Author: Pete R.
Date: 13:11:51 07/06/00
Go up one level in this thread
On July 06, 2000 at 14:19:41, Dann Corbit wrote: >I have proposed something similar to that in the Brainy interface. > >But to make it work right, I really need to make a database interface for >Brainy instead of just reading the whole glop into memory. With a million >rows, it already consumes 30+ megabytes of ram. With 6+ megs... Hmmm. That leads to a number of thoughts. I wonder what the typical correlation would be between positions and the ECO labels of the games that produced them. Assuming a reasonable correlation you could read into RAM only those rows with an ECO tag corresponding to the ECO variation being played in the current game. Obviously just determining whether or not the current position is in the Brainybase should be quite fast even looking into a disk-based database, but for more intensive operations having likely hits in main RAM would be a plus. The ideal would be if there were a way to grab all CAP positions that could be near-future descendents of the current position and stuff the position values into the hash table in semi-permanent form as is done with Crafty's "store" feature. This way Crafty would see much further than normal, e.g. if 10 ply down it spots a position X, and this position has a CAP value based on a 12 ply search, that value will be more accurate than what Crafy can assess it as from the root. They key issue would be the slowdown due to the lookup, hence the desire to have a way to read rows into RAM that will be likely hits. >In the proposed Brainy interface, besides inquire(position) there is also >inform(position) so that you can also write back to the database when you do a >better calculation or calculate a missing result. At least for the long time >control games it could be useful. And if there were millions of programs all >running using this interface, we could expand the knowledge base very rapidly. I would be entertained by just one for now. ;) In the not too distant future I will probably buy a GHz machine, and if I can get a cable modem I'd be happy to have it playing Blitz games on ICC to see how the Brainy interface works. You can bet I'll scrutine any lost games and pound rows accordingly. :) A few such machines playing each other would be a neat experiment to grow the book, millions is probably a bit much to hope for. ;)
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.