Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Hello from Edmonton (and on Temporal Differences)

Author: Vincent Diepeveen

Date: 14:59:04 08/05/02

Go up one level in this thread


On August 05, 2002 at 17:53:09, Sune Fischer wrote:

Play 500 games with TD and see how optimistic you are then.

Or make a huge evaluation and toy a lot with book learning which is
a similar problem.

>On August 05, 2002 at 17:18:24, Vincent Diepeveen wrote:
>>>It will adjust _all_ the weights that make a contribution in the eval.
>>
>>I know, but that is exactly the problem.
>>
>>*that* isn't working, because it doesn't know what it is adjusting,
>>so it doesn't draw the right conclusions at all. in fact an infinite
>>run of TD learning will only by random luck manage to find out
>>what a good parameter set it, that's exactly the problem here.
>>
>>Drawing conclusions in chess is a problem anyway, because results of
>>a game are not always determined whether something is better or worse.
>
>I don't think you understand what TDLeaf is doing at all, which is why you are
>so pessimistic about it, to you it is a mysterious black box and you don't
>believe in this 'magic'?
>Maybe I'm too optimistic, but I won't toss it out the window without trying it.
>I think very few people with the required knowledge in _both_ areas have tried
>it.
>
>>  THE REAL PROBLEM IN CHESS:
>>
>>Suppose it happens that the tuner (TD neural network or whatever
>>as long as it's optmizing a bunch of parameters at the same time)
>>have by accident chosen a starting set where
>>*all* my parameters tuned very well except the open file
>>bonus. Instead of positive +0.50 it has put it to negative -0.50.
>
>That would be an interesting challenge for the finished tuner, and I will do the
>test when I can. But I just don't see how you can ever conclude you have very
>well tuned parameters, in fact this is _the_ problem we are trying to solve.
>I would judge it by its strength only, if it achieves higher rating than before,
>then it's stronger and I would conclude that the values before where _not_
>optimal.
>
>>In nowadays chess that means a sure defeat.
>>
>>So the learner will draw the wrong conclusion, because the
>>next run where it tunes another x parameters wrong, the randomness
>>of the position makes the defeat less sure.
>>
>>It is a trivial fact in chess that if you play for random positions that
>>the chance you win or draw is bigger than with a good program with one
>>huge problem, because the randomness of a position is having a small
>>chance to confuse the opponent, because the loss by natural induction doesn't
>>apply.
>>
>>to explain this: if 2 players try to achieve nearly 100% the same thing then
>>obviously if 1 thing is completely *dead* wrong, you lose chanceless.
>>
>>If 2 programs are *completely different* from each other then this chance
>>is less.
>>
>>It's here where it is clear that domain dependant knowledge is required.
>>
>>No if you make an autotuner you are not going to 'guide' it each run of it.
>>you have a tuner out of LAZYNESS. Because you can do yourself a better job
>>anyway.
>
>I can't say if a rook is 5 pawns, 5.2 pawns, 5.6 pawns. Is a double pawn 0.8
>pawns or 0.73 pawns etc.?
>I have no idea what values to give, not even to 0.3 pawn accuracy.
>
>If I setup a position and the static score is off, of those 41 parameters that
>contribute, how do I find those that matter and how much to adjust them?
>
>-S.



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.