Not logged inRybka Chess Community Forum
- - By tturgut (**) Date 2009-07-31 04:38
Is the default function on Monte Carlo analysis the best?

I asked this question earlier, and if I remember correctly Vas said that the higher the number of games, the more accurate the evalution is.

Is this the case, or increasing the depth, braching etc, would increase the accuracy? (and if this is the case, increase to what settings?)
Parent - - By M ANSARI (*****) Date 2009-07-31 05:50
It depends how much time you spend.  The more depth the less games you have and the more time you need.  A lot also depends on the position, is it a complicated middle game or is it a technical endgame.  I have used Monte Carlo a few times where it found some very deep only moves to save a game, it is quite powerful once you learn how to use it but most likely it is still not mature as an application and will be much better with Rybka 4.
Parent - - By Vasik Rajlich (Silver) Date 2009-08-03 06:58
The data about this is limited but I'd stick with default settings (and low depths). # of games seems to be more important than quality of games.

Parent - By tturgut (**) Date 2009-08-05 10:40
I agree.

This program has some kind of learning function. The evaluations change and it reaches better conclusions by time. If it was randon, that wouldn't happen. It finds better moves with higher number of games, and plays worse moves less.

I know Vas said there is no learning function, but it is hard to explain the better moves and more accurate analysis if this was random.

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill