Author: Andrew Dados
Date: 13:23:45 10/25/99
Go up one level in this thread
On October 25, 1999 at 16:17:18, Andrew Dados wrote: >On October 25, 1999 at 14:45:29, Scott Gasch wrote: > >>Hello. >> >>I recently ran the LCTII test positions against my engine. Suffice to say it >>did badly. It solved almost all of the tactical positions rather quickly (where >>a material advantage or a mate was possible). However it did not get many of >>the others. This tells me that it is too materialistic (since it does not miss >>any ways to grab pieces) but it has nowhere near enough positional knowledge. >> >>I'd like to use the information from the ones it missed in order to better tune >>the eval. However, I do not know enough about chess to understand why one of >>the solutions is the best in some cases. Has anyone written a kind of "here is >>the reason the solution is the best move" document about this test suite or any >>test suite? >> >>I'm currently re-running the test with a larger lazy-eval and delta pruning >>window... If anyone is interested I will post the results here. >> >>Thanks, >>Scott > > There are at least 2 'positional' positions in LCTII I would consider >tactical: 'pos6' and 'pos9' (solutions e5 and Qd4, respectively). >I think turning your eval to positional part of LCTII is waste of time - >positions are pretty specific, they hardly address 'proper positional play' >issue, at least for programs of 'medium'(whatever that means) strength. I would >focus myself on such practical situations, like, for example, 2 sicillian >classic puter mistakes: avoiding e5 with black which would create weak pawn on >d6 and avoiding Nd4xf6 with white... once one has such a 'typical' positions >figured out - - it's then time to try LCTII, imo. I wouldn't tweak my eval much just to solve those, though. -Andrew- p.s. arrrgh - @#%^! browser!
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.