Death of Giraffe, but hopefully not ML in chess!

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

whereagles
Posts: 565
Joined: Thu Nov 13, 2014 12:03 pm

Re: Death of Giraffe, but hopefully not ML in chess!

Post by whereagles »

Cheers Mat. Good luck on your new job.

I'm sure you'll keep coming here from time to time :)
User avatar
cdani
Posts: 2204
Joined: Sat Jan 18, 2014 10:24 am
Location: Andorra

Re: Death of Giraffe, but hopefully not ML in chess!

Post by cdani »

Congratulations on your new work!! Maybe someone will take the witness of Giraffe...
JoshPettus
Posts: 730
Joined: Fri Oct 19, 2012 2:23 am

Re: Death of Giraffe, but hopefully not ML in chess!

Post by JoshPettus »

I managed to compile giraffe again on osx but it crashes with a segfault 11 when I send it the go command. Is it possible the eval.net/meval.net are out of date? or do I have to break out the debugger?
[edit]
It works as soon as I get rid of meval.net I take it that file is no longer needed? giraffe doesn't even ask for it. The eval.net in the september 8, 2015 windows binnary doesn't work. It crashes right away with a abort trap 6. But my old eval.net works just fine.
matthewlai
Posts: 793
Joined: Sun Aug 03, 2014 4:48 am
Location: London, UK

Re: Death of Giraffe, but hopefully not ML in chess!

Post by matthewlai »

GONeill wrote:Congratulations on the new job!

Any chance of a final Giraffe compile on the Downloads page of your site? The current one is from September and I think you've done some updates since then.

Thanks!
'

Thanks Graham!

I will try to clean it up a bit and make a final release when I find time for it.
Disclosure: I work for DeepMind on the AlphaZero project, but everything I say here is personal opinion and does not reflect the views of DeepMind / Alphabet.
matthewlai
Posts: 793
Joined: Sun Aug 03, 2014 4:48 am
Location: London, UK

Re: Death of Giraffe, but hopefully not ML in chess!

Post by matthewlai »

MikeB wrote:All the Best Matthew - out of all the engines released over the last twenty 20 years , your engine, by far, was the most interesting. It is easy to see why Google hired you. Any final source updates before we turn the lights out?
Thanks Michael! I really hope it would inspire others to continue exploring this path!
Disclosure: I work for DeepMind on the AlphaZero project, but everything I say here is personal opinion and does not reflect the views of DeepMind / Alphabet.
matthewlai
Posts: 793
Joined: Sun Aug 03, 2014 4:48 am
Location: London, UK

Re: Death of Giraffe, but hopefully not ML in chess!

Post by matthewlai »

Werewolf wrote:Sad news.

I hope you can develop it in your spare time as a private engine. Then one day you may leave Google and the techniques may become so well known that there's no conflict anymore and the Giraffe can live once more.

Set the Giraffe free....
Maybe, but I would much prefer sharing what I do in my free time. That makes my hobbies a lot more fun for me.
Disclosure: I work for DeepMind on the AlphaZero project, but everything I say here is personal opinion and does not reflect the views of DeepMind / Alphabet.
matthewlai
Posts: 793
Joined: Sun Aug 03, 2014 4:48 am
Location: London, UK

Re: Death of Giraffe, but hopefully not ML in chess!

Post by matthewlai »

JoshPettus wrote:I managed to compile giraffe again on osx but it crashes with a segfault 11 when I send it the go command. Is it possible the eval.net/meval.net are out of date? or do I have to break out the debugger?
[edit]
It works as soon as I get rid of meval.net I take it that file is no longer needed? giraffe doesn't even ask for it. The eval.net in the september 8, 2015 windows binnary doesn't work. It crashes right away with a abort trap 6. But my old eval.net works just fine.
A lot of things have changed since then, so I'm not surprised that it doesn't run with old net definitions.

meval has never been required. It just makes the engine stronger if it's available. But the format changed, that's why the old version doesn't work.

I'll try to retrain new nets and make a final release when I find some free time.
Disclosure: I work for DeepMind on the AlphaZero project, but everything I say here is personal opinion and does not reflect the views of DeepMind / Alphabet.
duncan
Posts: 12038
Joined: Mon Jul 07, 2008 10:50 pm

Re: Death of Giraffe, but hopefully not ML in chess!

Post by duncan »

do you have hopes it will improve on go

from wiki

Currently the company's focus is on publishing research on computer systems that are able to play games, and developing these systems, ranging from strategy games such as Go[25] to arcade games




https://www.youtube.com/watch?v=EfGD2qveGdQ
User avatar
lantonov
Posts: 216
Joined: Sun Apr 13, 2014 5:19 pm

Re: Death of Giraffe, but hopefully not ML in chess!

Post by lantonov »

Giraffe was the deep hope of many CC enthusiasts for starting a new phase off the beaten track.

The code looks like Chinese to me. I hope someone with more knowledge picks it up and improves it possibly utilizing what we know from other engines.
matthewlai
Posts: 793
Joined: Sun Aug 03, 2014 4:48 am
Location: London, UK

Re: Death of Giraffe, but hopefully not ML in chess!

Post by matthewlai »

lantonov wrote:Giraffe was the deep hope of many CC enthusiasts for starting a new phase off the beaten track.

The code looks like Chinese to me. I hope someone with more knowledge picks it up and improves it possibly utilizing what we know from other engines.
The code is a bit messy since that's research code written to a tight deadline after all. I did try to keep it as clean as possible though, and i would say it's cleaner than most research code.

I am happy to answer questions about the code if anyone wants to take a look at it. You do need to have fairly good understanding of neutral networks and TD learning to begin with, though. Otherwise there would be too much to explain.

If I forget about this and stop checking the forum, feel free to ping me at matthewlai@google.com.
Disclosure: I work for DeepMind on the AlphaZero project, but everything I say here is personal opinion and does not reflect the views of DeepMind / Alphabet.