ZBaduk - LeeLa Zero and KataGo from your webbrowser

spook · **#181**

@Maharani
The save-file problem should be fixed. With apologies for the inconveniences.

Maharani · **#182**

Thanks

Please do keep us posted if you employ the newest Kata net ;3

spook · **#183**

Just installed a new version of the Front-End. This one is actually a heavy one.
It contains the upgrade to Angular 9 (wait, isn't there a version 10 now ?)

A nifty addition, is a list of the biggest mistakes of the game, which I am sure most visitors will enjoy.
At first this list is hidden, It only shows up once a considerable part of the game has been analysed.

Attachment:

new features.jpg [ 63.55 KiB | Viewed 2915 times ]

For some time now I had the impression that there were not enough advantages for paying members. In fact about 80% of the capacity of ZBaduk goes to free users. Although I do plan to keep most features available for free, some lines had to be drawn. So, I went through the feature list and put some limitations in place. For paying members nothing changes, but free users, may want to check this feature table. Personally, I think they are pretty fair.

ZBaduk also has a new e-mail address. Feel free to write to: bram at zbaduk dot com.
Let me know what you think.

More features are coming up shortly.

lichigo · **#184**

Thank you very much,I suscribed for 3 months and the new feature is great ^^

Maharani · **#185**

Visually it looks great

However, the AI table is a bit broken now... it only shows the statistics sometimes, and even when it does it occasionally freezes up.

lichigo · **#186**

Hello, I was wondering if it possible to use katago with a komi of 0? When I try it doesn't work. but I tried with something like minus 10 and it works. Should I just use 0.5 komi ?

spook · **#187**

Thank you for your support everybody.

Maharani wrote:

Visually it looks great

However, the AI table is a bit broken now... it only shows the statistics sometimes, and even when it does it occasionally freezes up.

Strange, I'll look into that asap.

Cyan · **#188**

Sometimes a low playout high winrate move gets picked (N11 in this case) instead of the good move (L18).

Maharani · **#189**

spook wrote:

Maharani wrote:

Visually it looks great

However, the AI table is a bit broken now... it only shows the statistics sometimes, and even when it does it occasionally freezes up.

Strange, I'll look into that asap.

To be more precise, the AI table only shows up if I click on the bot name. It stays frozen until I click on the bot name again, which is when it updates.

lichigo · **#190**

Hello ,
Does someone know how to set up katago with 0 komi ? I can only do it with 0.5 or -1.

spook · **#191**

I just released a couple of fixes:

- The stats-table-refresh problem should be fixed. (reported by Maharani).

- A fix for a potential performance issue with the bot-dropdown in the connection wizard.

- There was already an option to filter only the top 32 moves, we've added some more options to that selection.
(i.e. top4, top8, top12, top16, top20, top24, top28, top32 or all.)

As of yesterday, ZBaduk has a stripe account, meaning that we can actually create invoices from there as well.
For now, it is mostly done manually though. So, if you wanted to upgrade your account earlier, but did not have access to paypal,
then this will probably be a solution. In that case, just drop me an e-mail: bram at zbaduk dot com

For the technical guys out there:
I also published some of the ZBaduk internal code as a public npm library: mostly the SGF parsing. (i.e. "zbaduk-commons")
And I am upgrading our back-end to use NestJS, which seems to offer a lot of advantages in scalability.

In parallel, I am also working on AQZ integration and on a redesign of the charts.
And I found a Chinese and Korean translator ! Hurray !
So, more good stuff coming up surely !

spook · **#192**

lichigo wrote:

Hello ,
Does someone know how to set up katago with 0 komi ? I can only do it with 0.5 or -1.

I think you can just type 0 komi in there.
Is that what you were looking for ?

Attachment:

Video_2020-07-26_020359.gif [ 176.36 KiB | Viewed 2509 times ]

spook · **#193**

Cyan wrote:

Sometimes a low playout high winrate move gets picked (N11 in this case) instead of the good move (L18).

You do have a point. And just for clarity, I do think KataGo would actually be playing that move if it had to make one, even though it has a low amount of playouts.

Afaik, by default KataGo just takes the move with the highest LCB, as long as it has more than 10 playouts. Although, I wouldn't be surprised if this was configurable to some degree. Then again, even if there is, tweaking KataGo, wouldn't be a solution, because Leela Zero has the same issues.

But again, I do think that you have a point. So, here's a proposal to move towards a solution.
ZBaduk could add an option in its UI settings:
"Min playouts for best move", with 5 possible options to pick:
>33% of visits of top 3 moves.
>25% of visits of top 4 moves.
>20% of visits of top 5 moves.
>10% of visits of top 10 moves.
no filter

Cyan · **#194**

spook wrote:

You do have a point. And just for clarity, I do think KataGo would actually be playing that move if it had to make one, even though it has a low amount of playouts.

Afaik, by default KataGo just takes the move with the highest LCB, as long as it has more than 10 playouts. Although, I wouldn't be surprised if this was configurable to some degree. Then again, even if there is, tweaking KataGo, wouldn't be a solution, because Leela Zero has the same issues.

But again, I do think that you have a point. So, here's a proposal to move towards a solution.
ZBaduk could add an option in its UI settings:
"Min playouts for best move", with 5 possible options to pick:
>33% of visits of top 3 moves.
>25% of visits of top 4 moves.
>20% of visits of top 5 moves.
>10% of visits of top 10 moves.
no filter

That's a good idea. Can anyone test if Katago actually plays N11 in the above situation? It sounds like a big problem for Katago if it's true. Although it initially thinks the winrate for N11 is 89%, when it is actually played, black's winrate drops to 47%, losing 7 points, a huge disadvantage at such a high level of play.

lichigo · **#195**

Thank you for the answer, but when I do this black starts with 45%. So I don't understand. Sorry

I just saw the score chart is not working very well. Seems freezing after few moves. Can't see next moves or thé score.

spook · **#196**

lichigo wrote:

Thank you for the answer, but when I do this black starts with 45%. So I don't understand. Sorry

The cause of the problem has been found, a fix is coming up within the next 24 hours.

As mentioned in one of my other recent posts, there is also a major update coming which replaces all charts.

Attachment:

charts preview.png [ 54.07 KiB | Viewed 2421 times ]

One of the advantages is that several charts can be mapped on top of each other and toggled visible/invisible.
And it also saves some space by doing so. But it needs a little more work. -
Hopefully, we have a release at the end of this week.
PS: Most importantly, this chart has smarter zooming. It zooms by selecting a rectangular area on it.

lightvector · **#197**

spook wrote:

Afaik, by default KataGo just takes the move with the highest LCB, as long as it has more than 10 playouts. Although, I wouldn't be surprised if this was configurable to some degree. Then again, even if there is, tweaking KataGo, wouldn't be a solution, because Leela Zero has the same issues.

Actually, both KataGo and Leela Zero require that the highest LCB move also has enough playouts compared to other moves, not just 10, otherwise LCB would be a great harm on the strength instead of a benefit. It's a bad idea to pick a barely-searched move when the barely-searched move might only falsely seem good due to not having been read out enough to find the correct tactics yet. For KataGo, it's currently required by default that the LCB move has at least 15% of the playouts that the top playouts move has.

But zbaduk probably should not hardcode any such rule on its own. The final word on the preference ordering of moves should be given by the "order" field returned by the analysis. Zbaduk should simply just be sorting by this "order" field, whose purpose is specifically to let the bot give its preference ordering according to whatever combination of value/playouts/policy or whatever internal metrics it deems best. I'd be surprised if the "order" field were ranking the 80-playout move at the top, unless KataGo has also been somehow explicitly configured to also override the 15% threshold as well, so this would fix all the issues like the one you just observed that arise from trying your own custom sorting logic.

niallgo · **#198**

Hi,

1. I've noticed an issue when viewing on my iPad. All the ai recommended moves are the same black colour with the numbers black on top so it's very hard to see. Not sure if this is just me as I'm using an old iPad.

2. Is there a way to see the current move's points effect rather than it's percentage increase or decrease when using katago? The chart lists the point difference of the recommended moves but not the effect of the move that was just played.

3. Is there a way to share my current analysis with someone else so we can view together online?

Thanks for making this site, it's really helpful!

lichigo · **#199**

Still cannot use the 0 komi ^^ Is it possible to change the rules too? Like 0 komi with JPN rules. I review a lot of jubango of Go seigen these days so Zbaduk is super useful

spook · **#200**

lichigo wrote:

Still cannot use the 0 komi ^^

Should be fixed now.
(Also 0 komi with japanese rules works fine now.)

PS: On a totally unrelated note: there was an issue with bot-games sometimes not being saved automatically.
--> also that issue should be fixed now.

ZBaduk - LeeLa Zero and KataGo from your webbrowser

Who is online