Nalimov EGTB problem related to DTM?

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
velmarin
Posts: 1598
Joined: Mon Feb 21, 2011 8:48 am

Re: Nalimov EGTB problem related to DTM?

Post by velmarin » Wed Feb 17, 2016 3:17 pm

Here refers to a table, two bishops, and a horse, it's correctly capture the horse before the 50 moves.
Perhaps there are few similar positions, but if some tournaments between engines use these tables how judges of the game, it is not known if we are doing the right.
With six pieces the risks of being wrong are greater, and the award will not be correct.

bob
Posts: 20392
Joined: Mon Feb 27, 2006 6:30 pm
Location: Birmingham, AL

Re: Nalimov EGTB problem related to DTM?

Post by bob » Wed Feb 17, 2016 7:37 pm

velmarin wrote:Here refers to a table, two bishops, and a horse, it's correctly capture the horse before the 50 moves.
Perhaps there are few similar positions, but if some tournaments between engines use these tables how judges of the game, it is not known if we are doing the right.
With six pieces the risks of being wrong are greater, and the award will not be correct.
As the # of pieces grows, DTM becomes more and more unreliable due to the 50 move rule. Trying for early exchanges while still maintaining the mate is at least an attempt to improve this somewhat. When I get around to trying this, I can play tens of thousands of games to see if it is better or worse overall.

User avatar
hgm
Posts: 23382
Joined: Fri Mar 10, 2006 9:06 am
Location: Amsterdam
Full name: H G Muller
Contact:

Re: Nalimov EGTB problem related to DTM?

Post by hgm » Wed Feb 17, 2016 8:08 pm

But an exchange in itself doesn't have to make things easier. (E.g. if you exchange B for N in KBBKN you turn a slow mate into no mate at all.) So if you have a 50-move problem now, you might have it again after the exchange. And then you should try to make as progress as possible before you exchange (but within the limits of 50-move rule), i.e. exchange as late as possible to start from a more favorable position in the next phase. So you should probably not hasten the exchange if you still have plenty of time.

This is a different issue from how the engine should use an EGT that advices it to make wrong moves, however. Normally you would cut off the branch after an EGT hit, assuming that the score it gives you is the correct one. So you would never see that it is steering you towards a draw.

bob
Posts: 20392
Joined: Mon Feb 27, 2006 6:30 pm
Location: Birmingham, AL

Re: Nalimov EGTB problem related to DTM?

Post by bob » Thu Feb 18, 2016 5:34 am

hgm wrote:But an exchange in itself doesn't have to make things easier. (E.g. if you exchange B for N in KBBKN you turn a slow mate into no mate at all.) So if you have a 50-move problem now, you might have it again after the exchange. And then you should try to make as progress as possible before you exchange (but within the limits of 50-move rule), i.e. exchange as late as possible to start from a more favorable position in the next phase. So you should probably not hasten the exchange if you still have plenty of time.

This is a different issue from how the engine should use an EGT that advices it to make wrong moves, however. Normally you would cut off the branch after an EGT hit, assuming that the score it gives you is the correct one. So you would never see that it is steering you towards a draw.
It would still be using the egtbs so it would know that is a draw... It would just prefer playing lines the egtbs say is winning, but trying for the earliest possible capture so long as the endpoint says "won".

The trick would be in the scoring. Rather than just "mate in N" as given by the EGTB, it would be "mate-in-n.fraction" where fraction is proportional to the 50 move counter. That way the farther back in the PV the capture occurs, the larger the fraction, while still keeping the same mate-in-n score...

Post Reply