Author: Ian Osgood
Date: 13:08:06 02/16/00
Go up one level in this thread
On February 16, 2000 at 06:28:36, Tom Kerrigan wrote: >On February 16, 2000 at 05:18:27, Steve Maughan wrote: > >>A couple of comments: >> >>1) Hardware has GREATLY improved over the last ten years. Palm Genius is >>running on a 16 MHz DragonBall. Richard says that this is the equivalent to > >Yes, I fully realize the hardware difference. But TSCP has a trivial evaluation >function. It's not well tuned and it's missing critical knowledge. A >world-champion program should be able to beat it, even if it has a huge hardware >advantage. Another thing is that TSCP was only searching 4 ply for the second >game, which translates to a few thousand nodes. So the difference isn't nearly >as big as you might think. I think the big difference is that Palm Genius uses a selective search. Isn't it a known phenomenon that dumb but complete searchers will find the holes in a smart but selective search? This is the reason usually given that Genius is usually at the bottom of the group of top programs on the SSDF. If you are testing against the unregistered version of PalmGenius, be aware that it is limited to 1 sec/move, which occasionally will obtain 1-3 ply searches instead of the average 3-5 ply search. I am curious also: how does TSCP do against the other simple public domain chess program, SCP (the ancestor of GnuChess and PocketChess)? I suspect your eval isn't as primitive as you think! :) >Besides, the hardware difference does not explain the bizarre moves, or why it >was beaten handily by my 1500 friend. > >I suspect that Genius either has some bugs, or Lang tried to rewrite it from >scratch and it will take some time to mature. > >-Tom This is a common complaint against the entire Genius line of programs: at blitz speeds they play mostly solid positional chess, with the occasional gaffe. The cause is the selective search Lang uses. Genius only generates a handful of moves for itself and *all* the responses for the opponent (which is why the PV advances in depth by 2 each iteration). This leads to conservative play, and the occasional "blind spot". I think Lang used his Roma engine unmodified because 1) it was already written in 68000 assembly, 2) it was designed for slow hardware, 3) it was the last of his dedicated programs not to use a hash table, and 4) by using the engine unmodified, he could claim that PalmGenius is a world champion program (which it was in 1987). One should remember that Roma was sold in 1987, when *all* processors were slow. The design of his search takes that into account. I have found that at faster speeds (2 sec/move) PlamGenius is competitive with a more modern portable unit, the Sapphire II. His selective search gets deep enough quickly enough to offset the Sapphire's hash tables (not so effective at short depths). This advantage seems to dissipate at longer time controls. Also, the opening book of the registered PalmGenius is still rather small compared to the Sapphire II, a shortcoming I expect to be resolved in the Pro version. Ian
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.