considered when evaluating piece
placement. Knights and (to a lesser extent) bishops are given a bonus
for blocking isolated/backward pawns on central files. Rooks are given
a bonus for occupying a half-open file in front of an isolated/backward
pawn. A blocked isolated/backward pawn on the same colour as a friendly
bishop is a factor in my bad bishop eval.
Mine is similar, but again I use a file-dependant penalty where central
files are more important.
This is all very similar to what I do, except that my base advancement
bonus is much more aggressive (12, 12, 24, 48, 84, 128). I also have
a small penalty for a king which is cut off along a file and cannot
move closer to an enemy passed pawn. This helps in some rook endgames,
where a common theme is to use the rook to cut off the defending king.
I don't do this, but it looks like an idea worth trying. But one special
case I have is that passed rook pawns are very strong against knights.
When one side has a passed rook pawn and the other side has a knight and
no other pieces, I increase the passed pawn bonus by 50%.
Same here, but unfortunately my unstoppable passed pawn detection is not
always 100% reliable.
Tord

[Uri Blass]

Geschrieben von:/Posted by: Uri Blass at 15 June 2004 16:12:45:
Als Antwort auf:/In reply to: Re: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Fabien Letouzey at 15 June 2004 15:45:41:

What is your definition of backward pawns?

In Fruit a pawn is backward if:
1) it is not isolated
2) friendly neighbour pawns are more advanced
3) it cannot get to the same rank as one of them in one move (e.g. the square is controlled by an opponent pawn)
Fruit 1.0 only used 1) and 2) and that visibly resulted in stupid moves in many games.
Other than that pawn evaluation was not modified.
Thanks for sharing your work on pawn evaluation. Working on it is now
one of my priorities now (pawn-structure knowledge is necessary for
the quiet playing style I am aiming at) and I think we will end up
having similar features.
Fabien.

I do not aim of special playing style and the only thing that I care is playing better.
I think that things like static panelty for isolated pawns and static panelty for doubled pawns are not the smartest thing to do and there are other stupid things that I do not do well.
As far as I see you may do something better by using different values for the opening and the endgame and you do not have piece square table for it.
I also see that you give bigger panelty for isolated pawns that can be more easily attacked by the opponent because there is no pawn behind them.
Uri

[Uri Blass]

Geschrieben von:/Posted by: Uri Blass at 15 June 2004 16:29:53:
Als Antwort auf:/In reply to: Re: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Tord Romstad at 15 June 2004 16:08:26:

What is your definition of backward pawns?
I guess that my pawn evaluation is relatively stupid.
Here is short descreption
In all stages of the game:
1)static panelty of 0.1 pawn for every isolated pawn that is not passed pawn
2)static panelty of 0.2 pawns for every double pawn(2 pawns in the same file get panelty of 0.2 and 3 pawns on the same file a panelty of 0.4 pawns)
3)bonus for passed pawns based on the rank of the pawn
I use values 4 8 16 20 30 30 as basic values(100 is an estimate for the value of a pawn and I have piece square table for that).
4)These values are multiplied by 2 if the passed pawn is protected by a pawn and they are divided by 2 if the passed pawn is blocked by an enemy piece.
5)Additional bonus for advanced passed pawns if the opponent does not control the square that they want to go
6)evaluating the passed pawn relative to the kings.
In endgames:
1)evaluate pawns relative to the kings(for example if all the pawns are on files a-b-c and one of the king is in file h than the king get a panelty based on the distance between file c and file h that is 5).

2)panelty for king that it's rank is behind the most advanced passed pawn(when the panelty is bigger in case of advanced passed pawn).
I think that it may be better if I replace it by panelty for king that is not in the square of the pawn.
My evaluation does not consider side to move there and I guess that it also should be improved.
3)bonus of 0.1 pawns for having passed pawn in the biggest or smallest file.
(If there are no pawns in file a then white can get a bonus for having passed pawns in file b).
In pawn endgames:some evaluation to detect unstoppable passed pawns in part of the cases to give them a big bonus.

My definition is identical to Fabien's.
I have a penalty which depends on the file. Central isolated pawns are
considered more serious than isolated pawns on the wings. The values
range from 4 units (rook pawns) to 20 units (e and d pawns). The penalty
is decreased if the pawn is blocked by an enemy pawn, or if the pawn
is free to advance (as judged by my SEE) and there is an enemy pawn a
knight's move in front of the pawn. My evaluation of backward pawns are
quite similar.
Isolated and backward pawns are also considered when evaluating piece
placement. Knights and (to a lesser extent) bishops are given a bonus
for blocking isolated/backward pawns on central files. Rooks are given
a bonus for occupying a half-open file in front of an isolated/backward
pawn. A blocked isolated/backward pawn on the same colour as a friendly
bishop is a factor in my bad bishop eval.
Mine is similar, but again I use a file-dependant penalty where central
files are more important.
This is all very similar to what I do, except that my base advancement
bonus is much more aggressive (12, 12, 24, 48, 84, 128).
a small penalty for a king which is cut off along a file and cannot
move closer to an enemy passed pawn. This helps in some rook endgames,
where a common theme is to use the rook to cut off the defending king.
I don't do this, but it looks like an idea worth trying. But one special
case I have is that passed rook pawns are very strong against knights.
When one side has a passed rook pawn and the other side has a knight and
no other pieces, I increase the passed pawn bonus by 50%.
Same here, but unfortunately my unstoppable passed pawn detection is not
always 100% reliable.
Tord

I have piece square table for pawns and pawns in bigger files get bigger value also thanks to their place so only the table of bonus for passed pawns may be misleading.

I also have
I prefer to be careful here and I detect only part of the cases.
I can do errors but they are usually only errors of not detecting unstoppable passed pawns.
I do not consider pawn endgames as very important so improving it is not in the top of my priorities.
Movei lost against phalanx in pawn endgame because of lack of knowledge but as far as I can see pruning promotions in pawn endgames is probably a good idea to solve the problem in the time control of WBEC and it will be the only change that I plan to add about it for the first division.
Maybe panelty for backward pawns can also help in that position.
Uri

[Uri Blass]

Geschrieben von:/Posted by: Uri Blass at 15 June 2004 16:39:05:
Als Antwort auf:/In reply to: Re: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Uri Blass at 15 June 2004 16:29:53:

What is your definition of backward pawns?
I guess that my pawn evaluation is relatively stupid.
Here is short descreption
In all stages of the game:
1)static panelty of 0.1 pawn for every isolated pawn that is not passed pawn
2)static panelty of 0.2 pawns for every double pawn(2 pawns in the same file get panelty of 0.2 and 3 pawns on the same file a panelty of 0.4 pawns)
3)bonus for passed pawns based on the rank of the pawn
I use values 4 8 16 20 30 30 as basic values(100 is an estimate for the value of a pawn and I have piece square table for that).
4)These values are multiplied by 2 if the passed pawn is protected by a pawn and they are divided by 2 if the passed pawn is blocked by an enemy piece.
5)Additional bonus for advanced passed pawns if the opponent does not control the square that they want to go
6)evaluating the passed pawn relative to the kings.
In endgames:
1)evaluate pawns relative to the kings(for example if all the pawns are on files a-b-c and one of the king is in file h than the king get a panelty based on the distance between file c and file h that is 5).

2)panelty for king that it's rank is behind the most advanced passed pawn(when the panelty is bigger in case of advanced passed pawn).
I think that it may be better if I replace it by panelty for king that is not in the square of the pawn.
My evaluation does not consider side to move there and I guess that it also should be improved.

My definition is identical to Fabien's.
I have a penalty which depends on the file. Central isolated pawns are
considered more serious than isolated pawns on the wings. The values
range from 4 units (rook pawns) to 20 units (e and d pawns). The penalty
is decreased if the pawn is blocked by an enemy pawn, or if the pawn
is free to advance (as judged by my SEE) and there is an enemy pawn a
knight's move in front of the pawn. My evaluation of backward pawns are
quite similar.
Isolated and backward pawns are also considered when evaluating piece
placement. Knights and (to a lesser extent) bishops are given a bonus
for blocking isolated/backward pawns on central files. Rooks are given
a bonus for occupying a half-open file in front of an isolated/backward
pawn. A blocked isolated/backward pawn on the same colour as a friendly
bishop is a factor in my bad bishop eval.
Mine is similar, but again I use a file-dependant penalty where central
files are more important.
This is all very similar to what I do, except that my base advancement
bonus is much more aggressive (12, 12, 24, 48, 84, 128).

I have piece square table for pawns and pawns in bigger files get bigger value also thanks to their place so only the table of bonus for passed pawns may be misleading.

I mean that pawn on bigger ranks get bigger value.
pawn in the 7th is 1.70 pawn so it practically is evaluated as +2 in the average case.
I could decide that it is 1.00 pawn and give 1.00 pawn bonus for passed pawn in the 7th but it is not the same because I give some bonus for having big positional advantage relative to simpler evaluation that is based almost only on piece square table(the idea is that the advantage is not linear and the some of positional advantages can be bigger than the linear sum of them).
If I decide to give big positional bonus for passed pawn in the 7th the result is that this side will be considered to have big positional advantage even in case that the only advantage is having a pawn in the 7th so it will get another bonus and I do not like it.
Uri

[Robert Allgeuer]

Geschrieben von:/Posted by: Robert Allgeuer at 15 June 2004 17:25:14:
Als Antwort auf:/In reply to: Re: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Uri Blass at 15 June 2004 14:52:47:

I cannot believe that 100 elo point improvement is only thanks to small speed improvements and better time management.

All the info you need to find out is here http://wbec-ridderkerk.nl/html/details/Fruit.html . You can even look at the source code to see how the program was improved.

Fruit1.0 was improved by better evaluation and it is not news but the main surprise is the improvement from Fruit X05/09 to Fruit1.5
I got the impression from fabian's words that the improvement from
Fruit X05/09 to Fruit1.5 is only thanks to better time management and better compiler.
I understand that fruit did not know to use more time in case of fail low in X05/09 and maybe there were more improvement in time management(I did not read the relevant part of fruit source to find out) but I did not think that it should give more than 30 elo improvement.
together with better compiler 40 elo improvement make sense but 100 elo seems too good to be correct.
It is possible that I underestimated the time management improvement and the meaning of it is the same of being twice faster.
Uri

I believe to remember (I do not have the readme here) that mobility was added to Fruit 1.5, and I have always been assuming that this was the change with the biggest impact.
Robert

[Fabien Letouzey]

Geschrieben von:/Posted by: Fabien Letouzey at 15 June 2004 17:40:33:
Als Antwort auf:/In reply to: Re: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Robert Allgeuer at 15 June 2004 17:25:14:

I believe to remember (I do not have the readme here) that mobility was added to Fruit 1.5, and I have always been assuming that this was the change with the biggest impact.
Robert

Biggest visible impact (on the playing style) no doubt. However it also made Fruit slower (by 20% if I recall).
After the addition of insufficient-material detection and single-reply extension (also some piece-square-table changes), the result was two similar versions Fruit X 05/05 and 05/09. Heinz has tested one of those intermediate versions, so they can be used for comparison.
Overall the addition of mobility perhaps accounts for 20 or 30 Elo increase max. (guesstimate).
Fabien.

[Dann Corbit]

Geschrieben von:/Posted by: Dann Corbit at 15 June 2004 21:28:01:
Als Antwort auf:/In reply to: Re: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Fabien Letouzey at 15 June 2004 15:32:43:

Hi,

Fruit1.0 was improved by better evaluation and it is not news but the
main surprise is the improvement from Fruit X05/09 to Fruit1.5
I got the impression from fabian's words that the improvement from
Fruit X05/09 to Fruit1.5 is only thanks to better time management and
better compiler.
I understand that fruit did not know to use more time in case of fail
low in X05/09 and maybe there were more improvement in time
management(I did not read the relevant part of fruit source to find
out) but I did not think that it should give more than 30 elo
improvement.
together with better compiler 40 elo improvement make sense but 100
elo seems too good to be correct.
It is possible that I underestimated the time management improvement
and the meaning of it is the same of being twice faster.
Uri

There was an intermediate version Fruit X 05/21 that played in some
tournaments (e.g. WBEC division qualifications).
Fruit X 05/21 and Fruit 1.5 are very similar in strength.
As far as I can remember, the only differences between Fruit X 05/09
and Fruit X 05/21 were:
1) time management
2) 20% speed improvement (on Linux)
I will check in my log files.
I mentionned the compiler as a way of thanking Slater Wold. I
obtained some speed improvement by changing low-level data structures.
Only Windows users can measure the actual difference in NPS. Joachim
Rang has the development executables and Dann probably does as well.
We can measure that if you want. You can't compare with Fruit 1.0
though because mobility slowed it down quite a lot.
I don't use aspiration search so I have no fail low. But Fruit 1.0
always stops the search when the time limit is reached, even if it
could "know" (with appropriate code) that the move is very bad.
Bear in mind that Fruit 1.0 has a hard time limit (no time management
by my standards). Fixing this is part of what I call "time management"

I think that this last change is probably what accounts for the big Elo jump.
If fruit avoids a bad move even one time in a game, it will make a very large change. If the bad move leads to a loss, it is worth it even to lose on time to resolve something better. If you lose a full pawn against a computer, it is very difficult to come back. If you lose a full piece, it will take a miracle. If you lose a major piece, then you can forget it.
So why not sit and think about it until you find something that minimizes the damage? Even if the clock expires you were going to lose anyway.
That is why I think it is not a good thing to remove programs that lose on time from tournaments. They are probably doing the right thing, sometimes.



my ftp site {remove http:// unless you like error messages}

[Uri Blass]

Geschrieben von:/Posted by: Uri Blass at 15 June 2004 23:01:52:
Als Antwort auf:/In reply to: Re: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Dann Corbit at 15 June 2004 21:28:01:

Hi,

Fruit1.0 was improved by better evaluation and it is not news but the
main surprise is the improvement from Fruit X05/09 to Fruit1.5
I got the impression from fabian's words that the improvement from
Fruit X05/09 to Fruit1.5 is only thanks to better time management and
better compiler.
I understand that fruit did not know to use more time in case of fail
low in X05/09 and maybe there were more improvement in time
management(I did not read the relevant part of fruit source to find
out) but I did not think that it should give more than 30 elo
improvement.
together with better compiler 40 elo improvement make sense but 100
elo seems too good to be correct.
It is possible that I underestimated the time management improvement
and the meaning of it is the same of being twice faster.
Uri

There was an intermediate version Fruit X 05/21 that played in some
tournaments (e.g. WBEC division qualifications).
Fruit X 05/21 and Fruit 1.5 are very similar in strength.
As far as I can remember, the only differences between Fruit X 05/09
and Fruit X 05/21 were:
1) time management
2) 20% speed improvement (on Linux)
I will check in my log files.
I mentionned the compiler as a way of thanking Slater Wold. I
obtained some speed improvement by changing low-level data structures.
Only Windows users can measure the actual difference in NPS. Joachim
Rang has the development executables and Dann probably does as well.
We can measure that if you want. You can't compare with Fruit 1.0
though because mobility slowed it down quite a lot.
I don't use aspiration search so I have no fail low. But Fruit 1.0
always stops the search when the time limit is reached, even if it
could "know" (with appropriate code) that the move is very bad.
Bear in mind that Fruit 1.0 has a hard time limit (no time management
by my standards). Fixing this is part of what I call "time management"

I think that this last change is probably what accounts for the big Elo jump.
If fruit avoids a bad move even one time in a game, it will make a very large change. If the bad move leads to a loss, it is worth it even to lose on time to resolve something better.
So why not sit and think about it until you find something that minimizes the damage? Even if the clock expires you were going to lose anyway.
That is why I think it is not a good thing to remove programs that lose on time from tournaments. They are probably doing the right thing, sometimes.

No
By losing on time you can be sure of losing the game.
If you play the oponnent can blunder later.
If you lose a full pawn against a computer, it is very difficult to come back. If you lose a full piece, it will take a miracle. If you lose a major piece, then you can forget it.
No
losing on time is never the right thing.
If you think that there is no practical chance then it is better to resign.
Uri

[Dann Corbit]

Geschrieben von:/Posted by: Dann Corbit at 15 June 2004 23:41:03:
Als Antwort auf:/In reply to: Re: now the impact--gantlet Fruit 1.5, 600 games (Blitz) geschrieben von:/posted by: Uri Blass at 15 June 2004 23:01:52:

Hi,

Fruit1.0 was improved by better evaluation and it is not news but the
main surprise is the improvement from Fruit X05/09 to Fruit1.5
I got the impression from fabian's words that the improvement from
Fruit X05/09 to Fruit1.5 is only thanks to better time management and
better compiler.
I understand that fruit did not know to use more time in case of fail
low in X05/09 and maybe there were more improvement in time
management(I did not read the relevant part of fruit source to find
out) but I did not think that it should give more than 30 elo
improvement.
together with better compiler 40 elo improvement make sense but 100
elo seems too good to be correct.
It is possible that I underestimated the time management improvement
and the meaning of it is the same of being twice faster.
Uri

There was an intermediate version Fruit X 05/21 that played in some
tournaments (e.g. WBEC division qualifications).
Fruit X 05/21 and Fruit 1.5 are very similar in strength.
As far as I can remember, the only differences between Fruit X 05/09
and Fruit X 05/21 were:
1) time management
2) 20% speed improvement (on Linux)
I will check in my log files.
I mentionned the compiler as a way of thanking Slater Wold. I
obtained some speed improvement by changing low-level data structures.
Only Windows users can measure the actual difference in NPS. Joachim
Rang has the development executables and Dann probably does as well.
We can measure that if you want. You can't compare with Fruit 1.0
though because mobility slowed it down quite a lot.
I don't use aspiration search so I have no fail low. But Fruit 1.0
always stops the search when the time limit is reached, even if it
could "know" (with appropriate code) that the move is very bad.
Bear in mind that Fruit 1.0 has a hard time limit (no time management
by my standards). Fixing this is part of what I call "time management"

I think that this last change is probably what accounts for the big Elo jump.
If fruit avoids a bad move even one time in a game, it will make a very large change. If the bad move leads to a loss, it is worth it even to lose on time to resolve something better.
So why not sit and think about it until you find something that minimizes the damage? Even if the clock expires you were going to lose anyway.
That is why I think it is not a good thing to remove programs that lose on time from tournaments. They are probably doing the right thing, sometimes.

No
By losing on time you can be sure of losing the game.
If you play the oponnent can blunder later.
If you lose a full pawn against a computer, it is very difficult to come back. If you lose a full piece, it will take a miracle. If you lose a major piece, then you can forget it.
No
losing on time is never the right thing.
If you think that there is no practical chance then it is better to resign.

The effect is identical. And the problem is that you have no idea how long it will take to resolve the search. To decide how much time to consume is a programmer decision, and the resolution of the time checks in the search could cause a failure to detect it has run out of time.
So if I see I have suddenly lost a huge part of my score, I may decide to allocate a great deal of time to resolve it. It would be nice if I never used all that remains, but I think it is excusable to do so.


my ftp site {remove http:// unless you like error messages}