Life In 19x19

If you see a slide that says "CAUTION", it's good to take that into account.

The word covered by the speaker's head is "self". Bot results in self-play are always massively exaggerated. It's not uncommon to see a 75% self-play winrate in selfplay to translate to 52% winrate against a third-party ...

I don't want get into nuanced argument here, but just to clarify my original post - I didn't say the computers will develop new theories, but that they'll help humans do it. Perhaps new styles of opening or tengen-based strategies will be possible to develop using automated analysis of results, or ...

It matters even for weaker players, since the computer...

* will be able to help humans develop new Go theories everyone can benefit from

* will become a much stronger teacher (to be taught well, it's not enough if the other player is just a stone or a few better than you; on the other hand ...

That's not a compressed neural network but just a frequency-based specialized pattern classifier.

First, there's a huge difference whether we talk about *training* the networks, or just *running* the program with pretrained networks. The requirements I mentioned are for training.

For running the network, you can certainly scale it down - it will just be able to search to a smaller depth. The ...

Just to train the SL network as in the paper, you'd need 150 gpu-weeks of time. And this extra accuracy grinding matters a lot, see Fig 2. Of course, maybe after 1 gpu-week you'd already be quite a bit above 50%, but I think the best open source solution right now is just about 50%.

I think that if ...

I think saying that the only advantage of KGS is the community is a little harsh. OGS has plenty of glitches, ranging from the semi-broken bot support, over many many confusing glitches connected to the multiple user ranks, to reviews just not being nearly as convenient as on KGS (due to a lot of ...

I still like KGS a lot, but using desktop Java is just such a pain. I think many people nowadays just do not install Java on their PC at all, maybe? It hogs memory, the applet keeps requesting updates and few desktop apps require it nowadays. Then, installing it just for KGS is something they ...

While simply rewriting the Python code into a different language might increase the speed (he gave an estimate of maybe 5x) that is not really the best way to go about it. You want to work from the algorithm in its abstract form, not the form where decisions about data representations, etc. have ...

Ok, let's open com/cyberoro/wbaduk/CGameApp.class in your favorite hexeditor (I used hed), find 0x06b9 in the file, and replace the 0x06 with 0x03,
I don't find 0x06b9 in the file.


I'm so sorry - I have meant 0xb6, not 0xb9 - you can verify the repr on the Wikipedia page I've linked. I'll also ...

Hi! So I wanted to try out CyberOro/WBaduk again after a while - on Linux, without Oracle Java. Well, that was a fun ride! This is a very raw and extremely technical guide, *not* really a step-by-step non-programmer thing, maybe someone will polish it up. (It took me about an hour to figure this out ...

quantumf: I think that Computer Go based approaches could actually produce pretty good sequence disambiguers, maybe better than a lot of human scribes! (...who can, though, ask other observers for help.) Also, even if a sequence is not perfect, if the end result is correct, some rate of mistakes may ...

One initial worry is that it's quite slow, despite the C optimization - takes a good few seconds to parse an image on an i7 machine. That may be an issue in certain situations. Not sure exactly what you mean by automatic game transmission, but moves are often played faster than a few seconds apart ...

Hi! Thanks for the feedback, nice testcases. Indeed, a lot of improvement will be probably needed for many uses, mainly more sophisticated stone detection I guess. (Then again, for some uses like automatic tournament game transmission where you can set things up a bit, I think Imago might be almost ...

Hi! A student of mine, Tomas Musil, has created open source Go board optical recognition software that seems pretty nicely usable. We have focused on completely automatic runs, so it automatically detects the board corners and then the stones on the board. You can find it at m together with a lot of ...