Life In 19x19

Posted: **Fri Jan 11, 2019 11:30 pm**

Personally I like #199 more. The games of #199 are so much more enjoyable to watch because #199 likes to attack. #200 maybe better than #199 in some aspects, but it always plays some boring moves such as solid defense, rarely tenuki or sacrifice few stones to gain influence.

Posted: **Wed Jan 16, 2019 11:33 am**

Vargo,
in the latest version AhQ Go (2.4.2) in the list of recommended weight appeared new network 192x5 Ms5. much stronger than f438268e. could you test it?

Posted: **Sun Jan 20, 2019 2:02 am**

Do you have a link to a network file usable with LZ0.16 ?

Posted: **Sun Jan 20, 2019 8:01 am**

https://drive.google.com/file/d/11uBKRz ... sp=sharing
when using LZ0.16 and Ms5-192x15 or f438268e sometimes LZ0.16 freezes. better lz0.15

Posted: **Sun Jan 20, 2019 9:23 am**

Thanks.
f438268 is OK with LZ016 and twogtp on my computer, but the combination Ms5, LZ(015 or 016), and twogtp is not OK, it crashes every time

Posted: **Sun Jan 20, 2019 11:12 am**

gogui-twogtp V1.4.9, leela-zero-0.15-cpuonly 20 games ok. I previously unpacked the network. I have a weak video card now, but I'll try to run the gpu version

Posted: **Sun Jan 20, 2019 11:33 am**

I started the gpu version (using sabaki) with f438268e - it plays normally, Ms5 - it works, but it makes stupid moves. It seems that Ms5 has problems with gpu

Posted: **Mon Jan 21, 2019 3:53 am**

40 games f438268e - Ms5 28-12 (10 sec/move, no pondering, LZ0.15-cpuonly, twogtp V1.4.9)

Posted: **Wed Jan 23, 2019 9:01 am**

60 more games, total
100 games f438268e - Ms5 63:37 (10 sec/move, no pondering, LZ0.15-cpuonly, twogtp V1.4.9)
Ms5 disappointed

Posted: **Thu Jan 24, 2019 9:28 am**

I heard networks bjiyxo f438268 (192x15) and LM_GX89 (192x15) were very strong. OK, I added two players : LZ_Elfv1 (224x290) and LZ#202 (256x40) to run a 4 players tournament.

Short time parity (2 min/game per player , with gpu:1x1080) no pondering , komi 7.5, every player playing 10 games against every other one.

Results :

202-Elf 3-7
202-f43 5-5
202-x89 2-8
x89-Elf 5-5
x89-f43 7-3
Elf-f43 5-5

I'm very much surprised by the poor result of #202 (too little time for 40b ?) and by the excellent result of GX89.

It's about 3h for the 60 matches. I'll try to run the same thing tomorrow, to see if I get similar results. Maybe (probably) I should let it run for a lot more games.

Complete tournament code :

Code: Select all

rem 202-elf

gogui-twogtp -black "C:\Users\jm\gogui150\leela-zero-0.16-win64ok\leelaz.exe --gtp --weights=C:\Users\jm\LZnetworks\202.gz --noponder" -white "C:\Users\jm\gogui150\leela-zero-0.16-win64ok\leelaz.exe --gtp --weights=C:\Users\jm\LZnetworks\elf2.txt --noponder" -games 10 -sgffile 202elf -time 2 -auto -komi 7.5 -alternate

gogui-twogtp -analyze 202elf.dat

rem  202-f43

gogui-twogtp -black "C:\Users\jm\gogui150\leela-zero-0.16-win64ok\leelaz.exe --gtp --weights=C:\Users\jm\LZnetworks\202.gz --noponder" -white "C:\Users\jm\gogui150\leela-zero-0.16-win64ok\leelaz.exe --gtp --weights=C:\Users\jm\LZnetworks\f438268.gz --noponder" -games 10 -sgffile 202f43 -time 2 -auto -komi 7.5 -alternate

gogui-twogtp -analyze 202f43.dat

rem 202-X89

gogui-twogtp -black "C:\Users\jm\gogui150\leela-zero-0.16-win64ok\leelaz.exe --gtp --weights=C:\Users\jm\LZnetworks\202.gz --noponder" -white "C:\Users\jm\gogui150\leela-zero-0.16-win64ok\leelaz.exe --gtp --weights=C:\Users\jm\LZnetworks\LM_GX89.txt --noponder" -games 10 -sgffile 202x89 -time 2 -auto -komi 7.5 -alternate

gogui-twogtp -analyze 202x89.dat


rem X89-elf

gogui-twogtp -black "C:\Users\jm\gogui150\leela-zero-0.16-win64ok\leelaz.exe --gtp --weights=C:\Users\jm\LZnetworks\LM_GX89.txt --noponder" -white "C:\Users\jm\gogui150\leela-zero-0.16-win64ok\leelaz.exe --gtp --weights=C:\Users\jm\LZnetworks\elf2.txt --noponder" -games 10 -sgffile x89elf -time 2 -auto -komi 7.5 -alternate

gogui-twogtp -analyze x89elf.dat

rem X89-f43

gogui-twogtp -black "C:\Users\jm\gogui150\leela-zero-0.16-win64ok\leelaz.exe --gtp --weights=C:\Users\jm\LZnetworks\LM_GX89.txt --noponder" -white "C:\Users\jm\gogui150\leela-zero-0.16-win64ok\leelaz.exe --gtp --weights=C:\Users\jm\LZnetworks\f438268.gz --noponder" -games 10 -sgffile x89f43 -time 2 -auto -komi 7.5 -alternate

gogui-twogtp -analyze x89f43.dat

rem elf-f43

gogui-twogtp -black "C:\Users\jm\gogui150\leela-zero-0.16-win64ok\leelaz.exe --gtp --weights=C:\Users\jm\LZnetworks\elf2.txt --noponder" -white "C:\Users\jm\gogui150\leela-zero-0.16-win64ok\leelaz.exe --gtp --weights=C:\Users\jm\LZnetworks\f438268.gz --noponder" -games 10 -sgffile elff43 -time 2 -auto -komi 7.5 -alternate

gogui-twogtp -analyze elff43.dat

all the stats.zip: (32.08 KiB) Downloaded 564 times

60 games.zip: (49.23 KiB) Downloaded 535 times

Posted: **Thu Jan 24, 2019 11:56 am**

Vargo wrote:

Short time parity (2 min/game per player , with gpu:1x1080) no pondering , komi 7.5, every player playing 10 games against every other one.

......

I'm very much surprised by the poor result of #202 (too little time for 40b ?) and by the excellent result of GX89.

I think this sort of thing requires REASONABLE time controls. Nothing faster than human blitz play. Saying that this irrelevant, that all that matters is equality, misses an important point.

Think of a race. Which is faster, a human or a race car?

a) Down a quarter mile straightway
b) Ten times around two pylons 50 feet apart

Posted: **Thu Jan 24, 2019 8:49 pm**

Well, a "long game" on my computer (GTX1050) with 20 seconds per move maybe just same as the "2 minutes" game on a GTX1080. #202 lost again to LM-GX89. Sometimes it seems bigger doesn't mean better in the AI world.

Posted: **Thu Jan 24, 2019 11:47 pm**

Is it possible to use the gx-89 with lizzie?

Posted: **Fri Jan 25, 2019 2:36 am**

lichigo wrote:Is it possible to use the gx-89 with lizzie?

Should be, I've used GX88 with Lizzie.

Posted: **Fri Jan 25, 2019 2:55 am**

Oh great, could you share the link. I would like to try it ^^

Life In 19x19

LZ's progression

Re: LZ's progression

Re: LZ's progression

Re: LZ's progression

Re: LZ's progression

Re: LZ's progression

Re: LZ's progression

Re: LZ's progression

Re: LZ's progression

Re: LZ's progression

Re: LZ's progression

Re: LZ's progression

Re: LZ's progression

Re: LZ's progression

Re: LZ's progression

Re: LZ's progression