Life In 19x19

And here is the journey of the training, from absolute beginner to pro.

2022-06-28 20:55:07-0400: Candidate won match, score 20.000 to 1.000 in 21 games, accepting candidate daoqi-s583680-d41745
2022-06-29 08:15:50-0400: Candidate won match, score 20.500 to 0.500 in 21 games, accepting candidate ...

Hi,

Just want to share the success of adapting KataGo to play Daoqi.

Recently, I've used the modified KataGo to train a Daoqi model. After >2 weeks of continuous training and approximately 100k games, I got a pretty strong model.

How strong? I'm an AGA 6D, and the model can beat me with 4 ...

I'm still in the process of figuring out what changes I need to make in order to get all pieces (self play, training, gtp etc) to work for Daoqi. It'll take some time. Any suggestions or help will be greatly appreciated.

Yes I posted it on Discord and got it answered. I included the responses below for reference:

cmf — 06/17/2022

multiStoneSuicideLegals is set to "false,true", I think this means half games will allow suicide, half won't. If you don't want suicide to be allowed, you should just set it to false ...

I've been training Daoqi model from my KataGo fork/branch. It seems working fine. However I noticed a few issues:

In some games, I noticed suicide moves are generated in some games. multiStoneSuicideLegals is set to "false,true". Not sure what that means but I suppose it won't allow suicide. Could ...

I tried to set cudaUseFP16 to false. Both gatekeeper and benchmark worked fine.

Thanks a lot. I did try to run benchmark and got same error. I'll try the downgrade and other suggestions.

Hi @lightvector,

Hope this finds you well!

Not sure whether you remember me. Two years ago I spent a few months trying to set up KataGo on my laptop to train a model to play Go and also worked on adapting KataGo to play one of Go's variants - Daoqi. However I wasn't able to get very far because I ...

I updated the opencl kernels and have created a PR. When you get a chance, I would appreciate if you can take a little time to review the PR and see whether I missed any place. If not, I'll pull latest code to my repo and start training.

Whole PR: https://github.com/gcao/KataGo/pull/3/files

OpenCL ...

Great! Thank you!

@lightvector

I'm looking into this again. Wonder whether below two places are where I need to update in order to support Daoqi in OpenCL. Do you mind giving some hints please? Thank you.

https://github.com/lightvector/KataGo/blob/master/cpp/neuralnet/openclkernels.cpp#L195
https://github.com ...

Congratulations to the new release. There is a lot of interesting enhancements in it. For me the most welcome feature is the synchronous training loop. When I find some time, I'll play with it to see whether I can train a net from my underpowered computer that is playable on an amateur level(even ...

I googled around but couldn't find much information regarding cyclic boundary in both CUDA and OpenCL. So I was going to make the changes you suggested in OpenCL code. Looks like I should update https://github.com/lightvector/KataGo/blob/master/cpp/neuralnet/openclkernels.cpp but don't know how. If ...

I guess diam=5 makes it a 5x5 convolution as mentioned in one of your post. That's why I changed it to 5. I'm not sure 3x3 is good enough for the net to handle the cyclic boundaries properly, even though it's good for Go.

Anyway you are totally right I should compare diam=3 and diam=5 in order to ...

I've updated the conv2d and dilated_conv2d to take diam parameter and add the cyclic boundary when necessary. I also updated diam to 5 wherever it's set to 3. However I noticed diam is set to 1 in many other places (e.g. p1/intermediate_conv, g1, p2 layers). Should I change the diameter in all ...