New android app "BadukAI"

[akigo]

the optimized 60b network appeared almost 8 months ago. is it possible to make a new network at least once every half a year? on some devices, such as mine, an unoptimized network does not work even 40b

8 months is quite a long time, indeed, so I made a new one, available here: https://github.com/aki65/aki65.github.i ... ag/v1.15.0

[And]

@akigo

Thanks a lot! very pleased!

[lovego]

Did you try yolov3 to recognize the stones on the board. ^-^

[akigo]

I had a closer look at the links you provided, but without much success.

The first one uses yolov3 and is quite understandable, but doesn't give a reference to the trained yolo weights, that are used for detection. And I would need these weights as I have neither the knowledge nor the computational resources to create them myself.

The second one uses yolov5 and doesn't include any documentation at all, so it's difficult to figure out which purpose each source has. Moreover it doesn't contain weights either (only a link to a baidu net disk and I couldn't download that file).

[Amigo]

akigo, could you add a variant of the game on a large board? a good size for several reasons is 27×27. https://senseis.xmp.net/?LargeBoards, etc.

[And]

Yes, that would be great! as far as I know, there are no game programs for android with support for large boards. 27x27 or 29x29 is very interesting!
https://senseis.xmp.net/?PlatinumDragon ... artingMove

[akigo]

Currently, I'm busy with "more practical" enhancements of BadukAI, so there won't be large boards in the next release. But maybe after that ...

[lovego]

Coooool!! Can you tell us what the "more practical" enhancements is? ^-^

[akigo]

It's about faster networks, additional configuration parameters, sgf editing and the ability to pass for the kyu-rank bot.

[Dragon]

Currently, I'm busy with "more practical" enhancements of BadukAI, so there won't be large boards in the next release. But maybe after that ...

great news! I often play on big boards. this is very interesting and helps the game on a regular board!

[lovego]

When will the new version come out? I cannot wait to see.

[akigo]

I made a new version (1.16) available on play store https://play.google.com/store/apps/deta ... r.baduk_ai and github https://github.com/aki65/aki65.github.i ... ag/v1.16.0 with the following changes:

1) Faster networks
The built-in Katago 20b network and the 40b network from https://github.com/aki65/aki65.github.i ... ag/v1.16.0 are about 30% faster on recent Snapdragon devices. I couldn't check the effect on other chipsets, but they should be at least as fast as before.

2) 27x27 and 29x29 boards

3) You can now choose the NPU accelerator, that BadukAI should use, in the "Edit Config" dialog. By default, BadukAI makes a reasonable choice, but that will not be optimal on all chipsets. So you can now try all available choices yourself and check which one is fastest.

4) Navigation into sgf variations
If the game tree contains several continuations from the current position, the button "next var" will be enabled. Tapping this button doesn't change the position, but switches to the next variation. As a result the circle, which indicates the next move, will change. If you then tap ">", you will advance into the chosen variation. If you are already on the last variation, "next var" will switch to the first one. The button "back to main" will always take you back to the main line (to the position where your chosen variation left off the main line).
When you are not at the end of the current game branch and you enter a move manually or let the AI generate a move, you create a new variation.
So you now have basic sgf editing functionality.

5) The kyu-rank bot now passes when appropriate.

[And]

@akigo

you're just a wizard!!! thank you very much, very much!!! all the best and good health!!!

[And]

on an old tablet (MTK8735P) network 20B ~45 visits/min, version 1.15 network 20B ~35 visits/min

[And]

when I want to download the network, via Choose *.txt.gz/*.bin.gz, I can't select any file. any networks are loaded through Choose *.tflite (bin and txt, and not optimized)! network g170small-b4c32-s709658624-d275852492 does not load in any way