Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Bionic vs Crafty, once again

Author: Djordje Vidanovic

Date: 08:43:10 01/23/99

Go up one level in this thread


On January 23, 1999 at 10:52:02, Robert Hyatt wrote:


>This is wrong.  The version of bionic that playedin the first 1/2 of the
>Dutch tournament matched crafty _exactly_.  Every move of every game except
>for 1 or 2.
>
>But that isn't nearly so important as one key thing they 'get'... that being
>a parallel search that no one else has.  Which gives them a 2x-3x speed boost
>over everyone else.  So to say 'it isn't crafty' is baloney.  A few eval changes
>don't make a new program.  I've also pointed out that anybody that takes the
>crafty source is _required_ to make that source public as part of the freeware
>project.   They've never done this.  IE I'd like to see a source version
>released that will _exactly_ match the Dutch tourney moves.  _then_ we could
>_know_ what is different.  They were going to do this, supposedly.  But nothing
>has been done.

Hello, Bob.

I am merely a go-between.  An intrigued and quite interested go-between, but a
go-between nevertheless.  I went over to the Bionic Home Page
(http://www.impakt.be/bionic) and here's what I read over there (I do, however,
stand by my previous observation that the Bionic I saw in action is simply very
much different from Crafty):

History
Bionic - Basic
BIONIC (Believe It Or Not, It's Checkmate !) is a chess program originally
designed by the flemish chess player Hans Secelle. Already a few months after
the first version was written, both Albrecht Heeffer and Matt Casters (author of
one of the very the first non-japanese Shogi programs) joined the author to form
the so-called "B-team" (the A-team had already retired). Albrecht continuously
provided fast hardware and Matt had the occasional 'good idea'.
At the time, Bionic was written 'from scratch' in Basic (compiled with
Quickbasic 4.5, later PDS 7.1) and was based on two famous books: 'Schaken voor
computers' (Van den Herik, Van Diepen) and 'Het computerschaakboek' (Kraas,
Schrufer, Bartels).

In 1994, Bionic entered the Dutch computerchess championship as first ever
Belgian fully competitive chessprogram. Although this version did NOT have:

iterative deepening;
null-move:
pvs-search;
hashtables;
permanent brain;
history or refutation tables;
the stalemate
3-fold repetition
50-move rules
it scored a fully deserved 3/11 pts. and got an award as "best" newcomer. (See
games archives for games scores).
Six months later, Bionic entered the 9th Aegon Tournament and scored its first
ever win against an expert class human player. It had a completely lost position
but capitalized on an overplay by his unfortunate opponent.

November 1995 saw Bionic in his second Dutch Open ComputerCchess Championship,
this time to score 4/11, the only point of interest being the win against
Impakt, the program of teammate Matt Casters.

April 1996 saw both Bionic and Impakt at the 10th Aegon tournament, were both
programs performed relatively well, scoring 2/6 though it must be said that
Impakt had the stronger opposition (IGM Ree, IM Hartoch).

In november 1996, Bionic played its third Dutch open and scored a mere 2,5/11,
having only a draw against the strong program 'Ant' (Tom Vijlbrief) to its
credit.

April 1997 featured the 11th and (unfortunately) last Aegon tournament. This
time, Bionic performed extremely poorly, scoring only half a point against Dutch
expert Nico Kuijff and losing the five remaining games. The downward spiral
continued with a disastruous 4th particpiation in the Dutch Open when Bionic
scored it's worst result ever: 1,5/11. At least 6 games were lost in the engame
due to the lack of hashtables, though positionally it showed some good sense in
its game against Rookie (Marcel Van Kervinck)

Bionic - Impakt
At this point it was decided that either the program would be written in C or
cease it's existence completely. Bionic had already been translated into C on a
1 to 1 base in a few weeks time by januari 1997, running 30.000 nps on a P133.,
but unfortunately, it contained a bug that even Chrilly Donninger could not
find. Therefore, some more drastic measures were needed. In november 1997, an
old Crafty-version (9.26) was downloaded, stripped of all it's knowledge
(evaluation, preevaluation, knowledge parameters, time control, openingbook,
endgame table-recognition, etc.) until only the search engine and part of the
pawn evaluation were left over.
Then, slowly, all knowlegde thet Bionic contained was added to the Crafty
skeleton but the first test was disastruous: Crafty 15.20 beat Bionic Impakt
3-1. Obviously, we took out much more than we put in, so some extra's were
needed.

As to the abovementioned "extra's" , I remembered a conversation that I had with
Nico Kuijff (see above) who said: "If I were to write a chess program, I would
teach it to "look" thoroughly at the position BEFORE doing anything else". Or,
as Max Euwe wrote in "Judgement and planning" : 'Make a plan that meets the
requirements of the position'

So, I started to write a completely new pre-scan routine (the details of which I
will publish in due time) that contained MORE knowledge than was left over in
Crafty's evaluation function.

When this was done, Albrecht persuaded me to port the source into a likewisely
stripped 15.20 version of Crafty, a version that supported multi-processing.
This second port was frustrating, in that all parameters had to be readjusted,
as the value of a pawn was reduced to 100 (1000 in Crafty 9.26) and all other
value's accordingly.

The results though were impressive: Bionic Impakt scored 2540 on the Louget test
(Crafty 15.20 scored 2455) on an SMP-machine that Albrecht had assembled and
which contained 2 P450-processors, overclocked to 500 MHZ, and enabling Bionic
Impakt to do 360.000 nps on average.

Belgian Fide-master Geert Vanderstricht offered to function as sparring partner
and played BI on two occasions. The first match, BI ran on a P350 and beat Geert
7-2. The second match, two weeks later, BI ran on the Dual and scored 7.5-1.5.
Thus BI arrived at a TPR of 2393, clearly approaching IM-level in playing
strenght.
--------------------------------------------------------------------------------

© 1998 by Hans Secelle & Albrecht Heeffer



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.