Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: PB-ON vs PB-OFF (final results)

Author: Jeremiah Penery

Date: 17:25:44 10/16/99

Go up one level in this thread


On October 16, 1999 at 19:24:58, Ratko V Tomic wrote:

>>>For example, if instead of different time controls you used
>>>depth D and D+1 versions, the D+1 will guess correctly the move of D
>>>version in every single move, 100% guess rate.
>
>>
>> I don't think this is true.  I think the D+1 will ponder different moves
>> more than 20% of the time, or maybe more.  Even D vs. D would have about
>> 20% pondering difference.  This is because in D vs. D, say one searches
>> to 8 ply, then begins to ponder.  The next one searches to 8 ply, but
>> it's already one ply ahead of the other, so effectively it was like
>> a 9-ply search from the first one.
>
>You're not assuming the _fixed_ D+1-ply vs fixed D-ply rules of game.

?

> In this
>mode as soon as the given depth is computed program makes a move, it doesn't
>take advantage of pondering to geain depth in some given time. If during
>pondering it happens to reach its prescribed depth it moves instantly when its
>turn comes.


I wasn't even thinking of pondering.  I'm thinking strictly search depth, no
matter how it's reached.

>In this mode of play (strictly fixed depth), D+1 has already computed the D-ply
>opponent's move and knows exactly, every single time, what it will get in
>return, since D-deptyh subtre was a part of his D+1 tree which he picked,


I don't think this is how it works.  (I'll try to explain my reasoning here, but
I'm having trouble thinking of exactly how to do it, so I'll use more
'practical' examples.)

Let's start from the initial position, for simplicity.  You have the D-ply
search as white and D+1 as black.  Say D is 8, and its line starts 1. e4 Nc6.
So it moves 1. e4.  D+1 searches 9 ply,  (This is already like 10 ply from the
initial position.) and so it makes a different move, since it sees deeper.  It
moves 1. ...e5, and starts pondering Nf3.  D is pondering on Nc6, so it has to
search over again.  It picks Nf3, and ponders Nf6.  D+1, again seeing more,
chooses d5, and ponders Nxe5.  D has to start over again, so it chooses d3, and
ponders dxe4.  D+1 chooses d4, however, and ponders on Bg5.
This continues for a while. <Extrapolate to any middlegame position.  In
tactical positions, the difference will be more pronounced.>

This shows how a different depth search can make even the same program have many
pondering misses.  In reality, this will happen about 20% of the time. (Based on
the articles I mentioned earlier.)

> and
>D-ply oponent (otherwise identical program except for depth) has no choice but
>to pick the depth D best move. Yet, the D-ply program will still manage to win
>at least 20% of the time.


>Yes, but I find it much more interesting that there are clearly quite a few
>positions where deepening the search will make things worse.

It only makes things worse because of mis-evaluation or possibly some pruning
mistake.  It will not happen in the same position with every program.


> And that's not one
>of those "one in million" patholgical constructs, but perhaps one in 5, in your
>everyday vanilla chess positions.


I don't think it's probably quite so high.  This would be another thing for
which it'd be nice to have some test results, because I could just as easily be
wrong about it. :)

> Which means, they can be sought or created
>more systematically than just intuitively when we're lucky.

For one particular engine, this may be true.  If you can determine the long-term
weaknesses, or positional shortcomings of a particular engine, you may be able
to determine where this can happen.  But as I said, this will not be the same
for all engines in any specific position.


Jeremiah



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.