Life In 19x19

For general interest, here's another KataGo analysis with 10000 playouts per move rather than just 1000. As expected, it's similar to the first graph, but with some differences in details.

it is not a "proper go" and my position had many weak points

I think this is 19x19 reasoning that is much less useful on 9x9. The reason is that in 19x19, the game is sufficiently long and open that the notion of "weak points" is a good heuristic for "it's hard to read out every specific sequence ...

And wrote:thank! waiting for new releases! We are also waiting the CPUonly version!

You can already run on the CPU if it supports OpenCL. It's quite slow though, since the cpu slowness compounds with the unoptimised opencl code.

Works very well...

Four handicap games against Crazy Stone Deep Learning :
First game : C.S. set at 2Dan, H3 (komi 7.5) LazyB won easily.
Second and third games : H2 against CS set at 5Dan. LazyB won both convincingly, with its winrate increasing steadily during the games (white histograms at the ...

Just released version 0.7 to Google Play, it should be available for download or upgrade very shortly. Major new features include:


⋅ Leela Zero updated to version 0.17, should be a bit faster.
⋅ Added region select buttons. When pondering, Leela Zero will only suggest ...

Superb program very easy to use, especially on a tablet.


Thanks!



Is there a way to play handicap go ? (with the 15b networks, LZ plays good H games)

Maybe a "pass" button would do the job ?


Right now there isn't, but at least adding a pass button would be easy enough (if I can find ...

I just added a 13x13 version, using the procedure discussed here for convering 19x19 weights for different board sizes. The weights used come from LZ network #205. Apparently they are very strong, but I think the network is also quite large and so might be a bit slower than you'd hope on a phone.

Wow. Looks nice.

Sadly I don’t have an Android device. Any chance that we can get this for iOS also?


Thanks!

Unfortunately iOS development is impossible without macOS, which I don't have (plus the developer account fee is far higher).

The app is written using the cross-platform Kivy framework ...

I've just released Lazy Baduk, a Leela Zero analysis tool for Android. The app is available on Google Play , or a few versions with different weights can be downloaded directly from Github (see below). It requires Android version 6.0 (Marshmallow) or higher.

https://raw.githubusercontent ...

It looks like (if someone could confirm, it'd be great) there was something there at the counter-hane at move 25. All my variations give a good result for black, but probably I am missing some variations.

Your main counter-hane variation line at M15 would be quite poor for black. When white ...

4) Nvidia or AMD or is there anything else?
5) any compatibility issues? I think LZ works with both, but didn't Facebook's ELF only work on Nvidia?


The important thing here is whether the applications use OpenCL or CUDA to run code on the GPU. OpenCL is an open standard that is supported by ...

I feel sure that it would be possible to construct a whole board position where Black should block at Q-17 instead of R-16.

I had a quick look and it seems there are a few professional examples. In some cases black blocks the 'wrong' way for the opposite reason to trying to live - if black is ...

Thank you for your thorough comments.

After going through them a little more detailed, there are some minor questions:

The comment on move 15 about having a white stone outside black's wall:

So, given the choices between
A :
$$Bc ...

Here's an SGF with some comments:

https://lifein19x19.com/forum/download/file.php?id=8783

Black does misplay the early fight a bit (although white does too), but the most interesting question seems to be around how black should get more out of white's invasion on the right side. I've picked out ...

At my current level, it feels like influence is soo easy to eradicate. Just toss in a stone, but not too close so it gets killed.

This is normal, and it's not a bad way to learn - I remember thinking similar things, and slowly adjusting the safe distance from that influence further and further ...