Author: Pham Hong Nguyen
Date: 01:52:26 08/14/04
Go up one level in this thread
On August 14, 2004 at 03:12:27, Michael Henderson wrote: >I'm having trouble understanding how the hash table creates "unsolvable" path >info problems for 50 move/rep draw detection. Could you avoid storing draw >scores, and check for draws before checking hash to solve this? Or is the >problem somewhere else and I am not seeing this...which I am sure of :) Null >move probably contributes its share of errors here too so I probably won't worry >about it. > >Michael If you search on r.g.c.c. you can find some examples and explanations of Dr. Hyatt and Bruce about this problem. The site of Bruce Moreland is also an excellent source about this. I will give you one more example: Suppose at a position you have three available moves A, B, C and they can gain scores of -10, 5, 30. When searching this position first time, suppose your engine finds that move C is a draw by rule (50 move/3 rep) so it returns 0 instead of 30. Hence, your engine thinks that the best move is B with score of 5 and save this value into hash table. Now, whenever you engine see this position again it gets a wrong score of 5 from hash table event though C is not a draw move in the new situations. From above example, you can see that your engine did not save directly the value of 0 into hash table but still got bad effects. Thus, storing draw or not is not a solution for this problem. Pham
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.