[ggo-discussion] Re: glGo 1.2 beta

David C. Oshel wrote:

> So how good is the one you have to use?

I don't "have to use" it. :*) I saw a demonstration of it about a year 
ago and asked: "Can I have this for glGo??"

About how good, tough to say. I compared a couple of complex positions 
(complex at least to my poor understanding of Go) with the Pandanet 
scorer, GNU Go and a quite old but good Amiga program named Primiview. 
GNU Go only outputs the final score value, I couldn't get it to display 
territory or marked groups (Is there a way?). GNU Go is good when the 
game is almost at its end, but during the middle game the output seems 
like rolled dice. The Primiview estimator could get a couple of middle 
game positions quite right, but often failed on some dead groups who had 
only 1 1/2 eyes. Apparently it couldn't determine the half eye properly. 
  The Pandanet scorer got those dead-with-1 1/2-eyes groups correct, and 
estimation over some middle game positions looked quite reasonable. But 
I could fool all three scorers with some Go Seigen game which had a huge 
dead dragon, and neither got that right.

Feel free to experiment with it a bit. Primiview runs fine in UAE emulator.

Then there is the CGoban2 territory estimator; though it's occasionally 
useful, I overall find it's output not very valuable.


 > In other words, is it a
> parimutuel algorithm trying to fix odds on the outcome, or is it just 
> counting territory if the game stops now?

I have no idea about the implementation, I don't have the source code. 
Even if I had it, no idea if I would understand much. I had a basic AI 
class at university, but that's all AI programming I ever did.


Peter

Other related posts: