Rayon (rn-3.9) is the strongest open source go AI

[Shoreline]

Why not post the home page here? It is open source right? I'm not so familiar with the dragon server and was lost in finding any info.

Here is source link:
https://github.com/zakki/Ray

When I compile this it says it is version 8.0.1, the same as a compile from one year ago.

There is link "6 releases", pick the latest one, rn3.10, which was released just couple of days ago

[roy7]

There is link "6 releases", pick the latest one, rn3.10, which was released just couple of days ago

Oh there's binaries too, at least for 3.3. I was trying to compile myself direct off the github.

I tried compiling

https://github.com/zakki/Ray/releases/tag/rn-3.10.2

for Windows with MWIN2 and gcc 6.3.0. To get past compiler errors in UctSearch.cpp I had to change "std::atomic_int64_t" to "std::atomic<std::int64_t>" in the appropriate spots, and also change atomic_fetch_add in some places to be "(long int)1) instead of "1)" at the end of the function all.

Unfortunately, it will seg faults a lot if I try running it after compilation. This is with the CNTK beta 11 CPU-only version. The last binary release was rn-3.3.

[ez4u]

Rn.3.3.1-4c0g doesn't use gpu. http://www.yss-aya.com/cgos/19x19/cross ... -4c0g.html
Other Rn.xx-4c uses i7-2600K (or 6700K) + GTX980ti

I assume you are one of the developers. Is there any chance you could add binaries for 3.10?

[Shoreline]

Just report to everyone here that Rayon Bot (RN3.9) at DGS achieved 6 dan level and slaughtered a human 5 dan player on an even game. I actually
do not know exactly the upper limit strength of this Rayon Bot at DGS, could be a lot stronger than 6 dan.

[Shoreline]

http://www.yss-aya.com/cgos/19x19/bayes.html

Leela 0.9.1 is out in the results. It appears that new leela is significantly weaker than RN3.9, The difference could be several stones.

Just got information from Rayon authoer that RN3.9-4c in the above link CGOS ranking uses 4core i7 and GTX-980Ti GPU. This setting is significantly beefer setting than my DGS Rayon Bot setup

[pookpooi]

Hard to compare between different program because of different hardware (4 cores vs. 1 core + 1 gpu)

[Garf]

http://www.yss-aya.com/cgos/19x19/bayes.html

Leela 0.9.1 is out in the results. It appears that new leela is significantly weaker than RN3.9, The difference could be several stones.

I think you misread, it is not Leela 0.9.1, it is Leela 0.9 running on a single core. It's stronger than Rn3.8 on 4 cores + GPU.

[Shoreline]

http://www.yss-aya.com/cgos/19x19/bayes.html

Leela 0.9.1 is out in the results. It appears that new leela is significantly weaker than RN3.9, The difference could be several stones.

I think you misread, it is not Leela 0.9.1, it is Leela 0.9 running on a single core. It's stronger than Rn3.8 on 4 cores + GPU.

Leela0.9.1c1g in the link. I suppose that is leela 0.9 1 CPU with 1 GPU. In the go AI, GPU calculation is the main calculator,
CPU does not matter that much. That is still a comparable hardware.

[Mike Novack]

Just a note --- if we are going to discuss the effect of hardware on the playing strength of these programs we can't just say things like "a 4 core I7". That encompasses a range in crunch power almost 4:1 between the strongest and the weakest I7 << about the same ratio as between the strongest CPU's being used by go programs and the strongest I7's >>

[roy7]

Leela0.9.1c1g in the link. I suppose that is leela 0.9 1 CPU with 1 GPU. In the go AI, GPU calculation is the main calculator,
CPU does not matter that much. That is still a comparable hardware.

Leela makes great use of the cpu though. GPU is only used for the neural network related computations, the rest of the program is all cpu based.

Yes GPU is important and no GPU means weaker play, but a strong cpu will make a much stronger Leela.

[Shoreline]

I got an alert from the author that Rn.3.10.2 windows binary version is available for download. But it requires GPU

[Gomoto]

Can you provide the link to the binary? Thanks.

[as0770]

Here is source link:
https://github.com/zakki/Ray

When I compile this it says it is version 8.0.1, the same as a compile from one year ago.

There is link "6 releases", pick the latest one, rn3.10, which was released just couple of days ago

Thanks but it won't compile:

g++ -O3 -Wall -std=c++11 -g -c src/UctSearch.cpp -o src/UctSearch.o
src/UctSearch.cpp:39:18: fatal error: Eval.h: Datei oder Verzeichnis nicht gefunden
#include "Eval.h"
^
compilation terminated.

[roy7]

Can you provide the link to the binary? Thanks.

https://github.com/zakki/Ray/releases

[roy7]

Thanks but it won't compile:

g++ -O3 -Wall -std=c++11 -g -c src/UctSearch.cpp -o src/UctSearch.o
src/UctSearch.cpp:39:18: fatal error: Eval.h: Datei oder Verzeichnis nicht gefunden
#include "Eval.h"
^
compilation terminated.

You have to download the CNTK as well and add the cntk/Include directory to the includes path when compiling. Either add to your PATH in general or add

Code: Select all

-I /path/to/cntk/Include

When you link at the end you'll need to add cntk/ctnk/EvalDll.dll (I think it was called) to the list of objects on the file command make does.