Page 20 of 21

Re: Google's AlphaGo team has been working on chess

Posted: Thu Dec 28, 2017 1:56 pm
by bhamadicharef
Have you tried to look in the source tree where is that function called what() !
https://bitbucket.org/waterreaction/gir ... at=default
Will try to re-compile when I find time ...

Re: Google's AlphaGo team has been working on chess

Posted: Thu Dec 28, 2017 2:06 pm
by bhamadicharef
A lot of the new AlphaZero versus established Stockfish reminds me of some old papers from Chellapilla and Fogel in 1999-2000 on Evolving neural networks to play checkers without relying on expert knowledge see at IEEE explorer http://ieeexplore.ieee.org/document/809083/ ... At time already the focus was to train / evolve neural network for game and without expert knowledge apart from the rules of the gaome, checkers in this case. Deep
Learning with Residual Nets are today advanced NNets of that time.
Loking forward to more detailed on AlphaZero topology to play ...

Re: Google's AlphaGo team has been working on chess

Posted: Thu Dec 28, 2017 4:12 pm
by Joost Buijs
Rebel wrote:It's still sold.

BrainMaker Professional for Windows $795

:lol:
Very nice you know that one too!
I was already playing with that package 25 years ago. I still have an old version 'BrainMaker v2.52' (written in Borland C) somewhere on my NAS. The floppies it originally came on are long time gone.

Edit:

Just tried to run it, but it does not work under Windows 10, I will keep it for nostalgic reasons.

Re: Google's AlphaGo team has been working on chess

Posted: Thu Dec 28, 2017 5:23 pm
by syzygy
Joost Buijs wrote:Just tried to run it, but it does not work under Windows 10, I will keep it for nostalgic reasons.
Linux + Wine probably offers better backward compatibility for old Windows programs :-)

Re: Google's AlphaGo team has been working on chess

Posted: Thu Dec 28, 2017 5:58 pm
by mhull
Joost Buijs wrote:
Rebel wrote:It's still sold.

BrainMaker Professional for Windows $795

:lol:
Very nice you know that one too!
I was already playing with that package 25 years ago. I still have an old version 'BrainMaker v2.52' (written in Borland C) somewhere on my NAS. The floppies it originally came on are long time gone.

Edit:

Just tried to run it, but it does not work under Windows 10, I will keep it for nostalgic reasons.
IIRC, DOS versions work under FreeDOS, maybe even Windows CMD window. These are the versions I have. I also have a Macintosh version that runs under System 7/8/9.

Re: Google's AlphaGo team has been working on chess

Posted: Thu Dec 28, 2017 6:19 pm
by mhull
Rebel wrote:It's still sold.

BrainMaker Professional for Windows $795

:lol:
CALSCI had a second generation of ISA/PCI accelerator boards that could process 2.3B NN connections per second (a Pentium 90 could do 1.9M connections per second). These boards shipped with either 64 or 128 CPUs and cost up to $8,345.

Current off-the-shelf multi-core systems could probably match those speeds these days.

Re: Google's AlphaGo team has been working on chess

Posted: Thu Dec 28, 2017 11:48 pm
by Rebel
Rebel wrote:I am trying to train Giraffe exactly how it is described here.

But it crashes,... [snip]
Anyone ever tried the training?

Re: Google's AlphaGo team has been working on chess

Posted: Fri Dec 29, 2017 4:23 am
by brianr
Rebel wrote:
Rebel wrote:I am trying to train Giraffe exactly how it is described here.

But it crashes,... [snip]
Anyone ever tried the training?
Works fine with Ubuntu for me. I could not get training to work under Win7.

Re: Google's AlphaGo team has been working on chess

Posted: Fri Dec 29, 2017 9:58 am
by Rebel
brianr wrote:
Rebel wrote:
Rebel wrote:I am trying to train Giraffe exactly how it is described here.

But it crashes,... [snip]
Anyone ever tried the training?
Works fine with Ubuntu for me. I could not get training to work under Win7.
Suspected already that it is OS related, Visual Studio debugger pointed to several WINDOWS\SYSTEM32 *.dll files.

Re: Google's AlphaGo team has been working on chess

Posted: Fri Dec 29, 2017 10:17 am
by Rebel
petero2 wrote:
Rebel wrote:
Daniel Shawul wrote:That is a start for sure -- proving a NN evaluation could be competive or even much better than a hand crafted evaluation function. The latency of evaluating the NN can be countered with a combination hardware (GPU/TPU) and software (async evaluations) which is what Google did for AlphaGo. Giraffe used only three layers of NN with chess specific inputs such as attack maps while AlphaZero used many more layers of CNN with just the rules of the game as input. Texel actually replaced its evaluation function with Giraffe's NN and showed that the eval is actually better but it would need a time odds to be competitive on the same hardware.
Statements like these could make me a believer.
The post describing this test is here.
Nice idea, a few points.

I tried the STS test and saw hardly any similarity between Giraffe (2016) and Texel GI. And so I ran the good old similarity test.

Code: Select all

Positions     8238                       Gira  Texe
{Giraffe w64 (time: 100 ms  scale: 1.0)} ----- 4.72
{Texel Gi    (time: 100 ms  scale: 1.0)} 4.72 -----
Only 4% where 65+% was to be expected?

Never seen such a low percentage, running it now at 1 second.

Last, I started a match, TC=40/60, I stopped after 40 games, 37.5 - 2.5 in favor of Texel GI while NPS favored Giraffe (2016) with approx. 20-25%.

------

Unless I have done something wrong I don't see how one can conclude the NN is on par with SF or your evaluation.