Life In 19x19
http://www.lifein19x19.com/

New networks Leela Zero, a Sensible Artificial Intelligence
http://www.lifein19x19.com/viewtopic.php?f=18&t=17023
Page 1 of 5

Author:  And [ Wed Oct 30, 2019 5:39 am ]
Post subject:  New networks Leela Zero, a Sensible Artificial Intelligence

http://sai.unich.it

Author:  Vargo [ Fri Nov 01, 2019 4:22 am ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

Very nice, thanks !

sai9x9 seems very strong, about as strong as katago1.2 ?
9x9 Sabaki games between katago and sai are often very close.
(here, both programs play around 2-5 sec/move)



sai9x9 can win against gnugo at H2 and H3, and even sometimes at H4 (komi 0.5)





Attachments:
sgh4.sgf [986 Bytes]
Downloaded 2761 times
sgh3.sgf [731 Bytes]
Downloaded 2724 times
sw1.sgf [1.03 KiB]
Downloaded 2761 times

Author:  And [ Fri Nov 01, 2019 5:24 am ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

if I understood correctly, sai9x9 uses a network 94619dea? but the network was received on 12/21/2018? almost a year ago! but strange, I haven’t seen this network before

Author:  And [ Mon Nov 04, 2019 10:44 am ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

last network 848fe9be (10 sec/move, cpu version) http://sai.unich.it/networks/848fe9be1b ... 99fbf84.gz
against zen 7 1 dan (gtp4zen -t4 -s1800 -n1 -o1.3 -p0.35) - zen 7 surrendered on move 109

Attachments:
sai 848fe9be - zen 7.sgf [964 Bytes]
Downloaded 569 times

Author:  And [ Tue Nov 05, 2019 10:38 am ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

last network 43268842 (10 sec/move, cpu version) http://sai.unich.it/networks/43268842e5 ... 5d364fe.gz
against ZEN 7 2 dan (gtp4zen -t4 -s2000 -n1 -o1.0 -p0.4) - SAI won +2.5 (302 moves)
20 more games SAI & ZEN - 4:16

Attachments:
sai 43268842 - zen 7.sgf [2.43 KiB]
Downloaded 537 times

Author:  And [ Thu Nov 07, 2019 5:54 am ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

SAI #64 fe1cfa3c (10 sec/move, cpu version) http://sai.unich.it/networks/fe1cfa3c81 ... 95f2dfb.gz
against ZEN 7 3 dan (gtp4zen -t4 -s2200 -n2 -o2.0 -p0.45) - 2:18
zen surrendered in two games:

Attachments:
fe1cfa3c - zen 7 2.sgf [1.55 KiB]
Downloaded 547 times
fe1cfa3c - zen 7.sgf [1.46 KiB]
Downloaded 524 times

Author:  Vargo [ Fri Nov 08, 2019 5:33 am ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

Two more Sabaki 9x9 games Sai-Katago (both at 40000 playouts)
Maybe Katago 1.2 wins slightly more often than Sai





Attachments:
saikata2.sgf [752 Bytes]
Downloaded 2507 times
saikata.sgf [883 Bytes]
Downloaded 2475 times

Author:  Vargo [ Fri Nov 08, 2019 6:14 am ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

Two Sabaki 9x9 games Sai-gtp4zen (40000 playouts for Sai, -T10 -s40000 for Zen)
Sai wins even games and can win H2 with komi 7.5 (but not always)




Attachments:
saizen_h2_k75.sgf [612 Bytes]
Downloaded 2478 times
saizen.sgf [998 Bytes]
Downloaded 2442 times

Author:  frmor [ Fri Nov 08, 2019 7:10 am ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

Hi, SAI main author here!

What options are you using with SAI?

Be aware at least of the following ones:

--lambda [0.5]
--mu [0]
--nrsymm

The latter is the simplest. Merge symmetrical positions in order to explore more deeply. Highly recommended when playing at fixed visits as it could give quite a lot of strength increase

lambda and mu are real parameters intended as 0<=mu<=lambda<=1. They tell SAI to consider a range of komi values when playing, instead of just the real komi of the game. If you are playing with an opponent of equal strength it is recommended to set both to 0 (notice that this is NOT the default for lambda).

If you are playing with a weaker opponent, and/or there is handicap, the default values should be ok, at least for 9x9

Author:  And [ Fri Nov 08, 2019 10:49 am ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

frmor, thank you so much for this project!
I correctly understood, these recommendations for the case if SAI are used to play against another program (human)?
which values are better if the opponent is stronger? and mu = 0.5 or 1 can be used, and what will it give?

Author:  Vargo [ Fri Nov 08, 2019 8:32 pm ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

frmor wrote:
--lambda [0.5]
--mu [0]
--nrsymm
and --noponder -p 40000 --komi x.x
Thank you for your detailed explanation about lambda and mu. I like SAI very much :tmbup:

Author:  frmor [ Sat Nov 09, 2019 4:19 am ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

And wrote:
frmor, thank you so much for this project!
I correctly understood, these recommendations for the case if SAI are used to play against another program (human)?
which values are better if the opponent is stronger? and mu = 0.5 or 1 can be used, and what will it give?


It's a bit complicated, and you should read the paper to get full details. Let us say that lambda and mu control the two sides of the interval of komi values over which the winrate is averaged.

If mu=0 it means to keep the real komi as one side.

If lambda=1 it means to use, for the other side, the komi value that SAI thinks would make the current position fair (in the sense of 50% winrate).

Intermediate values are to be intended in the winrate scale, not komi scale, so that for example, if SAI thinks it is ahead, with winrate 80% and lambda=0.3, than it would compute the weighted mean 71% = 0.3 * 50% + 0.7 * 80%, look for the komi value that would give itself a 71% winrate, and then use this as the side.

So if SAI is black and real komi is 7.5, then this virtual komi value would be something more, say 9.3. Then SAI for the exploration of the current move would compute the nodes winrates as averages over the komi range [7.5 ; 9.3] and choose its move accordingly.

You can put mu=lambda>0 and in this way SAI would play as if the komi was the one computed, hence giving itself increasing score handicap when ahead. (And the opposite when behind.)



My recommendation is to use lambda=mu=0 when against a strong opponent, to try hard to win with the real komi.

You should use values larger than 0 when against a weaker opponent, both to improve the final score, when playing a fair game, and to improve the strength of play when starting with handicap.

When SAI is behind with a weaker opponent because of handicap, positive values of lambda (and maybe mu) were showed to improve its ability to recover from the initial disadvantage.

Author:  And [ Sat Nov 09, 2019 6:48 am ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

SAI #70 c74e6c7d (10 sec/move, cpu version) http://sai.unich.it/networks/c74e6c7d84 ... 2c385ee.gz
against ZEN 7 4 dan (gtp4zen -t4 -s2400 -n2 -o1.5 -p0.5) - 1:19
zen surrendered in one game:


Attachments:
sai c74e6c7d - zen 7.sgf [1.58 KiB]
Downloaded 2363 times

Author:  And [ Mon Nov 11, 2019 1:42 pm ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

SAI 3ca44219 (10 sec/move, cpu version) http://sai.unich.it/networks/3ca44219f0 ... cd19998.gz
against ZEN 7 5 dan (gtp4zen -t4 -s2700 -n2 -o1.0 -p0.55) - zen surrendered:


Attachments:
sai 3ca44219 - zen 7.sgf [1.41 KiB]
Downloaded 2289 times

Author:  splee99 [ Mon Nov 11, 2019 7:20 pm ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

Zen behaved weird in the last few moves, which seem to be low level blunders.

Author:  And [ Tue Nov 12, 2019 6:30 am ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

he probably had reasons to walk like that, but unfortunately he won’t explain it :)
I continued the game after move 149 several times, ZEN (gtp4zen) wins more often than loses. ZEN original is the same

Author:  And [ Tue Nov 12, 2019 11:58 am ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

SAI #82 8ee88aa5 (10 sec/move, cpu version) http://sai.unich.it/networks/8ee88aa51f ... 526963b.gz
against Crazy Stone Deep Learning 5 dan (WINDOWS) - SAI B+19.5:


Attachments:
sai 8ee88aa5 - CSDL.sgf [2.49 KiB]
Downloaded 2264 times

Author:  And [ Sat Nov 16, 2019 2:02 am ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

SAI #92 a520f71b (-v 1000, cpu version) http://sai.unich.it/networks/a520f71b10 ... 2c663ca.gz
against ZEN 7 6 dan (gtp4zen -t4 -r15 -s3000 -n3 -o4.4 -p0.6) - 1:19
zen surrendered in one game:


Attachments:
sai a520f71b - zen7 6d.sgf [1.68 KiB]
Downloaded 2159 times

Author:  And [ Sat Nov 16, 2019 4:54 am ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

SAI developer:

"I believe SAI will get superhuman in a few months, with the support of the community. Anyhow we don't presume that it will ever be as strong as the latest LZ networks. That's not its purpose.

The aim of SAI is to get rid of the suboptimal moves LZ or AG may play when they are too far ahead of their opponent (i.e. almost always against humans) or when they are far behind (i.e. at the beginning of a game with handicap).

When SAI will be superhuman, it could show really how much handicap (in points or maybe in stones) it could give to a strong player. Or how many points difference it could score in a fair game.
In endgame, it may be able to show unknown smart plays or exhibit optimal choices in difficult situations.

This is presently from very difficult to impossible under the LZ/AG structure, because you would need to tailor positions in which the perfection of endgame moves really would matter for winning, and not just for score."

https://github.com/leela-zero/leela-zer ... -548264646

Author:  And [ Mon Nov 18, 2019 3:28 am ]
Post subject:  Re: New networks Leela Zero, a Sensible Artificial Intellige

SAI #99 109c93bc (-v 1000, cpu version) http://sai.unich.it/networks/109c93bcf1 ... babe7ef.gz
against ZEN 7 7 dan (gtp4zen -t4 -r15 -s3500 -n3 -o2.8 -p0.65) - 1:19
zen surrendered in one game:

TIME_B: 999.5, TIME_W: 751.4

Attachments:
sai 109c93bc - zen7 7d.sgf [1.57 KiB]
Downloaded 2069 times

Page 1 of 5 All times are UTC - 8 hours [ DST ]
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/