New Giraffe (Sept 8)

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

Moderator: Ras

Guenther
Posts: 4718
Joined: Wed Oct 01, 2008 6:33 am
Location: Regensburg, Germany
Full name: Guenther Simon

Re: New Giraffe (Sept 8)

Post by Guenther »

matthewlai wrote:This one only has some small changes. No 100+ Elo jump like last time.

https://bitbucket.org/waterreaction/gir ... 150908.zip

Changes:
1) Score reporting is now scaled down by 1/10. Hopefully that will play better with score-based adjudication systems. Score magnitudes are now more inline with what other engines report, though it's still probabilistic.
2) Time management has been improved (thanks to Graham Banks' testing that led me to notice this). About 33 Elo gained.
3) Gaviota tablebases now supported. See https://bitbucket.org/waterreaction/giraffe for details. I want to switch to a neural-network based endgame system later, but EGTB will do for now.

By the way, Giraffe is now open source, and the thesis is now available online. See thread on the programming forum if interested - http://talkchess.com/forum/viewtopic.ph ... 136#640136

This is probably going to be the last Giraffe release for a while. I have just submitted and defended my thesis. I think my brain needs a break from working on Giraffe full time for 3 months.

Will be travelling for a few weeks before getting back to Giraffe.

Thanks
Hi Matthew,

I played a few games against it now after I followed its reported progress since the first release and I must say I am really surprised how good it plays openings now
compared to the first few released games I have seen here. My congratulations it is already quite strong against Humans at least.
Memory usage seems hardcoded to be around 300MB right?
What would be the commandline param for WB to always load the Gaviota bases at start?
I noticed that the eval/meval files still are ascii - can you gain something later by a binary approach? (I guess the files will grow quite a bit when you do more training/learning?)

Guenther
User avatar
fern
Posts: 8755
Joined: Sun Feb 26, 2006 4:07 pm

Re: New Giraffe Engine (not Working)

Post by fern »

Thanks for your polite and understanding answer.

Fern
matthewlai
Posts: 793
Joined: Sun Aug 03, 2014 4:48 am
Location: London, UK

Re: New Giraffe Engine (not Working)

Post by matthewlai »

Henk wrote: Does your engine improves itself too ? So if it plays many games it plays better and better without human intervention.
Only in special training mode that's basically highly optimized self-play. Not during normal gameplay.
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: New Giraffe Engine (not Working)

Post by matthewlai »

Daniel Anulliero wrote:Girafe 64bits works great here in Arena
No book , I don't care arena have it 😊
I added girafe in my engine's set for testing Isa 😉
I like your new approach , very original !
All the best
Dany
Awesome thanks Dany!
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: New Giraffe (Sept 8)

Post by matthewlai »

Guenther wrote:
matthewlai wrote:This one only has some small changes. No 100+ Elo jump like last time.

https://bitbucket.org/waterreaction/gir ... 150908.zip

Changes:
1) Score reporting is now scaled down by 1/10. Hopefully that will play better with score-based adjudication systems. Score magnitudes are now more inline with what other engines report, though it's still probabilistic.
2) Time management has been improved (thanks to Graham Banks' testing that led me to notice this). About 33 Elo gained.
3) Gaviota tablebases now supported. See https://bitbucket.org/waterreaction/giraffe for details. I want to switch to a neural-network based endgame system later, but EGTB will do for now.

By the way, Giraffe is now open source, and the thesis is now available online. See thread on the programming forum if interested - http://talkchess.com/forum/viewtopic.ph ... 136#640136

This is probably going to be the last Giraffe release for a while. I have just submitted and defended my thesis. I think my brain needs a break from working on Giraffe full time for 3 months.

Will be travelling for a few weeks before getting back to Giraffe.

Thanks
Hi Matthew,

I played a few games against it now after I followed its reported progress since the first release and I must say I am really surprised how good it plays openings now
compared to the first few released games I have seen here. My congratulations it is already quite strong against Humans at least.
Memory usage seems hardcoded to be around 300MB right?
What would be the commandline param for WB to always load the Gaviota bases at start?
I noticed that the eval/meval files still are ascii - can you gain something later by a binary approach? (I guess the files will grow quite a bit when you do more training/learning?)

Guenther
Thanks!

Yes, hash table is hardcoded at 256MB, and it also uses some memory for neural networks, eval cache, etc. The hash table size actually doesn't matter too much for Giraffe, because it's a very slow searcher. It already takes forever to fill up 256MB.

There are no commandline params (in Giraffe) for loading Gaviota, but it's exposed to the GUI as a standard option, so the GUI should be able to remember settings and apply them every time. I may add a command line option (or more likely, a configuration file) if it's more convenient, once I get back to my development machine (unless anyone wants to submit a patch).

I have tried using a binary file. It would be slightly smaller, but load time is about the same, and I'd have to worry about things like endian-ness, etc, so I decided to not bother.

It won't become bigger with training. The file sizes are basically neural network sizes. The numbers will just become more accurate as training progresses.
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.
Daniel Anulliero
Posts: 772
Joined: Fri Jan 04, 2013 4:55 pm
Location: Nice

Re: New Giraffe Engine (not Working)

Post by Daniel Anulliero »

supersharp77 wrote:Matthew Lai

New Giraffe Engine (not Working)
supersharp77 wrote:
Loads in w32 and tried to run used 300 k ram (but no moves made)
has eval.net file and meval.net files enclosed but no ini file......no config file
and no readme file......also no opening book......just how is this chessengine
supposed to work?..where do you set the parameters? and wheres the cookbook/instructions?? LMAO Thx..........AR

What GUI are you using and the engine communication log?

eval.net and meval.net need to be in the working directory of the engine. There is no config file. No options to configure. Also no opening book. It's good enough to play openings without a book.
----------------------------------------------------------------
No parameters you say Mr Developer? (Even the flawed 'Brainless' had a config file)
So pray tell........how does one set the ponder and the ram without parameters?? Does your "artificial intelligence" decide that all by itself?
Thx AR...... :D
Mr ross is full of "very kinds" commentaries ...
Better you Post nothing about the great Mathew's work ...
Have you tried to wrote an engine ? Do you know how dificult is?
Play chess guy and "basta" !! 😆😆
Henk
Posts: 7251
Joined: Mon May 27, 2013 10:31 am

Re: New Giraffe Engine (not Working)

Post by Henk »

matthewlai wrote:
Henk wrote: Does your engine improves itself too ? So if it plays many games it plays better and better without human intervention.
Only in special training mode that's basically highly optimized self-play. Not during normal gameplay.
Why not ? Everybody who downloads your engine gets an original one after playing many games. Or would they all converge to the same values of the parameters.
User avatar
hgm
Posts: 28395
Joined: Fri Mar 10, 2006 10:06 am
Location: Amsterdam
Full name: H G Muller

Re: New Giraffe Engine (not Working)

Post by hgm »

supersharp77 wrote:So pray tell........how does one set the ponder and the ram without parameters??
Through the GUI, of course! You must live in pre-historic times that you think such things would require a config file...

As long as I can remember GUIs have a checkbox that you can tick to switch pondering on and off.

If Giraffe would ever implement pondering or a variable-size hash table, you can be pretty sure it still would not use a config file. Of course in this special engine memory use might be dominated not by the hash table but by the learned eval parameters.
supersharp77
Posts: 1265
Joined: Sat Jul 05, 2014 7:54 am
Location: Southwest USA

Re: New Giraffe Engine (not Working)

Post by supersharp77 »

H.G.Muller
Location: Amsterdam Posted: Sun Sep 13, 2015 11:34 New Giraffe Engine (not Working)
supersharp77 wrote:
So pray tell........how does one set the ponder and the ram without parameters??
Through the GUI, of course! You must live in pre-historic times that you think such things would require a config file...
As long as I can remember GUIs have a checkbox that you can tick to switch pondering on and off.

If Giraffe would ever implement pondering or a variable-size hash table, you can be pretty sure it still would not use a config file. Of course in this special engine memory use might be dominated not by the hash table but by the learned eval parameters.
------------------------------------------------------------
No box to check in the Fritz gui running the engine as winboard to uci
setup with limited ram.....sure I would love to give kudos to the developer like you guys.....but fact is...the engine is not working...yet
and his last engine needed special pyrotechnics to get working also (aka Brainless)..... :wink:
Daniel Anulliero
Posts: 772
Joined: Fri Jan 04, 2013 4:55 pm
Location: Nice

Re: New Giraffe Engine (not Working)

Post by Daniel Anulliero »

supersharp77 wrote:H.G.Muller
Location: Amsterdam Posted: Sun Sep 13, 2015 11:34 New Giraffe Engine (not Working)
supersharp77 wrote:
So pray tell........how does one set the ponder and the ram without parameters??
Through the GUI, of course! You must live in pre-historic times that you think such things would require a config file...
As long as I can remember GUIs have a checkbox that you can tick to switch pondering on and off.

If Giraffe would ever implement pondering or a variable-size hash table, you can be pretty sure it still would not use a config file. Of course in this special engine memory use might be dominated not by the hash table but by the learned eval parameters.
------------------------------------------------------------
No box to check in the Fritz gui running the engine as winboard to uci
setup with limited ram.....sure I would love to give kudos to the developer like you guys.....but fact is...the engine is not working...yet
and his last engine needed special pyrotechnics to get working also (aka Brainless)..... :wink:
Seems giraffe is not working... just for you lol
Fix that instead of your "kinds" answers
Are you a chess programmer ? Where is your great engine?
Play Chess and "basta"!
😆😤