New android app "BadukAI"

[akigo]

I made a new version (1.10) available on play store https://play.google.com/store/apps/deta ... r.baduk_ai and github https://github.com/aki65/aki65.github.i ... ag/v1.10.0 , which includes updated KataGo, updated weights and a few other improvements (see release notes on github).

[lovego]

Please update the No.985 weight! It`s very strong i have tested

[floating]

Installed baduk_ai__arm64-v8a-rel-1.10.apk with the .bin.gz and .tflite in the same release on Xiaomi Redmi note 7 running Lineage OS 17. It crashed and rebooted the phone first, but now a few hours later I open it and the downloaded .bin.gz hash shows in settings and it works, however the board is not visible fully. I can see the row 3 at bottom and the bottom margin is at row 2, so I am missing a bit there (can't see the first row at all). There is plenty of blue area above the board and the long white/black win rate columns could be shorter too. I wonder how I should fix this? I tried searching the forum here but couldn't find good words to search if this problem has been solved already

edit: Ok, I went into "edit config" menu, and the badukAI was unresponsive for 5 seconds (I saw the options there and I tried clicking cancel) and then the phone rebooted itself again. I guess I won't be touching that menu at least for a while

[akigo]

Nobody has reported similar problems yet, so at first glance I would attribute these issues to Lineage OS. As I don't have this on my phone, I can't analyze these problems. So I'm sorry, I can't help you here.

[lovego]

The official weight of 60b has been released!

https://katagotraining.org/networks/

[akigo]

I published an optimized version of the currently strongest 40b network (s985) here: https://github.com/aki65/aki65.github.i ... ag/v1.10.1

Next I will have a look at 60b ...

[lovego]

Amazing! Thanks very much

[akigo]

An optimized version of the first official KataGo 60b network is now available at https://github.com/aki65/aki65.github.i ... ag/v1.10.2

[akigo]

In response to some user requests, I made a new version (1.11) available on play store https://play.google.com/store/apps/deta ... r.baduk_ai and github https://github.com/aki65/aki65.github.i ... ag/v1.11.0.

It includes optional time control for the human player and some optional sounds (placing of stones, byoyomi warning).

[lovego]

The katago 1.10.0 is released, Many minor improvements!
Will BadukAI be updated in the near future?

[akigo]

The katago 1.10.0 is released, Many minor improvements!
Will BadukAI be updated in the near future?

I went through the changes in KataGo 1.10, and I don't think any of them will be significantly noticeable in BadukAI. So I won't make a new release just for the sake of upgrading KataGo. But I have some new features in the making, and when they are ready, I will publish a new version. And this one will probably also include KataGo 1.10.

[akigo]

I made a new version (1.12) available on play store https://play.google.com/store/apps/deta ... r.baduk_ai and github https://github.com/aki65/aki65.github.i ... ag/v1.12.0 with the following changes:

1) BadukAI is now built with android SDK 30 (necessary for google to accept the update in play store). This implies, that on recent android versions BadukAI doesn't get global storage access anymore. Therefore external files (sgf's or network files) now have to be chosen with the system's native file picker app. So don't be surprised if you seemingly leave BadukAI when you are asked to select a file. Just choose the file in the native system app, and you will automatically get back to BadukAI.
In particular, all this means, that loading an external optimized network (which consists of two files) now requires you to select both files, one after the other.

2) After tapping "Load sgf" you can now not only load sgf's but also images. BadukAI will then detect the go board and stones in the image and load the resulting position. You can also produce that image "on the fly" with the camera (for best results use 4:3 portrait mode). This comes in very handy if you want to analyze a position you encounter in a book or a video or on a real board. Depending on lighting conditions etc., the detection is not always perfect, but usually very good.

3) You can also load a picture into BadukAI by sharing it to BadukAI from within another app. In that way you can quickly load a screenshot.

4) When no moves have been played yet, you can now edit the board: After tapping "add B stone" you can put any number of black stones on the board (in the same way as you would play them). Same goes for white stones. After tapping "remove stone" you can place a cross on the board (in the same way as a stone). By placing it on an existing stone, you can remove that stone.
By tapping the marked editing button again (whichever it is), you can stop its editing function. In that state, touching the board will play a stone as usual.

5) BadukAI has been upgraded to KataGo 1.10.

6) Captures are now displayed below the histogram.

These were a lot of changes, therefore some bugs may have crept in. So please report if anything goes wrong, so that I can try to fix it.

[WinPooh]

Is it possible to select strength levels with kyu/dan scale, like in CrazyStone or AhQ Go?

[akigo]

Is it possible to select strength levels with kyu/dan scale, like in CrazyStone or AhQ Go?

Yes, there are several possibilities as described here: https://aki65.github.io/#strength

Using "kyu-rank" in settings (with the 20b network in use) is the easiest way, but then playing style is sometimes a little inconsistent.
If you don't like that, load a weaker network (available from https://katagotraining.org/networks/) and set a small number of playouts by "edit config".

go4thewin has thankfully compiled a table of estimated ranks for a lot of network/playout combinations here: https://docs.google.com/spreadsheets/d/ ... sp=sharing

Furthermore there is this excellent site: https://github.com/breakwa11/GoAIRatings, which shows the strength of a lot of KataGo networks. The ranks there refer to 1200 playouts. As a rule of thumb, halving the playouts reduces strength by 200 Elo. In that way you can roughly tune it to your level.

[lovego]

The new version is great! The image recognition speed is very fast, but the accuracy needs to be improved, I found that sometimes there will be extra pieces on the edge of the board. In addition, whether it can support the recognition of 9 or 13 board!