Author: Dann Corbit
Date: 14:44:28 05/26/99
Go up one level in this thread
On May 26, 1999 at 17:26:30, Robert Pawlak wrote: >Hello all, > >I was looking through some of the CAP data using Chess Assistant, and my very >quick impression was that the data seemed to indicate that all gambits were bad. It should not be too surprising to see gambits with a score of -1 or piece sacrifices with large negative values. In such cases, the one using the data will simply have to use the old brain-pan or update the data for his/her own purposes. Program's understanding is generally 99% tactial. >Now, do we know the reason for this? Is it that the comps don't understand the >positional compensation invovled? They do have some positional understanding. But not a lot. And if the payoff is more than 13 plies down the road, the program probably won't see it [that's the average depth for C.A.P.]. At this stage, the primary value will be to avoid tactical pitfalls that were not seen by humans in the initial analysis. Before long, project "connect the dots" will reveal things that are orders of magitude deeper [cough -- I believe]. >Also, weren't different programs used for analysis in the CAP project? If so, >how were the scores normalized? That is correct. The entries are normalized using depth in plies. It's unfair, but it is the best I could come up with. All "second opinions" are still retained anyway. So you can look at the alternates data and see what the other programs thought about the position.
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.