A Chess forum. ChessBanter

Go Back   Home » Chess Newsgroups » rec.games.chess.computer (Computer Chess)
Site Map Home Register Authors List Search Today's Posts Mark Forums Read Web Partners

ply depth

LinkBack Thread Tools Display Modes
Old August 28th 03, 03:35 PM
Leonardo Ljubicic
external usenet poster
Posts: n/a
Default ply depth

I would like to comment on a relative ability of newer vs. older chess
software to spot the key moves in acceptable time frame.
I recall my Saitek Simultano on 120/40 levels, it would think for a couple
of minutes, being sure of the response found so far, then on depth of 3 it
would come up with something dramatically "improved". Seeing this, I would
think, hey, this is not such a bad chess computer after all.
Then came Mephisto Berlin (R. Lang) and Saitek RISC 2500 (J. de Koning). An
impressive positional knowledge (seemed at a time) on one side, and
sparkling attacks and tactics on the other. Still, they took a minute or two
and ply depth of 3/15 and 2/6 to come up with something "even stronger".
Now today, we have Juniors X, Shredders X, Fritzs X, Hiarcss X, Chess Tigers
of all kinds, not to mention their "deep" brothers. A change in CPU speed
pushed them towards depth 11/29 or 17/38. No doubt they are stronger by
sheer numbers. But STILL they take a minute or two, depth of 12/26 or so, to
come up with something really different!
I am puzzled. It is not that I argue that further search deminishes the
chance of sudden "discovery" in given position (although it seems so), but
my question is: what Shredder X was really doing during the first 11 plies??
How come it did not found anything dramatical between ply 4 and 11???
Looking back at engine output, the main variation stayed the same!
Can anyone please help explaing this?


Old August 28th 03, 06:44 PM
Derek Wildstar
external usenet poster
Posts: n/a
Default ply depth

"Leonardo Ljubicic" wrote in message

I would like to comment on a relative ability of newer vs. older chess
software to spot the key moves in acceptable time frame.

To help you try to answer your own question:

If you ran a significantly old program on current hardware, would you expect
it to do worse, the same, or better as a current program, on current

I suspect you would answer 'worse', since better search algorithms, move
ordering and other enhancements in newer programs have enhanced evaluations,
and pruned searches so that they can look at more relevant positions more
quickly and make better assesments.

The same comment and reply will occur 5 years from now, when future programs
on future hardware will be most likely able to brute force 8ply instantly,
and only selectively search past that, probably to a depth of 20 ply in
tournament conditions with an ELO of 3600.

Old August 29th 03, 04:54 AM
Derek Wildstar
external usenet poster
Posts: n/a
Default ply depth

"Leonardo Ljubicic" wrote in message

What is the probability of the program to change the so-far best move on
n-th ply search?

That's a good question, specifically I do not know, but clearly doing so
improves the moves made. Again, back to the original example of dissimilar
programs on similar hardware, if both have brute forced to 6-ply, the
difference in score will result in differences of evaluation possibly
including a q-search (if that's the proper abbr. for quiesence search) of
greater depth or precision. I'm sure one of the talented programmers here
can answer this much better.

How does the probability of finding the new best move changes as search

on deeper and deeper?

That's another good question, I do not know what the probability of finding
a new best move is as ply increases, clearly, intuition suggests that it
follows a decreasing harmonic cycle, to wax math-o-sophic, that first will
be observed a wild variation then significantly less options based on the
rebuttal, and so on and so forth, like the runner covering halfway the
distance to the finishline as ply increases until finally, impossibly, there
is but one move to win. Is there a hard and fast formula? The programmers
might know. I think it's a really interesting math problem.

And still: how come new programs tend to move 5 or 6 or even more ply

deeper without finding anything that would make them change their mind????

That must be entirely specific to each engines method of move ordering, once
it finds the best move, it tends to hold onto it to the top of the search
tree, or near the top on the assumption that 'if it's a good move at 5 ply,
it's probably a good move at 7-ply, so let's stick with what works for now."

For a move to *not* be good at 7-ply but *be* actually better at 9-ply, and
it's still not so good at 5-ply, would send it farther down the move order
list, so that only time enough to get it to the point, let's say 11-ply,
that the 9-ply move is vetted as better over the 7-ply best choice, hence
showing it as the higher score, again, if you want specific code examples,
I'm not your analyst, sorry.

Also, don't discount your observational bias, maybe the phenomenon you are
describing is a minoroty of positions and you're just more attuned to
picking them out, I'm not saying that's likely, but for the most part,
programs act pretty predictably, and a consistent search depth of 1-ply
deeper in produces better game results than a 1-ply shallower search, with
'superior' chessic evaluation.


Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

vB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On

All times are GMT +1. The time now is 12:30 PM.

Powered by vBulletin® Copyright ©2000 - 2018, Jelsoft Enterprises Ltd.
Copyright 2004-2018 ChessBanter.
The comments are property of their posters.