For discussing go computing, software announcements, etc.
Garf
Dies in gote
Posts: 34 Joined: Mon Jun 06, 2016 1:22 amGD Posts: 0Been thanked: 10 times
Post
by Garf Wed Feb 22, 2017 1:00 pm
Code: Select all
50 Leela-0.9-1c1g 3320 107 107 154 67% 2935 0%
62 Zen-11.4n-1c 3244 40 40 633 71% 2861 0%
76 ELO ahead of the commercial Zen
So Leela (with GPU) is between half a stone and a stone stronger than Zenith Go 6. johnsmith called it, Gomoto was wrong.
goame
Lives with ko
Posts: 132 Joined: Sat Jan 28, 2017 12:25 amGD Posts: 0Been thanked: 7 times
Post
by goame Wed Feb 22, 2017 1:17 pm
Garf wrote: Code: Select all
50 Leela-0.9-1c1g 3320 107 107 154 67% 2935 0%
62 Zen-11.4n-1c 3244 40 40 633 71% 2861 0%
76 ELO ahead of the commercial Zen
So Leela (with GPU) is between half a stone and a stone stronger than Zenith Go 6. johnsmith called it, Gomoto was wrong.
Is it possiple to use 4x nvidia geforce gtx 1080 ti with Leela?
And how big is the benefit from 1 to 2 to 4 gpus?
Garf
Dies in gote
Posts: 34 Joined: Mon Jun 06, 2016 1:22 amGD Posts: 0Been thanked: 10 times
Post
by Garf Wed Feb 22, 2017 1:19 pm
goame wrote:
Is it possiple to use 4x nvidia geforce gtx 1080 ti with Leela?
No, because you cannot buy those yet
pookpooi
Lives in sente
Posts: 727 Joined: Sat Aug 21, 2010 12:26 pmGD Posts: 10Has thanked: 44 times Been thanked: 218 times
Post
by pookpooi Wed Feb 22, 2017 6:30 pm
It's hard to compare between Zen 6 (11.4) and Leela 0.9 in cgos, While Leela 1c1g is above Zen 1c by 76 elo, it's below Zen 2c by 77 elo. We need Leela without GPU (maybe 1 or 2 c or both) to have a better comparison.
yoyoma
Lives in gote
Posts: 653 Joined: Mon Apr 19, 2010 8:45 pmGD Posts: 0Location: Austin, Texas, USAHas thanked: 54 times Been thanked: 213 times
Post
by yoyoma Wed Feb 22, 2017 7:31 pm
I saw somewhere (compgo?) that the Leela author said he is not running Leela on CGOS. So it is an unofficial test by someone else.
Garf
Dies in gote
Posts: 34 Joined: Mon Jun 06, 2016 1:22 amGD Posts: 0Been thanked: 10 times
Post
by Garf Thu Feb 23, 2017 1:49 am
yoyoma wrote: I saw somewhere (compgo?) that the Leela author said he is not running Leela on CGOS. So it is an unofficial test by someone else.
I think it's a test by pookpooi actually.
The most interesting would probably be to run Leela with 1 cpu and no GPU. Then we know how much the GPU is worth. I'm not sure it affects the conclusion vs Zenith because Zenith can't use the GPU anyway, but that is not Leela's fault.
Zen 14.0 is now also on CGOS!
pookpooi
Lives in sente
Posts: 727 Joined: Sat Aug 21, 2010 12:26 pmGD Posts: 10Has thanked: 44 times Been thanked: 218 times
Post
by pookpooi Thu Feb 23, 2017 2:13 am
If it was test by me I wouldn't use gpu in the first place
Gomoto
Gosei
Posts: 1733 Joined: Sun Nov 06, 2016 6:56 amGD Posts: 0Location: EarthHas thanked: 621 times Been thanked: 310 times
Post
by Gomoto Thu Feb 23, 2017 5:18 am
Gomoto was wrong.
not true, he
was right
(Leela 0.8 was not stronger than Zenith 6)
Gomoto
Gosei
Posts: 1733 Joined: Sun Nov 06, 2016 6:56 amGD Posts: 0Location: EarthHas thanked: 621 times Been thanked: 310 times
Post
by Gomoto Thu Feb 23, 2017 7:28 am
Leela 0.9.0 defeats Zen again (this time without pondering, still using much more time than Zen)
Zenit Go 6 [setting: 20 seconds per move]
Leela 0.9.0 GPU [setting: 20 minutes for the game] (used more than 40 minutes)
CPU Intel 2500K GPU Nvidia GTX 970
Leela's advice towards Zen: Move 91 should have been at J10. Zen looses with an empty triangle
(52% => 47% win)
Zen was still quite sure to win the game at this time (60% win)
Pippen
Lives in gote
Posts: 677 Joined: Thu Sep 16, 2010 3:34 pmGD Posts: 0KGS: 2dHas thanked: 6 times Been thanked: 31 times
Post
by Pippen Thu Feb 23, 2017 1:28 pm
This is a typical diagramm of an opening I play with pretty good success. After analyzing it for 20 minutes, Leela gives it only about 41.5% for White with the slow 'a' as best continuation for White.
I am interested in what CS or Zen say to this position, and how they would continue for White. I usually continue with 'b' or 'c'.
$$B
Click Here To Show Diagram Code [go]$$B
Garf
Dies in gote
Posts: 34 Joined: Mon Jun 06, 2016 1:22 amGD Posts: 0Been thanked: 10 times
Post
by Garf Fri Feb 24, 2017 3:36 am
Given that you are 5D, can't you just play this out vs Leela, and see what happens after a bunch of moves? Either it'll show you a refutation, or its score will go towards 50%.
johnsmith
Lives with ko
Posts: 136 Joined: Fri May 27, 2016 3:55 amRank: KGS 6 danGD Posts: 0Has thanked: 47 times Been thanked: 21 times
Post
by johnsmith Fri Feb 24, 2017 6:24 am
Pippen wrote: This is a typical diagramm of an opening I play with pretty good success. After analyzing it for 20 minutes, Leela gives it only about 41.5% for White with the slow 'a' as best continuation for White.
I am interested in what CS or Zen say to this position, and how they would continue for White. I usually continue with 'b' or 'c'.
$$B
Click Here To Show Diagram Code [go]$$B
Zen suggest k4 with 53% for white. But I'd rather trust Leela. If you look more closely at its Net Win %, it says w is winning by 28% (i.e. losing the game) which is, I believe, more accurate than Win %. At least that's what those Deep Mind guys were looking at when they were testing AlphaGo.
Pippen
Lives in gote
Posts: 677 Joined: Thu Sep 16, 2010 3:34 pmGD Posts: 0KGS: 2dHas thanked: 6 times Been thanked: 31 times
Post
by Pippen Fri Feb 24, 2017 7:42 am
Zen suggest k4 with 53% for white. But I'd rather trust Leela. If you look more closely at its Net Win %, it says w is winning by 28% (i.e. losing the game) which is, I believe, more accurate than Win %. At least that's what those Deep Mind guys were looking at when they were testing AlphaGo.
Interesting. How can there be such a difference between the two probabilities?
I rechecked and here is what Leela says in detail: D2(best move): Win% 42, MC win% 51.7, Net win% 31.8, Net prob% 29. What is the meaning of all these numbers anyway? I always thought that win% matters, the rest not so much.
Uberdude
Judan
Posts: 6727 Joined: Thu Nov 24, 2011 11:35 amRank: UK 4 danGD Posts: 0KGS: Uberdude 4dOGS: Uberdude 7dLocation: Cambridge, UKHas thanked: 436 times Been thanked: 3718 times
Post
by Uberdude Fri Feb 24, 2017 7:49 am
Did everyone set the same komi, and ensure the different bots all respect it?
Bill Spight
Honinbo
Posts: 10905 Joined: Wed Apr 21, 2010 1:24 pmHas thanked: 3651 times Been thanked: 3373 times
Post
by Bill Spight Fri Feb 24, 2017 9:22 am
Pippen wrote: This is a typical diagramm of an opening I play with pretty good success. After analyzing it for 20 minutes, Leela gives it only about 41.5% for White with the slow 'a' as best continuation for White.
I am interested in what CS or Zen say to this position, and how they would continue for White. I usually continue with 'b' or 'c'.
$$B
Click Here To Show Diagram Code [go]$$B
BTW, I like "a". It is pretty much sente, and after White "b", instead, it is a good point for Black.
The Adkins Principle:At some point, doesn't thinking have to go on?
(52% => 47% win)
I am interested in what CS or Zen say to this position, and how they would continue for White. I usually continue with 'b' or 'c'.