KataGo's releases are here. https://github.com/lightvector/KataGo/releases

The executable version that comes with Lizzie is v1.2 and is out of date. It doesn't support all the features that were released in the last few weeks (japanese rules, stronger handicap play, accurate lead estimation) and is not capable of running the latest stronger neural nets either.

Just to be clear: v1.3 does work with Lizzie. You need to install Lizzie first (if you haven't already), then install KataGo 1.3, then use the settings menu in Lizzie (or edit the Lizzie config file directly if you prefer) to replace KataGo 1.2 with 1.3.

Are you saying that KataGo is at 9d+ strength even at small number of visits/move (like 128 visits/move)?

Probably not that small?

I was referring more to numbers like your 1600. Of course not everyone has a good GPU, but many of the higher-end GPUs from the last couple of years are very easy to get much more than 1600 visits in reasonable amounts of time. Very top-end ones should be able do it in a second or less.

There's a reasonable chance 1600 for LZ/KataGo is large enough to beat any human in the world more than 50% of the time (except for the chance that maybe there is a specific repeatable exploit in some joseki), at relatively quick time controls? But it's not like I've tested that either, I could be way off.

And here we go. The run is still going. Another week passed so let's have some new neural nets!

https://github.com/lightvector/KataGo/r ... 1.3.1-nets

A stronger 20-block net, and a very strong 15-block net that is almost as strong as ELFv2's 20 block net at equal visits despite being a smaller and faster net. Maybe not exactly as strong? Not sure, my testing so far is noisy. It should be fairly close though.

Nice!

If we're going to see a plethora of KataGo nets (which I would welcome), would you like to propose a naming/numbering convention? Perhaps something a little more concise than "I did this analysis using g170-b20c256x2-s1420141824-d350969033"?

@lightvector
This!

_________________
Dave Sigaty
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21

Maybe you could say "g170 20 block s1.42G"? (i.e. 1.42 billion training samples)

s = samples that net was trained on
d = data rows generated at that point in selfplay

So s and d are two indicators of exactly where it is in the run.

At this point, I'm not going to change the convention mid-run, or try to rename the nets and files that have been uploaded, that would be a mess.

If at some point later we started a public distributed run, we could try a different naming convention then.

Edit: Maybe there's something yet better, but I went ahead and updated the all the Github release pages (https://github.com/lightvector/KataGo/releases) suggesting this convention as at least a shorter-hand name.

There are some ranking systems that try to figure this out. According to those rankings, katago 1.2 at 100 playouts was at least 6 Dan.
https://github.com/breakwa11/GoAIRatings Breakwa's rankings estimate leela zero 157 net at 100 playouts ranked at 6d human. CGOS has katago 1.2 at 100 playouts ranked higher (Bayle's ELO) than a leela zero 157 version (9006c708) at 200 playouts. http://www.yss-aya.com/cgos/19x19/cross ... _v100.html
These rankings also estimate that katago 1.3.1 at 800 playouts would rank higher than any human player in the world in even games, though I can't believe that. These are kind of just guesses.

_________________
set katago to play at your level https://docdro.id/sHZU1ti or experiment with gtp4zen ( https://rb.gy/kx2ilb )

When playing handicap games against Katago, the initial board positions are not always the same, but look like this:

Click Here To Show Diagram Code

[go]$$W
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . X . . . . . . . . . . . . . |
$$ | . . X , . . . . . , . . . X . , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X , . X . . . , . . . X . , X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

Is there a way to choose myself where to place the handicap stones?

Isn't that up to the GUI you use, and/or your own choice of where to put your stones?

The kind of handicap configuration you post is what KataGo itself would might choose if *it* was the one playing with that many stones. I'm not sure though why your GUI would ask KataGo to place the stones as if KataGo were the one getting the extra stones, rather than letting you do so.

I am using the last version of Lizzie. I unpacked the last version of KataGo in the directory Lizzie\katago, downloaded the weight g170e-b15c192-s1305382144-d335919935.txt.gz in that directory. In the list of engines, I entered the command "katago\katago.exe gtp -model katago\g170e-b15c192-s1305382144-d335919935.txt.gz -config katago\gtp_example.cfg".

That's all. So maybe it's in the configuration of Lizzie, but I couldn't find out where.

Ah, I see. Apparently the GTP spec indicates that bots should implement a fixed handicap placement method in addition to a free one, according to exact rules defined in the spec that vary with board size oddness/evenness and a few other things. (Yuck. )

KataGo never implemented this, since Chinese rules allow free placement, and until recently KataGo didn't support any other rules, so it wasn't necessary anywhere that KataGo could work. I'll go ahead and add this for next minor release.

The interesting thing is that even if I implement this now, Lizzie won't pick it up automatically because Lizzie, rather than either checking for the availability of the bot's placement methods, or placing the stones itself, appears to be hardcoded to call KataGo's free placement method.

Anyways, as a workaround until Lizzie also itself gets a new version, you can just play all the stones yourself (passing as white), and then just resume play from there?

Yes, I will do that.

Lizzie with KataGo "g170 20 block s1.42G" or "g170e 10 block s1.14G" quickly find a winning move:

https://lifein19x19.com/viewtopic.php?p=244851#p244851

SAI and ELF could not find this move in an hour

Someone is running a katago 1.3.1 bot on the KGS server using AWS and getting 1000 playouts per second. The bot is ridiculously strong: https://www.gokgs.com/gameArchives.jsp? ... Accounts=y

Username: katagobot

You can view all the sgfs in the link. This is a fun bot to watch play! Hopefully in the future, these bots make human pros obsolete.
Katagobot uses newest g170 20 block s1.42G net

_________________
set katago to play at your level https://docdro.id/sHZU1ti or experiment with gtp4zen ( https://rb.gy/kx2ilb )

See also user "petgo2", which is the latest katago running on an RTX 2080 Ti. Petgo2 is 9d. Up to now katagobot is only "9d?" because it has not yet lost a rated game!

Edit: 45 minutes after posting the above, katagobot finally lost a rated game by 6.5 points, giving 6 stones to a 5d.

_________________
Dave Sigaty
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21

What's more, katagobot is one of only two players in the KGS top 10 that's achieved their rank by playing humans, rather than by learning to beat up HiraBot on H6. (The other is stakeout at number 10).

@lightvector

Do you have some additional details on your internal rating tests btw?

On CGOS, some users have put up 800 visits katago bots.

This one I would guess is the first 1.3 release, and it ranks slightly higher than LZ ELFv2 v800, but is worse in the head-2-head matchup.

Another one should be the 2nd release. It has not played that much, and has been taken offline, but had even poorer results against LZ ELF v2

We of course don't know the settings of these, but what do you make of it?

Do you mean "LZ_05db_ELFv2_p800"? Taking a wild guess based on the name I would guess that this is ELF with 800 playouts, not visits? If so, then it may be getting somewhere around 2x the number of playouts that KataGo is getting. KataGo and LZ-ELF being almost an even match despite that difference seems like KataGo is doing well.

How much compute time that takes depends on hardware and a lot of other details. Some users have reported that LZ-ELF does get almost 2x as many playouts due to LZ having had more GPU optimization work, making it close. Some users have reported a much smaller gap than that already, such as on FP16 tensor core cards, and with a smaller gap KataGo easily outperforms.

(And if you wait a week or so, some more performance improvements should be coming too - working on it...).