Author: Vincent Diepeveen
Date: 06:23:06 06/21/05
Go up one level in this thread
On June 21, 2005 at 02:54:06, m.d.hurd wrote: Any form of pruning will work of course at such quick levels and testing the same engine against itself is not a good sign. For example i have statistics that if i add extra check extensions and mating extensions that cripple my search depth but make diep tactical stronger, that it will work horrible against other engines at a level of 3 hours a game. It will however win from the same diep version without those tactical extensions. Any form of rude forward pruning will work at 30 0 i suspect. 128 KB cache is a bit small. 128MB ram for each engine is better. Here is the 4 runs you should do: a) fruit 2.1 standard at 3 hours a game against several other engines. Crafty, Yace and a few of such engines. I especially mention crafty and yace here because both are very bugfree engines. b) fruit 2.1 with futility pruning and without delta pruning c) fruit 2.1 without futility pruning and with delta pruning d) fruit 2.1 with all dubious stuff turned on. I suspect that it is going to be very hard to prove that delta pruning works for fruit at longer time controls. Now the real bad news. 10 games just really says shit. You need 1000 games in total. 250 for each run. I didn't take a look how futility has been implemented in Fruit yet. But i'm expecting that at longer time controls where it gets that huge search depth anyway, that fruit+futility or fruit without both turned on will perform the best. At the blitz you play i'm very convinced that a huge search depth within those few seconds you have is important. You'll see that many games get decided when time runs out at 30 0 and then quickly getting that extra depth matters really a lot. 10 games just says nothing is the whole problem. Even when it is 10-0. It just gives zero indications on what happened. Statistical accurate testing is REAL important in computerchess. >Just for fun I set Fruit 2.1 standard versus Fruit 2.1 with Futility and Delta >pruning switched on. > >2 rounds of 5 games at game in 30 minutes. 128 kb cache for each engine using >the supplied book_small.bin > >I wanted to see how differently they played and whether the extra settings made >much difference. > >10 games is obviously not enough to say one way or another, however maybe >someone would like to test this further. > >For some strange reason I called the Fruit 2.1 with the extra settings Fruit 2.1 >vr/dp, anyway .... > >The games were run on a single AMD 2400 XP with 1 gb memory. > >Round one Fruit 2.1 vr/dp scored 3.5 and Fruit 2.1 standard 1.5 > >Round two Fruit 2.1 vr/dp scored 3.5 and Fruit 2.1 standard 1.5 > >Total 7 points out of 10 for Fruit with the extra settings. > >Fruit 2.1 standard settings never won a game. > >If anyone wants the pgn just let me know. > >Regards > >Mike.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.