Author: Miguel A. Ballicora
Date: 17:28:59 09/15/02
Go up one level in this thread
On September 14, 2002 at 19:50:51, Omid David wrote: >I'm currently looking at the results of several algorithms on ECM test suites. I >find many positions that the program has played the correct move for wrong >reasons, without finding the correct variation at all. I know that almost all >researchers automatically count these positions as solved, but is it wise to do >so? When it comes to tactical tests, IMHO, the wise thing to do is to use a better test. In other words, to use positions were it is not possible (or very likely) to get the right solution for wrong reasons. However, it takes a lot of time to select them. >For example if one algorithm find the correct move without seeing the >combination behind it (e.g. score = -1.50), and the other algorithm finding the >whole combination (e.g. score = +7.50 or even seeing the checkmate), is it fair >to to count the position as solved in both cases?! > >In addition, there are a good number of positions in the ECM, in which better >moves than the mentioned ones, do exist; that's why IMO automatically counting >the correct moves is not enough, and a manual scan is necessary (although takes >lots of time)... ECM is known for being buggy. Giancarlo Pascutto selected in the past some positions, but still they are not 100% perfect. Yet, they are much better than the whole ECM. If you are interested, this are 30 positions that I personally selected after a while of testing from the test WCSAC (the original contains 1000 positions). They are tactical, contain only one solution, the solution is clearly the best move, if the computer is playing the solution most probably is seeing the outcome and they are not "super easy" (i.e. mate in 2). Regards, Miguel >Omid.
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.