Author: Will Singleton
Date: 11:41:39 02/05/99
Go up one level in this thread
On February 05, 1999 at 02:28:35, David Blackman wrote: >On February 03, 1999 at 17:05:33, jonathan Baxter wrote: > >>I think you can only get so far with TD learning and then you have to look at >>ways to get tactically stronger (which really means get deeper). TD learning is >>great for learning the eval, but the eval only predicts positional things, it >>can't even predict simple tactics like a potential fork. After a certain point >>your program becomes fundamentally limited by its tactical strength, no matter >>how strong it is positionally. > >>Cheers, >> >>Jon > >Are you sure TD learning can't learn tactics? I would have thought it could. >TD works by comparing eval for shallow and deep searches, doesn't it? And if the >deep search finds a big difference to the shallow one, the difference is because >of tactics. > >Of course you have to have factors in your eval that can be combined in some way >to predict tactics, because all you're doing is optimising the way the combining >is done. Are you talking about chess or backgammon? Seems that in backgammon, TD can result in play that optimizes the chances for success in a given position, but it is based on factors other than pure tactics, ie repetition and the roll of the dice. It just *looks* like good tactics. Note that TD has a tough time "bearing off", and that calculating algorithms are used at that stage of the game. In chess, where probabilility doesn't play a role, and the interaction of the pieces is much more complex, and the opportunity for training over hundreds of thousands of games is impossible, it would seem to me that TD learning would be limited to tuning positional eval parameters. Will
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.