FruitFly 1.1

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

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
velmarin
Posts: 1600
Joined: Mon Feb 21, 2011 8:48 am

Re: FruitFly 1.1

Post by velmarin » Fri Jul 20, 2012 9:48 am

Executables for testing, x32 and x 64.

http://www.mediafire.com/?bpds1sg2819qfqv



Libraries that you may need:
https://sites.google.com/site/chessbouq ... -microsoft

ZirconiumX
Posts: 1327
Joined: Sun Jul 17, 2011 9:14 am

Re: FruitFly 1.1

Post by ZirconiumX » Fri Jul 20, 2012 9:53 am

Thank you very much jose!

Matthew:out
Some believe in the almighty dollar.

I believe in the almighty printf statement.

ZirconiumX
Posts: 1327
Joined: Sun Jul 17, 2011 9:14 am

Re: FruitFly 1.1

Post by ZirconiumX » Fri Jul 20, 2012 11:07 am

Well, some super quick testing of ten whole games shows that:

FruitFly 1.1 is about 100 Elo stronger (My calculator shows 200 Elo, but I know the error margin is huge) than Fruit 2.1, scoring 75%.

FruitFly 1.1 is about 300 Elo stronger (calculator = 400 Elo) than FruitFly 1.0 scoring 90% (1.0 didn't win a single game; it drew two however)

Matthew:out
Some believe in the almighty dollar.

I believe in the almighty printf statement.

Richard Allbert
Posts: 777
Joined: Wed Jul 19, 2006 7:58 am

Re: FruitFly 1.1

Post by Richard Allbert » Fri Jul 20, 2012 2:19 pm

Why not run a test of a few '000 games vs a set of oppenents, and then release if there is an improvement that is within the error margin? This would make more sense.

I'm not insulting you, btw. Just a suggestion.

If you don't have resource for such a test, I volunteer to do it for you.

Richard

ZirconiumX
Posts: 1327
Joined: Sun Jul 17, 2011 9:14 am

Re: FruitFly 1.1

Post by ZirconiumX » Fri Jul 20, 2012 4:27 pm

The minimum my GUI tests at is game in 1 minute plus 1 second per move. So 'a few 000's of game' would take around 60 hours. I don't have that kind of time.

So I gladly accept your offer of a gauntlet. Could you have FruitFly 1.1, FruitFly 1.0, Fruit 2.1 and DiscoCheck in there please? Nothing above 2900 CCRL Elo would be greatly appreciated!

Matthew:out
Some believe in the almighty dollar.

I believe in the almighty printf statement.

Richard Allbert
Posts: 777
Joined: Wed Jul 19, 2006 7:58 am

Re: FruitFly 1.1

Post by Richard Allbert » Fri Jul 20, 2012 7:45 pm

Ok, this time yes. Can you pm me the links to the executables, please?

But again, there's no point in introducing a feature that fruit 2.1 doesn't have and assuming it must be stronger when it has it. If it takes 60 hours to complete the test, then wait the couple of days until the test is complete. Even if it takes a week of starting and stopping. If the strength increase shown is positive, and outside the error margin, then you can say it's stronger.

Anyway, when I have the executables, I'll run a 1+1" test vs a range of opponents and post the results, assuming FF is stable.

Richard

Richard Allbert
Posts: 777
Joined: Wed Jul 19, 2006 7:58 am

Re: FruitFly 1.1

Post by Richard Allbert » Sat Jul 21, 2012 1:40 pm

A test is running to Compare Fruit 2.1 and Fruitfly 1.1.

Oppnents are:

Spike 1.4
Komodo 3
Gaviota 0.85
Critter 1.6
Bison 9.11
Arasan 14.3

They play 1000 games vs each, TC is 20s+0.2s

I'll post the results when done

Richard

Richard Allbert
Posts: 777
Joined: Wed Jul 19, 2006 7:58 am

Re: FruitFly 1.1

Post by Richard Allbert » Mon Jul 23, 2012 5:42 am

FF has played it's 6000 games, Fruit 2.1 has played 2000.

So far FF is -30 elo vs Fruit 2.1, test will be complete in 24 hours or so.

ZirconiumX
Posts: 1327
Joined: Sun Jul 17, 2011 9:14 am

Re: FruitFly 1.1

Post by ZirconiumX » Mon Jul 23, 2012 7:29 am

Richard Allbert wrote:FF has played it's 6000 games, Fruit 2.1 has played 2000.

So far FF is -30 elo vs Fruit 2.1, test will be complete in 24 hours or so.
I don't program for elo, I program for style. If you want elo, use Toga.

Another thing. Most of these programs are post 2009. Fruit came out in 2004. Did you really expect a Fruit type program not to be crushed to pieces?

Matthew:out
Some believe in the almighty dollar.

I believe in the almighty printf statement.

Sven
Posts: 3894
Joined: Thu May 15, 2008 7:57 pm
Location: Berlin, Germany
Full name: Sven Schüle
Contact:

Re: FruitFly 1.1

Post by Sven » Mon Jul 23, 2012 8:07 am

ZirconiumX wrote:
Richard Allbert wrote:FF has played it's 6000 games, Fruit 2.1 has played 2000.

So far FF is -30 elo vs Fruit 2.1, test will be complete in 24 hours or so.
I don't program for elo, I program for style. If you want elo, use Toga.

Another thing. Most of these programs are post 2009. Fruit came out in 2004. Did you really expect a Fruit type program not to be crushed to pieces?

Matthew:out
Hi Matthew,

I'm not sure about your feelings when you replied to Richard but I'd suggest not to feel attacked by Richard's report about his test results. He ran his test almost the same way as you had requested, except for his different choice of opponents. And since your own "quick test" report after 10 games (if I got you right) focussed mainly on the rating differences between FF and Fruit I see nothing wrong in Richard's report focussing on ELO, too.

Furthermore, -30 ELO compared to Fruit 2.1 is not like "being crushed to pieces" ;-)

If Richard's result is confirmed then it shows again that it is really hard to improve an already strong engine without very intensive testing of each change. The state-of-the-art method for that is indeed to play many games (usually implying very short TC) instead of few (long TC) games, and also to run such tests after applying single (or few) changes, not after a whole bunch of them, since in the latter case you'll never know which changes were good and which weren't. It is clear that the whole thing takes a lot of time. There is no quick way to get a strong engine.

Sven

Post Reply