KataGo Distributed Training and new networks

[johnsmith]

How to beat Zen without even thinking.

[And]

but in reality it was not easy - after 157 moves Zen winrate showed 83, and KataGo (if it is a winrate) W -85.92c
therefore this victory is more expensive!

[And]

KataGo s585 playouts 1 - Zen 9d H2 1:3

[go4thewin]

And, sorry probably stupid question but do you think s585 would beat zen 9 dan >50% in even games? Thanks for the games!

[And]

interesting! I'll play the match for the night!
I find it difficult to answer, since victories with a sufficient share of "luck"

[go4thewin]

Thanks And!

[And]

KataGo s585 playouts 1 - Zen 9d, komi 7.5, 14:6, KG W 7:3, B 7:3. 70%!

[And]

will the network be strong enough to win against 9 dan with a 9 stone handicap with 1 playouts? it seems incredible!

[go4thewin]

I couldn't even believe your results. When I first saw it, I thought you must have set maxPlayouts to 1 but forgot to limit the numSearchthreads to 1 so it ran extra playouts, but when I ran two game I got the same result, 2:0, wow. The next bot to defeat is golaxy's elephant bot on 19x19.com, which they rate as "1p". I thought the katago policy would never defeat it or zen 7, but now I think soon it will defeat the elephant. This is amazing. Thanks And!

[And]

It seemed to me that the LZ is performing more than the set playouts, if you do not limit numSearchThreads. in KataGo, I did not limit this parameter, and the log file shows that only 1 playouts are being executed (always)

[And]

I thought that it was always, but after looking at the last log, I found that there were 2 playouts 2 times per game. thanks for the info!
2 times per game is not much, but now, of course, I'll change numSearchThreads

[go4thewin]

I wonder if using this getconfig.cpp when compiling would fix the issue
https://github.com/portkata/KataGo/tree ... pp/program
probably not, but it uses the line

Code: Select all

if(maxPlayouts = 1)                  replace("$$NUM_SEARCH_THREADS", "numSearchthreads = 1")

at line 217, that's the only difference from the master branch getconfig.cpp . It feels like maybe that command is not in the right syntax or in the right place?

You're right, not much difference between 1 & 2 playouts
s580 1 po won 5/6 against gtp4zen7.

Screenshot_20210130-205017.png (366.63 KiB) Viewed 49501 times

Edit: Quick testing (6 games) with new optimized net, it seems to be atleast as strong as the last 40b net of the non-distributed run. optimized s580 policy (t1 p1 ncachesize = 0) against 20b net 5 playouts (t1 p5) was 4 wins - 2 losses, non optimized s503 40b was 2 wins - 4 losses . The optimized net is almost 1/4 the size and 4 times faster. Really amazing stuff.

[And]

You're right, not much difference between 1 & 2 playouts

I did not state this

[go4thewin]

My mistake, I misread your post. In the nets I have tested though, usually over sample sizes of 50 or a 100 games, at the larger net sizes, there is a very minimal difference between 1 and 2 playouts. Sometimes, because of noise the 2 playouts may seem weaker.

But what I'm really curious about is whether the optimized net is stronger than the s503 net. I haven't run enough games yet, but I'm guessing it is? That would really be incredible given it's size and speed. It is almost the same size as Leela Zero 157! As well as the 15b katago nets

[And]

look at here viewtopic.php?p=263170#p263170
I don't know what the accuracy of the stored weights is, but it looks like 99 or even 100%!