Author: Bruce Moreland
Date: 13:12:54 07/26/01
Go up one level in this thread
On July 26, 2001 at 14:30:02, Kevin Stafford wrote: >>Ok, Joshua. My apologies to you since i didn't read correctly your question. >>BTW Where is the point in getting a "slow searcher" when everybody knows that >>the true power of a chess program is in superior tactics, not certainly in a >>better positional understanding. If I remember well, several tries to optimize >>the positional knowledge disadvantaging the search deepness gave unsatisfactory >>outcomes. >> >>Regards. > >'Everybody' seems to know this besides the chess programmers apparently. This >tactics vs. positional debate is your personal crusade only, so you really >shouldn't make it sound like a commmon understanding. A good example of why >you're wrong is fritz 6. It is a slower searcher than fritz 5, but also plays >better chess, due to a more advanced evaluation function that takes into account >more positional factors. Taken to the extreme, your logic would give us an >ultra-fast engine which simply counts material (resulting in highly tactical >play). The problem is, we've seen these engines before, and they lose to slower >engines which understand passed pawns, king safety, rooks on open files, etc. >Extra speed at a certain point has diminishing returns. The >sacrifice-all-knowledge-for-one-extra-ply approach simply doesn't work that >well. > >I'm not saying that purely positional engines are the way to go (fritz 6 is >obviously still one of the fastest engines out there), just that you seem to >misunderstand the fact that the extra speed really doesn't gain you much. An >engine that is twice as fast won't reach anywhere near twice the search depth >because the tree is growing exponentially at each step. > >-Kevin An engine that is twice as fast will find tactical shots in half the time. This is a *good* thing. But high NPS doesn't necessarily imply tactical speed, nor do low NPS necessarily imply more positional knowledge. If a program does expensive stuff in order to try to improve move ordering, nodes per second will decrease but the program will search faster. This has nothing to do with positional knowledge. Also, it's possible to add "knowledge" to an eval function that behaves like search. Here is an example: 1) Program X searches 10 plies at a very high node rate, but doesn't see two-ply tactics in its eval function. 2) Program Y searches 8 plies and can see two-ply tactics in its eval function. Its node rate is much lower because it is looking for two-ply tactics. Program Y is not smarter, it's just increasing its effective depth through eval. Sometimes it's possible to increase effective depth by a whole lot, and sometimes you can search 20 plies and have no effective depth, but a low node-rate program does not necessarily get more effective depth just because it runs at a low node rate. bruce
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.