Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: likelihood instead of pawnunits? + chess knowledge

Author: Vincent Diepeveen

Date: 19:03:09 10/25/02

Go up one level in this thread


On October 25, 2002 at 19:20:44, Ingo Lindam wrote:

>
>>I agree with Bob here. Somewhere in the program you must have a function
>>to combine all the probabilities into a comparision with an other
>>set of probabilities.
>>
>>In order to save yourselve loads of system time you could also implement
>>that function into the evaluation function, in which case you only
>>have to return an integer.
>>
>>Then you have again a chess engine working at the same way. Away probabilities.
>>Just return an integer :)
>
>Yes, at this point I have to agree as well... ofcourse to compare the
>scores/probabilities of two positions I have to obtain a single score at this
>moment. The advantage I see in following points:
>
>a) I can in different situations change the function giving my score out of P+
>and P= (and P-).

It's trivial to do that in the evaluation function. Easy to debug
compared to doing it in the search. Also 100% the same.

>b) more important for me seems the aspect of having clearly interpretable scores
>(also in the meaning of distinguishing draw positions and positions with equal
>chances)

But in the end somehow you must make your chance expectation concrete.
The evaluation function is doing that.

>c) making decisions based on statistical laws

You must see a clear distinction between statistics and search.

The search is a brute force method of correctly backtracking the
statistical estimation (evaluation function) to the root.

>And btw probabilities and integers don't have to be the opposite. You can
>ofcourse compare and add -log(p) instead of compare and multiply p.

Of course it's complete idiocy to use such functions in a chessprogram,
but apart from that, you could do that in the evaluation function as well.

>Even more important for me was the suggestion of let the computer use its self
>generated patterns derived from a huge data base of games.

If you can create an automatic evaluation function that's good and above
800 elo rating, then you can also automatically generate a human from
0s and 1s, because the big step is to make an automatic evaluation
function, the next step to apply its intelligence to mankind is
just a 5 minute transformation.

I do not see how to achieve the first goal however.

>Best regards and thank you for feedback.
>Ingo





This page took 0.01 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.