Author: Pete R.
Date: 14:16:32 02/02/00
Go up one level in this thread
On February 02, 2000 at 16:40:12, Robert Hyatt wrote: >On February 02, 2000 at 14:39:18, Pete R. wrote: > >>I ran both Fritz 5.32 and Crafty 17.7 to 16 ply, and while they agree that Bg5 >>is best they do not eval a clear win until you feed it to them. If you give >>them 1. Bg5 Nc5, Crafty finds the Bxd8 Nxa4 O-O-O +- continuation before Fritz >>does. But neither one of them sees this from the root position. This is yet >>another example of a position where the search depth is sufficient to cover the >>solution, but the programs don't see it until you give them a hint. >>Specifically the position after 1. Bg5 Nc5 takes 12 ply for both Crafty and >>Fritz to see the win, but since I ran them both to 16 ply, they should have seen >>the win from 2-ply back, but didn't. Of the programs I have, Hiarcs 7.32 is >>fastest to see the win (and eval it as such) in the position after Bg5 Nc5, but >>is as blind as the others to see this winning line from the root. >> >>In theory alpha-beta should cover this, i.e. a 16 ply search should find >>something that a 12 ply search finds from a position 2 ply down the road, but >>reality is different. This is another example where a bunch of shorter searches >>backsolved would be superior to a deep search from the root. > > >One possiblility is that Nc5 is not the best reply to Bg5? That will make this >behave as you explain. Actually I was going to post a followup with a corrected title to the effect that Crafty handles this much better than Fritz. Somewhere between 12 and 16 ply Crafty saw Bg5 Nc5 was bad news, and switched to Qxg5 instead. That still loses, but the eval is less than with Nc5.
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.