Author: Steve Schooler
Date: 01:44:07 03/24/01
Go up one level in this thread
On March 23, 2001 at 15:22:59, José Antônio Fabiano Mendes wrote:
> [D]8/8/8/5N2/4kpK1/5N2/8/8 w
> White mates in 62 moves
> ++++++++++++++++++++++++++++
> [D]8/8/8/5N2/4Kpk1/5N2/8/8 w
> White plays and Black draws
Amazing. I would like to understand WHY. To facilitate my understanding, I
hope chess interface(s) are enhanced as follows:
1. As an initial task, interface will accept tablebase position and then
generate pgn readable tree identifying "optimal moves" for each side, down to
mate. When two (or more) moves are equally optimal, both variations should be
included in tree.
2. Interface will allow user to specify "tree depth". When a "short" route to
mate is anticipated, user would (typically) specify infinite (i.e. down to
mate). For a 62 mover, user might opt for an initial depth of 10-15 moves
(19-29 ply).
3. Interface will also allow user to specify "branching factor". For example,
user might specify +3 moves (5 ply). This means:
a. For "winning side", any suboptimal move move that increases the (forced)
path to mate by <= 3 moves is (also fully) explored in the tree.
b. For "losing side", any suboptimal move that lessens the path to mate by <= 3
moves is also explored.
Actually, I suspect that I'm badly oversimplyfing the options for branching
factor methods that a user will find helpful.
4. In drawn positions, since it's often helpful to also explore sub-optimal
(losing) moves in the tree, the interface should provide this as an OPTION.
5. Since inclusion of suboptimal moves exponentially affects tree size (re # 3
and 4 above), interface should allow option to LIMIT DEPTH of suboptimal
variations. For example, user might opt for SUBOPTIMAL-DEPTH of 2 ply. This
would imply that in a drawn position, all losing moves are explored to a depth
of 2 ply (i.e. present each losing move, along with opponent's optimal reply).
--------------------------
Is the above approach marketable enough to justify the effort? Ideally, the
interface + tablebases allow the user to glean positional insights that are
normally available ONLY in endgame texts. I think that if extravagant branching
methods (re # 3 and 4 above) facilitate reasonable tree pruning, then mass
appeal is plausible.
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.