OliThink 5.4.0 has been published with an big leap in strength for only 3 lines of code
Moderators: hgm, Rebel, chrisw
-
- Posts: 12540
- Joined: Wed Mar 08, 2006 8:57 pm
- Location: Redmond, WA USA
Re: OliThink 5.4.0 has been published with an big leap in strength for only 3 lines of code
I guess with shared hash and threads, about 3500 Elo{*].
i have 64 threads on my box.
[*] CCRL or CEGT scale
i have 64 threads on my box.
[*] CCRL or CEGT scale
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
-
- Posts: 41433
- Joined: Sun Feb 26, 2006 10:52 am
- Location: Auckland, NZ
Re: OliThink 5.4.0 has been published with an big leap in strength for only 3 lines of code
I will be starting an OliThink gauntlet for CCRL 40/15 this week.
I was going to use 5.6.6, but I guess that I should now use 5.6.8 or later if you release again today?
I was going to use 5.6.6, but I guess that I should now use 5.6.8 or later if you release again today?
gbanksnz at gmail.com
-
- Posts: 725
- Joined: Tue Dec 18, 2007 9:38 pm
- Location: Munich, Germany
- Full name: Dr. Oliver Brausch
Re: OliThink 5.4.0 has been published with an big leap in strength for only 3 lines of code
Hi Graham,Graham Banks wrote: ↑Sun Aug 30, 2020 1:08 am I will be starting an OliThink gauntlet for CCRL 40/15 this week.
I was going to use 5.6.6, but I guess that I should now use 5.6.8 or later if you release again today?
Which OS are you using? Windows? I have just released a 5.6.8 using popcnt hardware support for Windows 64:
http://brausch.org/home/chess/OliThink568pop.win64.zip
I am looking forward for your tourney.
Last edited by OliverBr on Sun Aug 30, 2020 6:04 am, edited 1 time in total.
-
- Posts: 41433
- Joined: Sun Feb 26, 2006 10:52 am
- Location: Auckland, NZ
Re: OliThink 5.4.0 has been published with an big leap in strength for only 3 lines of code
Windows.OliverBr wrote: ↑Sun Aug 30, 2020 5:58 amHi Graham,Graham Banks wrote: ↑Sun Aug 30, 2020 1:08 am I will be starting an OliThink gauntlet for CCRL 40/15 this week.
I was going to use 5.6.6, but I guess that I should now use 5.6.8 or later if you release again today?
Which OS are you using? Windows? I have just released a 5.6.8 using popcnt hardware support for Windows 64:
http://brausch.org/home/chess/OliThink568pop.win64.zip
I'll use 5.6.8 unless there is a later one once I'm ready to start.
gbanksnz at gmail.com
-
- Posts: 725
- Joined: Tue Dec 18, 2007 9:38 pm
- Location: Munich, Germany
- Full name: Dr. Oliver Brausch
Re: OliThink 5.4.0 has been published with an big leap in strength for only 3 lines of code
I know, I am releasing a lot of versions nowadays, that's because I have a new toy, a 32-core test computer, so I could improve quite a bit. Anyway the 5.6.8 is good, a newer version is not expected very soon.Graham Banks wrote: ↑Sun Aug 30, 2020 5:59 am Windows.
I'll use 5.6.8 unless there is a later one once I'm ready to start.
-
- Posts: 725
- Joined: Tue Dec 18, 2007 9:38 pm
- Location: Munich, Germany
- Full name: Dr. Oliver Brausch
Re: OliThink 5.4.0 has been published with an big leap in strength for only 3 lines of code
Hi Graham.. I actually found a little change which improves 5.6.8 another 20 ELO points.. So, yes another version (5.6.9) is coming soon.OliverBr wrote: ↑Sun Aug 30, 2020 6:07 amI know, I am releasing a lot of versions nowadays, that's because I have a new toy, a 32-core test computer, so I could improve quite a bit. Anyway the 5.6.8 is good, a newer version is not expected very soon.Graham Banks wrote: ↑Sun Aug 30, 2020 5:59 am Windows.
I'll use 5.6.8 unless there is a later one once I'm ready to start.
Code: Select all
# PLAYER : RATING ERROR POINTS PLAYED (%) W D L D(%) CFS(%)
1 OliThink 5.6.8b : 22 9 2002.0 3765 53.2 1290 1424 1051 37.8 100
2 OliThink 5.6.8a : 0 ---- 1763.0 3765 46.8 1051 1424 1290 37.8 ---
White advantage = 21.06 +/- 4.45
Draw rate (equal opponents) = 38.06 % +/- 0.79
-
- Posts: 725
- Joined: Tue Dec 18, 2007 9:38 pm
- Location: Munich, Germany
- Full name: Dr. Oliver Brausch
Re: OliThink 5.4.0 has been published with an big leap in strength for only 3 lines of code
5.6.9 has been released. It gains about 25 ELO to 5.6.8, thus OliThink finally has more than ELO 2600.
The trick was a revert of the following change in OliThink 5.1.7 from the year 2008:
5.1.6 has in line 1549:
5.1.7 has:
5.1.7 has been 25 ELO weaker than 5.1.6 after modification. I couldn't know those days, but now I found it out with a big tournament.
So in 5.6.9 I reverted it, again using a factor 8.
PS: Funny fact: When starting small tournaments with fast timecontrol and another engine like Fruit2.1, it always plays very strong at the beginning.
After about 100 games for each engine:
But with more games this is converging to the expected result:
This is reproducible. What could be an explanation for this?
The trick was a revert of the following change in OliThink 5.1.7 from the year 2008:
5.1.6 has in line 1549:
Code: Select all
for (i = 0; i < 64; i++) nmobil[i] = bitcnt(nmoves[i])*8;
Code: Select all
for (i = 0; i < 64; i++) nmobil[i] = (bitcnt(nmoves[i])-1)*6;
So in 5.6.9 I reverted it, again using a factor 8.
PS: Funny fact: When starting small tournaments with fast timecontrol and another engine like Fruit2.1, it always plays very strong at the beginning.
After about 100 games for each engine:
Code: Select all
# PLAYER : RATING ERROR POINTS PLAYED (%) W D L D(%) CFS(%)
1 OliThink 5.6.9 : 70 69 50.5 90 56.1 34 33 23 36.7 68
2 Fruit 2.1 : 54 65 47.5 90 52.8 33 29 28 32.2 95
3 OliThink 5.6.8 : 0 ---- 38.0 92 41.3 21 34 37 37.0 ---
White advantage = 10.54 +/- 25.14
Draw rate (equal opponents) = 35.92 % +/- 4.31
Code: Select all
# PLAYER : RATING ERROR POINTS PLAYED (%) W D L D(%) CFS(%)
1 Fruit 2.1 : 92 21 613.5 1008 60.9 518 191 299 18.9 100
2 OliThink 5.6.9 : 28 21 478.5 1008 47.5 325 307 376 30.5 100
3 OliThink 5.6.8 : 0 ---- 420.0 1008 41.7 270 300 438 29.8 ---
White advantage = 28.64 +/- 7.94
Draw rate (equal opponents) = 27.16 % +/- 1.17
-
- Posts: 70
- Joined: Tue Dec 31, 2019 2:52 am
- Full name: Kieren Pearson
Re: OliThink 5.4.0 has been published with an big leap in strength for only 3 lines of code
This is very strange. Does the GUI restart each engine after each match?OliverBr wrote: ↑Sun Aug 30, 2020 11:52 am 5.6.9 has been released. It gains about 25 ELO to 5.6.8, thus OliThink finally has more than ELO 2600.
The trick was a revert of the following change in OliThink 5.1.7 from the year 2008:
5.1.6 has in line 1549:5.1.7 has:Code: Select all
for (i = 0; i < 64; i++) nmobil[i] = bitcnt(nmoves[i])*8;
5.1.7 has been 25 ELO weaker than 5.1.6 after modification. I couldn't know those days, but now I found it out with a big tournament.Code: Select all
for (i = 0; i < 64; i++) nmobil[i] = (bitcnt(nmoves[i])-1)*6;
So in 5.6.9 I reverted it, again using a factor 8.
PS: Funny fact: When starting small tournaments with fast timecontrol and another engine like Fruit2.1, it always plays very strong at the beginning.
After about 100 games for each engine:But with more games this is converging to the expected result:Code: Select all
# PLAYER : RATING ERROR POINTS PLAYED (%) W D L D(%) CFS(%) 1 OliThink 5.6.9 : 70 69 50.5 90 56.1 34 33 23 36.7 68 2 Fruit 2.1 : 54 65 47.5 90 52.8 33 29 28 32.2 95 3 OliThink 5.6.8 : 0 ---- 38.0 92 41.3 21 34 37 37.0 --- White advantage = 10.54 +/- 25.14 Draw rate (equal opponents) = 35.92 % +/- 4.31
This is reproducible. What could be an explanation for this?Code: Select all
# PLAYER : RATING ERROR POINTS PLAYED (%) W D L D(%) CFS(%) 1 Fruit 2.1 : 92 21 613.5 1008 60.9 518 191 299 18.9 100 2 OliThink 5.6.9 : 28 21 478.5 1008 47.5 325 307 376 30.5 100 3 OliThink 5.6.8 : 0 ---- 420.0 1008 41.7 270 300 438 29.8 --- White advantage = 28.64 +/- 7.94 Draw rate (equal opponents) = 27.16 % +/- 1.17
-
- Posts: 725
- Joined: Tue Dec 18, 2007 9:38 pm
- Location: Munich, Germany
- Full name: Dr. Oliver Brausch
Re: OliThink 5.4.0 has been published with an big leap in strength for only 3 lines of code
No, cutechess-cli is not restarting the engine. There is actually a parameter in the xboard protocol ("reuse") , it's the recommended "1", the engine will not restartet. Actually this should be standard.Kieren Pearson wrote: ↑Sun Aug 30, 2020 12:39 pmThis is very strange. Does the GUI restart each engine after each match?
I guess that Fruit is a good learner just alone by filling the Hash(TT)Tables. OliThink always replaces hash-entries, so they don't live very long.
After more 3000 games, there is +22 ELO, same as in the battle between the OliThinks
Code: Select all
# PLAYER : RATING ERROR POINTS PLAYED (%) W D L D(%) CFS(%)
1 Fruit 2.1 : 91 12 1931.0 3156 61.2 1644 574 938 18.2 100
2 OliThink 5.6.9 : 22 12 1474.5 3158 46.7 1015 919 1224 29.1 100
3 OliThink 5.6.8 : 0 ---- 1330.5 3158 42.1 868 925 1365 29.3 ---
White advantage = 28.37 +/- 4.38
Draw rate (equal opponents) = 26.28 % +/- 0.69
-
- Posts: 70
- Joined: Tue Dec 31, 2019 2:52 am
- Full name: Kieren Pearson
Re: OliThink 5.4.0 has been published with an big leap in strength for only 3 lines of code
I find it hard to believe that fruit is able to use transposition entries from different games to observe a noticeable increase in strength over time. There’s no way those entities are surviving that long and if they were future games should have had a different opening and the entires wouldn’t even be useful anyways.
Do you use an opening book for your tournaments?
Two more possibilities: It was just a coincidence that your engine did better for the first 100 games, or maybe your engine has a bug which causes it to have lowered nps over time. I say this because my engine had a similar bug early in its development.
Do you use an opening book for your tournaments?
Two more possibilities: It was just a coincidence that your engine did better for the first 100 games, or maybe your engine has a bug which causes it to have lowered nps over time. I say this because my engine had a similar bug early in its development.