Author: GuyHaworth
Date: 17:27:01 05/02/03
Go up one level in this thread
To clarify slightly, in the 'forward pass' method, the program looks at every position to see if it can set a depth or improve the possibly-interim depth already assigned. Actually, there is a good efficiency finesse here. If the 'interim' depth on cycle N is N or less, this cannot be improved if the metric is DTC or DTM. In both the forward- and retro- algorithms, when the metric is DTC, cycle N propagates losses in N-1 to wins in N. Because of this neat property that all depths set in cycle N are 'N', the bitvector approach - holding the status of a position in one bit, e.g. 'position needs depth setting to N' - can be used. In both the forward- and retro- algorithms, when the metric is DTM, _if_ you choose to delay propagating losses in N-1 until cycle N, the same bitvector approach works. If you choose to propagate losses in (DTM) N-1 as early as possible, maybe even in cycle 1, then the bitvector approach cannot be used - and it uses either 8 or 16 times as much RAM. The Wu-Beal algorithm was first used on (Western) chess endgames by John Tamplin in 2001. He also combined the algorithm with Nalimov-indexing to create Nalimov-compatible DTC EGTs. Guido seems to have cut some good code to do EGT-generation. I must confess to being a theorist onlooker and not a hands-on programmer. - guy
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.