Author: Peter McKenzie
Date: 16:55:54 05/22/00
Go up one level in this thread
On May 22, 2000 at 10:03:14, Albert Silver wrote: >On May 22, 2000 at 03:22:37, Peter McKenzie wrote: > >>Recently I've noticed that my program tends to overvalue pawns, so I've been >>playing with reducing the pawn value. How does one test such a change? Not >>easily thats for sure! >> >>One thing I did was play 150 games vs the old version (new version scored 56% - >>yay), but I'm a little suspicious about how useful this information is. For >>starters the time control was fast (2min + 2sec inc, on P133), although does >>this really matter for this type of change? >> >>I guess I'll probably just go with the new version and keep an eye on its play >>on ICC... >> >>cheers, >>Peter > >The value of pawns can vary wildly in accordance with the structure most >especially. Unless the value of the actual pawns (material values) is >disproportianate in relation to the other pieces, I would look more closely at >the value of the surrounding positional factors and their interaction with each i am happy with the positional factors i have, and wish to tune my material values to be aligned with the rest of my evaluation function. In any case, the main point of my post was related to how such tuning (whether it be related to material or positional weightings) can be accurately tested. So far nobody has commented anything about the validity or not of selfplay testing! >other. In other words not just the base value of doubled pawns, but how this >will be modified when in conjunction with other aspects. Many positional >aspects/problems change in intensity depending on the presence of others (I'm >trying to be clear but this is beginning to sound like stereo instructions...). >A classic one is the pair of isolated doubled pawns on an open file. Take a pair >of doubled pawns, it's bad though how bad can depend. Isolated pawns are bad >though they can also procure greater piece freedom (Baburin wrote a whole book >just on the isolani d-pawn) but with doubled isolanis, you're in a whole lot >more trouble. Put them on an open file, and you might even be better off if you >were down those pawns (Pachman shows a great game illustrating this in 'The >Complete Strategy' Vol.2). In a sense, the value of the pawns here has changed >greatly, but what is changing the eval isn't the value of the pawns but the >positional factors surrounding it. Still, I am no programmer so I could be way >off, though the above does seem logical. > > Albert Silver
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.