Author: Marc Bourzutschky
Date: 11:46:30 01/11/03
Go up one level in this thread
On January 11, 2003 at 11:38:51, Rafael Andrist wrote: >On January 11, 2003 at 10:45:14, GuyHaworth wrote: > >>I haven't studied the Nalimov code but if the 'subgame reverse-inheritance' >>phase is changed as follows: >> >>a) side mated ==> DTC = DTM = 0 >>b) stm converts to win ==> DTC = 1 rather than DTM = dm+1 >>c) stm forced to convert to loss ==> DTC = 0 rather than DTM = dm >> >>then Eugene's code generates DTC EGTs rather than DTM EGTs. >> >>Maybe the EN-program-cognoscenti know where this coding is? > Actually, KT does not use c) above, which may have been the source of the confusion. He explains this in the Dec. 1996 ICCA Journal. What it boils down to is that I believe c) ends up being stored as dm-1 in some cases. An example is the position: [D] 8/8/6Qk/8/8/8/8/2R4K b This would be dtc=0 using c) above, but KT has it as DTC=10, while DTM=11. >Changing the code would not be a problem, but it would be necessary to >_generate_ the EGTB again with DTC metric and this means re-generating already >existing information. A direct conversion would be much faster. > >>This seems a safer route to creating the DTC EGTs. > >If you do not trust the Thompson data, several cross-checks are possible after >conversion: > >1) win(DTC) = win(DTM) etc. comparing illegal positions is not possible because >Thompson metric does not make a difference between draw/illegal. > >2) DTM >= DTC > >3) running the self-consistency check for EGTB > >All this together would still be faster than re-generating the database. > >regards >Rafael B. Andrist I have some code that might be cobbled together to do the conversion. However, I agree with Guy that it might be easier to just generate DTC from scratch, especially since the Thompson set is incomplete (particularly missing endings with more than one pawn) and does not have wtm. I'm also not sure how many people still have the Thompson data. -Marc
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.