Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Hash and first Fail High

Author: Ralf Elvsén

Date: 19:18:46 08/14/03

Go up one level in this thread


On August 14, 2003 at 21:55:53, Sune Fischer wrote:

>On August 14, 2003 at 21:43:42, Ralf Elvsén wrote:
>>
>>Hello
>>
>>I haven't done any chess programming in over a year now (cured :). As you say
>>yourself below, this could be a statistical fluctuation. But I'm with Uri here:
>>I remember implementing a special improvement in move ordering in my program.
>>It worked fine, the time to search to a certain depth decreased. However, the
>>measure you call 1FH got worse. I made the same interpretation as Uri, namely if
>>you produce crappy moves at one ply you can will get a lot of FHs at the next
>>and your statistics looks fine :)
>
>I think it is a different effect, because if you do have a bad move ordering, it
>is expected to be consistently bad so you won't quickly get a FH at the next
>ply.
>
>It is true however that having a bad ordering could possibly produce a much
>larger tree, and this subtree might contain a lot of easy FHs.
>E.g. at a node you have 1 FH move and 30 FL moves, now you search these 30 FL
>moves first (assuming worst possible move ordering), so that gives you the
>chance to get 30 FHs at the next ply at the expence of 1 missed FH at the top
>ply.
>
>This is what you mean I guess, but note the inherent contradiction of
>assumptions.
>
>-S.
>

Well, we are trying to understand effects on large numbers by simple examples,
sometimes it works and sometimes it doesn't. In my case (IIRC) it was like this:
I penalized moves that would put a piece on a square where it could be captured.
That meant that a lot of easy captures were weeded out and thus the statistics
looked worse. Note that it is two independent aspects of move ordering: Avoiding
giving away pieces and detection thereof if it happens. There is no
contradiction in improving the former, thereby making it harder for the latter
to shine in the statistics. I'm not dogmatic however, it was a year ago and I
didn't investigate it like my life depended on it... It was just enough make me
seriously doubt the usefulness of that number.

Ralf



This page took 0.01 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.