Chess Master 3000 is Underated
Komputer Korner
Because CM is a selective search program, it benefits more from the
last 4 years of computers speed increases than full width searchers
have. Whether this will continue in the next 4 years is anybody's
guess, but it depends on the knowledge algorithms. However based on
the Deeper Blue experiment, I would say that knowledge and selective
search is the way to go. It is interesting that Ed Schroder has cut
back on the selectivity of Rebel 10, but in general the more lines
that you can prune out the better off you should be. As speeds of
micros increase towards the speed of Deep Blue, I predict that
selective search will make another big comeback.
--
--
Komputer Korner
The inkompetent komputer
To send email take the 1 out of my address. My email address is
kor...@netcom.ca but take the 1 out before sending the email.
DJWhitfill wrote in message
<199806180340...@ladder01.news.aol.com>...
>The faster the computer, the stronger the program will be. I have
followed the
>CM seties since it started. I recently downloaded Rebel Decade 2.0
and I was
>surprised at how strong it is. Of course I am just an average player,
so I tend
>to get kicked around a little more than some.
Yesus
I once heard Chessmaster 3000 described as a "terribly weak chess program" I
don't understand this. Im expert strenth myself (uscf 2030) and I lose to
chessmaster 4of 5 games!! In my opinion it can legitimately claim atleast a
2250 rating easily. I think only a small percent of players in the world can
call chessmaster weak.
DJWhitfill
Robert Hyatt
Komputer Korner <kor...@netcom.ca> wrote:
: Because CM is a selective search program, it benefits more from the
: last 4 years of computers speed increases than full width searchers
: have. Whether this will continue in the next 4 years is anybody's
: guess, but it depends on the knowledge algorithms. However based on
: the Deeper Blue experiment, I would say that knowledge and selective
: search is the way to go. It is interesting that Ed Schroder has cut
: back on the selectivity of Rebel 10, but in general the more lines
: that you can prune out the better off you should be. As speeds of
: micros increase towards the speed of Deep Blue, I predict that
: selective search will make another big comeback.
: --
: --
: Komputer Korner
: The inkompetent komputer
I'm still not sure I buy that argument. I don't see anything that suggests
that a selective searcher makes better use of faster hardware than full-
width searchers. 30 years of experience says the opposite is true. A
selective searcher will go deeper, but for every ply of selective search,
you add to the probability of making an error by overlooking something. A
full-width searcher doesn't have this particular problem. This is why, in
1976, that full-width replaced selective search and no one has looked back
since.
--
Robert Hyatt Computer and Information Sciences
hy...@cis.uab.edu University of Alabama at Birmingham
(205) 934-2213 115A Campbell Hall, UAB Station
(205) 934-5473 FAX Birmingham, AL 35294-1170
Komputer Korner
You are forgetting one crucial thing. Even selective searchers do a
full width search to some minimum depth. As the time controls get
longer or the hardware speed increases, the selective searchers are
able to do a higher number of ply of full width as well as a deeper
selective search. Thus the errors caused by selective search go down
as speed improves. Of course the tradeoff is still there to some
extent, but programs like the King engine are a living proof that
selectivity can work and work well. Even Deeper Blue decided that 14
ply full width was enough and thus the remaining part of the search
was poured into extensions. The last match against Kasparov saw
Deeper Blue miss at least one long variation tactic but the extra
extensions more than made up for this. If micros had Deeper Blue's
speed, I predict that programs like CM5500's THE KING would give
Deeper Blue a run for it's money. It is not the same when slowing
down Deeper Blue to a micro speed as giving CM5500 1000 times the
speed hardware. With that huge speed the full width part of the search
will cover almost all the tactics and let the selectivity shine.
--
--
Komputer Korner
The inkompetent komputer
To send email take the 1 out of my address. My email address is
kor...@netcom.ca but take the 1 out before sending the email.
Robert Hyatt wrote in message <6ma4ro$rug$1...@juniper.cis.uab.edu>...
Robert Hyatt
Komputer Korner <kor...@netcom.ca> wrote:
: You are forgetting one crucial thing. Even selective searchers do a
: full width search to some minimum depth. As the time controls get
: longer or the hardware speed increases, the selective searchers are
: able to do a higher number of ply of full width as well as a deeper
: selective search. Thus the errors caused by selective search go down
: as speed improves. Of course the tradeoff is still there to some
: extent, but programs like the King engine are a living proof that
: selectivity can work and work well. Even Deeper Blue decided that 14
: ply full width was enough and thus the remaining part of the search
: was poured into extensions. The last match against Kasparov saw
: Deeper Blue miss at least one long variation tactic but the extra
: extensions more than made up for this. If micros had Deeper Blue's
: speed, I predict that programs like CM5500's THE KING would give
: Deeper Blue a run for it's money. It is not the same when slowing
: down Deeper Blue to a micro speed as giving CM5500 1000 times the
: speed hardware. With that huge speed the full width part of the search
: will cover almost all the tactics and let the selectivity shine.
: --
: --
: Komputer Korner
that doesn't change a thing. A full-width searcher still searches
more plies full-width than does the selective searcher. So that the
selective searcher still has just as good a chance of making a mistake
because selective searchers still have a fixed probability of making a
mistake in each of their selective plies...
I'm simply don't believe that they get "more" from faster hardware than
the non-selective programs. They may get just as much...
Komputer Korner
we will agree to disagree on this until definitive tests have been
conducted.
--
--
Komputer Korner
The inkompetent komputer
To send email take the 1 out of my address. My email address is
kor...@netcom.ca but take the 1 out before sending the email.
Robert Hyatt wrote in message <6mb4ch$4ov$4...@juniper.cis.uab.edu>...
bruce moreland
Only a very very small percentage of people could call you weak, too,
but as a chess program you'd be hopeless at 2030 uscf.
bruce
bruce moreland
On Wed, 17 Jun 1998 23:50:04 -0400, "Komputer Korner"
<kor...@netcom.ca> wrote:
>Because CM is a selective search program, it benefits more from the
>last 4 years of computers speed increases than full width searchers
>have. Whether this will continue in the next 4 years is anybody's
>guess, but it depends on the knowledge algorithms. However based on
>the Deeper Blue experiment, I would say that knowledge and selective
>search is the way to go. It is interesting that Ed Schroder has cut
>back on the selectivity of Rebel 10, but in general the more lines
>that you can prune out the better off you should be. As speeds of
>micros increase towards the speed of Deep Blue, I predict that
>selective search will make another big comeback.
Chessmaster <= 3000 were Kittinger engines, I believe, and the new
ones have been The King.
bruce
bruce moreland
>You are forgetting one crucial thing. Even selective searchers do a
>full width search to some minimum depth. As the time controls get
>longer or the hardware speed increases, the selective searchers are
>able to do a higher number of ply of full width as well as a deeper
>selective search. Thus the errors caused by selective search go down
>as speed improves. Of course the tradeoff is still there to some
>extent, but programs like the King engine are a living proof that
>selectivity can work and work well. Even Deeper Blue decided that 14
>ply full width was enough and thus the remaining part of the search
>was poured into extensions. The last match against Kasparov saw
>Deeper Blue miss at least one long variation tactic but the extra
>extensions more than made up for this. If micros had Deeper Blue's
>speed, I predict that programs like CM5500's THE KING would give
>Deeper Blue a run for it's money. It is not the same when slowing
>down Deeper Blue to a micro speed as giving CM5500 1000 times the
>speed hardware. With that huge speed the full width part of the search
>will cover almost all the tactics and let the selectivity shine.
The painful thing about computer chess is that it is like a religion
where there are N religious texts, N being the number of commercial
programs out there, and nobody has access to the actual contents of
these religious texts, and so anyone can draw any conclusions that
they want from any one or several of these texts, and nobody else can
argue with the conclusions because nobody else can read the texts
either.
Just say whatever you want, nobody can prove anything.
bruce
bruce moreland
>we will agree to disagree on this until definitive tests have been
>conducted.
Don't hold your breath.
bruce
Johan Havegheer
On Thu, 18 Jun 1998 17:40:10 GMT, bru...@seanet.com (bruce moreland)
wrote:
>On Wed, 17 Jun 1998 23:50:04 -0400, "Komputer Korner"
><kor...@netcom.ca> wrote:
>
>>Because CM is a selective search program, it benefits more from the
>>last 4 years of computers speed increases than full width searchers
>>have. Whether this will continue in the next 4 years is anybody's
>>guess, but it depends on the knowledge algorithms. However based on
>>the Deeper Blue experiment, I would say that knowledge and selective
>>search is the way to go. It is interesting that Ed Schroder has cut
>>back on the selectivity of Rebel 10, but in general the more lines
>>that you can prune out the better off you should be. As speeds of
>>micros increase towards the speed of Deep Blue, I predict that
>>selective search will make another big comeback.
>
>Chessmaster <= 3000 were Kittinger engines, I believe, and the new
>ones have been The King.
>
Chessmaster 2000 = Kittinger
Chessmaster 2100 = Spracklens
Chessmaster 3000 = Sprcklens wirth HT
Chessmaster >4000 : The king by J. De Koning
Johan
>bruce
Komputer Korner
Well my theory doesn't apply in this case then because the Spracklen
engines were surely full width search engines. So CM3000 must exhibit
just the same general increase in strength as all the other full
width search engines on faster hardware.
--
--
Komputer Korner
The inkompetent komputer
To send email take the 1 out of my address. My email address is
kor...@netcom.ca but take the 1 out before sending the email.
Johan Havegheer wrote in message <35895a3c...@news.uunet.be>...
Komputer Korner
It shouldn't be too hard to test this. Take 2 sets of programs that
were fairly = in strength at a period of time. One set would be
selective searchers and the other full width and then 5 years later
run the same programs on the latest computers to see if there is a
difference in strength. This can be done backwards instead of waiting
5 years from today.
--
--
Komputer Korner
The inkompetent komputer
To send email take the 1 out of my address. My email address is
kor...@netcom.ca but take the 1 out before sending the email.
bruce moreland wrote in message
<3591520b....@news.seanet.com>...
Don Getkey
In article <6mbpra$76s$1...@tor-nn1.netcom.ca>, "Komputer Korner"
<kor...@netcom.ca> writes:
>Well my theory doesn't apply in this case then because the Spracklen
>engines were surely full width search engines. So CM3000 must exhibit
>just the same general increase in strength as all the other full
>width search engines on faster hardware.
Don't be too sure "KK", as obsevered, CM3000 is highly selective, and very
fast.
As for your "theory", I tend to side with Bob Hyatt, but not for the reason he
stated.
I have observed that the pre-Pentium era CM3000, even on my P150mhz, gains very
little in strength over my P75mhz. It seems that the problem is a bit more
basic than even you were hypothesizing.
At a certain point, even for a program like CM3, very little return in strength
gains can be had, because it is so limited in actual chess knowledge. If
CM3000 doesn't know which road to go down, it will simply go down the same
wrong road, only quicker.
BTW, any USCF Expert worth his salt, should have his way with CM3 most of the
time. When I see someone say what they are rated, and that they lose more than
they win to a certain program with the "same" rating, it makes me think that
either they don't have such a rating, or that they have particular difficulty
with the style of the computer in question. CM3 is no stronger than 2150 USCF
on my P150, despite an original rating in the mid to low 2200's, on a 486/66mhz
cir.1991.
yours in chess,
Don
Coon Rapids MN USA
OHall86418
What are you basing this assumption on that any uscf expert should have his way
with chessmaster 3000? Can you point to any specific matches? Or is this just
your personal Opionion? I am rated 1934 uscf and have played numerous matches
against chessmaster 3000 and although I do beat it occasionally In a ten game
matches it usually beats me 7of 10!!! My machine is a 486/100. I have played
many experts and masters in tournament play and to me chessmaster in atleast
low master strength even on my machine
Don Getkey
In article <199806190626...@ladder01.news.aol.com>,
ohall...@aol.com (OHall86418) writes:
No, I know of no published games of CM3000 vs any USCF Expert. Just my personal
match play tests against a Master candidate USCF Expert friend of mine. Not
only does he usually win against CM3, he also kills CM4 with regularity.
As an aside, in the first Harvard Cup, none of the comments by the human
contingent even remotely suggested that CM3000 was up to actual Master
strength.
My own feeling is that CM3000 is slightly weaker than say Rebel Decade 2.0. If
I had to give CM3000 a rating, the most I could warrant is average to weak USCF
Expert level.
silverwolf
When you chess masters out there say that CM3000 is at a 2000ish rating,
under what rules do you play it under when you lose "x" number of times out
of so many games? I play it on a P75 with its default 5 second per move
deal and I can usually get ahead on material and/or position early by
changing around opening lines and making it "think"--I just wish I could
always finish the job. :)
Example: e4, e5; Nf3, Nc6; Bb5, a6; Bxf3, b6xf3 (lopez exchange variation)
then, instead of castling--which is "book," I go d2-d4 and it usually
replies by bringing out its knight to f6. Now I castle, it trades pawns on
the d file then takes my pawn on the e file with its knight and I pin it
(because of its king) with my rook and end up going up in material and
position on it. It does vary on its reply to the exchange variation, but
there are also many other openings where it makes "mistakes" like this with
the 5 second rule. It also plays stupid in end games sometimes by taking
100's of moves :) to get a mate instead of losing some material to get a
quick mate--it will still get mate, but it sure is obnoxious about material
sometimes. Anyway, under these time constraints, what would you rate
CM3000? I play it no other way. On a side note--is that chess rater on it
accurate? I spent up to 15-20 minutes studying the position on it on some
questions and got an outlandish 1670ish score. I would never spend that
much time on a move in a game!
Thanks,
Mike
silverwolf
Ax...@my-dejanews.com
In article <6mk20h$54su$1...@newssvr04-int.news.prodigy.com>,
"silverwolf" <John...@microsoft.com> wrote:
>
> When you chess masters out there say that CM3000 is at a 2000ish rating,
> under what rules do you play it under when you lose "x" number of times out
> of so many games? I play it on a P75 with its default 5 second per move
> deal and I can usually get ahead on material and/or position early by
> changing around opening lines and making it "think"--I just wish I could
> always finish the job. :)
>
> Example: e4, e5; Nf3, Nc6; Bb5, a6; Bxf3, b6xf3 (lopez exchange variation)
> then, instead of castling--which is "book," I go d2-d4 and it usually
> replies by bringing out its knight to f6. Now I castle, it trades pawns on
> the d file then takes my pawn on the e file with its knight and I pin it
> (because of its king) with my rook and end up going up in material and
> position on it. It does vary on its reply to the exchange variation, but
> there are also many other openings where it makes "mistakes" like this with
> the 5 second rule. It also plays stupid in end games sometimes by taking
> 100's of moves :) to get a mate instead of losing some material to get a
> quick mate--it will still get mate, but it sure is obnoxious about material
> sometimes. Anyway, under these time constraints, what would you rate
> CM3000? I play it no other way.
Personally I find Cm3000 rather weak even for me compared to the more modern
programs.. At 5 seconds per move..I can usually beat the program quite
easily.. I can also win up to 30 seconds per move.altough it becomes close..
Compare that to a modern program like cm5500 or virtual chess..At 5 seconds a
move it's still a challenge..(altough i can sometimes still catch it out.in
tactics)
I know everyone says that chess computers can beat 99% of all
chessplayers..But this is assuming the program is at full strenght..I howver
prefer to play chess against computers that can beat me even if i lower their
thinking time to 5-30 seconds which is nearly instanteously since I am
usually impatent to wait for programs to think..Strangely this doesn't apply
against Humans..
. On a side note--is that chess rater on it
> accurate? I spent up to 15-20 minutes studying the position on it on some
> questions and got an outlandish 1670ish score. I would never spend that
> much time on a move in a game!
I don't find chess raters accurater at all.. I get scores between 1900 to
2000+ usually on most chess raters..whether from chessmaster rating suites or
those from magazines..But I'm only 1700+ on chessnet and 1600+ on FICS..
Anyone has the same experience..
> Thanks,
> Mike
>
>
-----== Posted via Deja News, The Leader in Internet Discussion ==-----
http://www.dejanews.com/ Now offering spam-free web-based newsreading