Author: Michael Henderson
Date: 11:01:36 09/14/04
Go up one level in this thread
On September 14, 2004 at 09:43:06, Stuart Cracraft wrote: >On September 12, 2004 at 23:37:05, Michael Henderson wrote: > >>On September 12, 2004 at 23:29:12, Michael Henderson wrote: >> >>>On September 12, 2004 at 22:58:22, Stuart Cracraft wrote: >>> >>>>I did this and the results are worse. >>>>Clearly something is wrong >>>> >>>>I took the triangular table, once the iteration finishes, >>>>and grabbed pv[0][0..N] to get the pv. I then made each >>>>of these moves on the board and used my hash store function >>>>to store the move that's in pv[0][i] from 0 through N into >>>>the hash table. The reason I made them was so that the hashkey >>>>was calculated since it is stored from a global value into >>>>the table when a store is done. After all this, the pv[] moves >>>>made is unmade and the position is back where it was. >>>> >>>>Since depth, a score, and whether it is an upper/lower/exact >>>>are required, I threw in some ideas, but nothing worked well. >>> >>>I also made the mental error of thinking you had to do this...but you don't! >>>You store the depth relative to the depth of the iteration you just did. >>>Storing the store is optional -- if you can get the score relative to the root >>>position score after making moves. What I do is store no score and no bound, so >>>nothing happens. I only store the best move from the PV and the depth. >>> >>>good luck, >>>Michael >> >>actually the score to store would just be the root score because it was backed >>up from leaf to root node and is the value for every position in the PV. Stored >>with exact bound. > >I tried this and it dropped my results from 249/300 on Win at Chess >to 78/300. did you remember to negate the score based on the side to move? i.e. Variation 1 is worth +500, but in black's case the score would be -500. Putting in 500 for every entry will mess it up badly! Michael
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.