Author: Will Singleton
Date: 21:27:22 11/22/05
[d]5r1k/1pq4p/p3Q3/8/6R1/7P/6P1/6K1 b - - This position illustrates two effects of differing methods of rep draw scoring. The first method is used by many programs, including (apparently) Fruit, in which the program scores a position found in the search that is equal to one in the game history as a draw. This method ignores the fact that making the move which gives 0.00 does not actually constitute a draw, and the game can continue with any score possible. The second method is to assign a draw score only to those positions reached in the search or the game history that actually represent real draws. The problem here is that the program, due to horizon effects or other circumstances I don't fully understand, will choose to play a useless repetition before making the move which it could have (should have) played in the first place. Problems here include wasting time, plus sometimes incorrectly assessing the rep draw, plus looking kinda stupid. In the above position, black should probably play Qd8. But my program, being a "second method" program, somehow likes the line Qc1+ Kh2 Qc7+ Kg1 and then Qd8. If you analyse with Fruit, it gives a draw score after the rep, which is fairly silly since it has to admit its error after Qd8. But which is sillier, Fruit's error or Amateur's? And, for those who subscribe to method 2, will your program duplicate Amateur's pseudo-rep or play Qd8 directly? Will ps Here's another interesting tidbit about Fruit (and perhaps other method 1 progs). Put the position 5r1k/1p5p/p3Q3/8/6R1/7P/6PK/2q5 b - - into Fruit. It will want to play Qc7+ Kg1 Qd8 etc. But, if you preload the hash with the above rep (that is, paste the original position into Fruit, then using analyze mode in winboard force Qc1+ Kh2, thus arriving at the tidbit position), you find Fruit will want to play the ridiculous Qb2. Some might say that's not fair play, since I've loaded the hash with a position Fruit would not have played by itself. Still, one cannot make that claim if assessing analysis capabilities. btw, Ruffian behaves the same as Fruit.
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.