AlphaGo's evaluation function

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

Kappatoo
Posts: 15
Joined: Sat Sep 23, 2017 10:33 pm

Re: AlphaGo's evaluation function

Post by Kappatoo »

How you can evaluate without knowing the result and how you know the result without a search?
Sorry for being dense, but I'm not sure what you mean. The result of what?
User avatar
fern
Posts: 8755
Joined: Sun Feb 26, 2006 4:07 pm

Re: AlphaGo's evaluation function

Post by fern »

Sorry if I was not clear. What I mean is that if any investigation, so to say, you can at once selectan best option as a matter of fact, that is, you already know what is best, that means that previously a sear was conducted precisely to make possible that a priori knowledge. As in ending tables.
Hope this time I make myself clear.
PK
Posts: 893
Joined: Mon Jan 15, 2007 11:23 am
Location: Warsza

Re: AlphaGo's evaluation function

Post by PK »

@Kappatoo: my impression is that I read/calculate much more while playing go, even though I am clearly better at chess. So please take my explanations with a grain of salt. However, big difference between both games is that in Go I do not calculate in a single tree. Rather, I need a couple of "what if" scenarios for playing or not playing in different regions of the board. Furthermore, if a local sequence is not a game-changer, I need to estimate who gets to play the first move elsewhere (it's best, if not always viable, to maintain initiative).
Kappatoo
Posts: 15
Joined: Sat Sep 23, 2017 10:33 pm

Re: AlphaGo's evaluation function

Post by Kappatoo »

fern wrote:f any investigation, so to say, you can at once selectan best option as a matter of fact, that is, you already know what is best, that means that previously a sear was conducted precisely to make possible that a priori knowledge. As in ending tables.

Sounds like a philosophical question. However, heuristics usually aren't based on insights derived from the very position you are currently considering. So even if they have somehow been 'learned' on the basis of some kind of search (which I believe isn't necessary), this search wasn't done on this position.
User avatar
fern
Posts: 8755
Joined: Sun Feb 26, 2006 4:07 pm

Re: AlphaGo's evaluation function

Post by fern »

well, no search in that position, maybe, but then We equally could say we consult logaritmic tables without need to search nothing in that "position" because the calculation was already made in XVIII century.
I mean, how something can be known a priori without previous experience?
Even if search does not mean specific calculations of every factor, some previous examination of structures or something similar must have been done before.
If not, please explain me how.

Fern