Author: Pete R.
Date: 10:08:43 09/30/99
Go up one level in this thread
On September 29, 1999 at 13:18:31, John R. Menke, Sr. wrote: >It would be very useful if chess software offered an interactive analysis >feature. It would work something like this... > >First, the proposed analysis tree is constructed by computer. Human can then >view and prune the tree before initiating analysis. Also, human should be able >to prune while analysis is in progress, if computer gets stuck in irrelevant >branches. > >This would be great for analyzing games or positions, possibly speeding up the >analysis several times compared to currect software. It also offers the >possibility for a new human+computer chess playing standard for competitions >where they are allowed to freely consult. Surely human+computer is potentially >much stronger than either alone, maybe 200+ ELO points stronger!? Isn't that >the next quantum leap for chess software that everyone is looking for? I suggested a Crafty feature idea to Dr. Hyatt that would allow this sort of thing in a different way. The big problem with computers is the horizon effect, and when analyzing something you will often follow a variation the computer thinks is promising, only to find when you get to a certain point down the road that it sucks. The fact that it sucked was too deep to be seen at the beginning of the variation. Now however if you go back to the beginning point to go for an alternate, the computer will quickly forget that the variation you just looked at ends up being bad, and you are back to square one with the computer recommending the bad variation. Transpositions to the bad variation are also a problem then. The simple idea is to allow the human operator to enter a manual assessment of a position, and have the program remember that permanantly. E.g. you believe white is winning on a certain position, so you enter a command like "assess +5". This value is stored in the hash table, so that the program now has a correct eval when it hits that position in the search. I think this will make for highly effective human-computer analysis, since the program can accept input from the human and use it immediately, and the human has tremendous flexibility to guide the search. Naturally it would be desirable to save the assessed positions from a session to a file (EPD format naturally) with a command like "assess save [filename]" and read them back in at your next session with a command like "assess load [filename]". Dr. Hyatt mentioned that this should be easy to do since his hash table entries already have a "permanence" bit, used for position learning. I don't know if that means it's on his to-do list though. ;) But I think it would be simple and very powerful.
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.