What is "randomness" for a CM9k personality?
Wilma
Wilma
Major Cat
>
> I'm sure JM has explained this before, but it's just not clear to me.
This is a _great_ question. I hope that
JM reads this...
In early versions of the product, the
documentation led the reader to conclude
that the program would randomly choose
one among several candidate moves of ap-
proximately equal desirability. My experiment-
ation indicated otherwise. Namely, from
time to time, the program would make _truly_
_inferior_ moves. It was a play strength
watering down feature that was presented
as a positive thing that just made play
more interesting with no downside whatso-
ever to it...
>
> Wilma
Regards,
Major Cat
Magnulus
moves. The more random you get, the more often it will choose suboptimal
moves.
For stronger engines it's probably unnecessary, but on weaker engines I
believe it helps to make the program a little more human playing. After
all, a weaker player doesn't have the "best" move in mind when they play a
particular move, just an assortment of good moves. Without it, you can see
the chess program tending to play the same moves over and over very
mechanicly.
Major Cat
>
> I think it does just what it says, it randomly picks among a few "best"
> moves. The more random you get, the more often it will choose suboptimal
> moves.
I am not familiar with the particulars
pertaining to Chessmaster 8, 9 or 10. Do
these newer versions just make somewhat
weaker moves or do they actually make bad
mistakes like in earlier versions?
>
> For stronger engines it's probably unnecessary,
You mean programs playing at their strongest
settings, right?
> but on weaker engines I
> believe it helps to make the program a little more human playing.
However, the _simulation_ of human play may
_still_ be far off... 8>)
> After
> all, a weaker player doesn't have the "best" move in mind when they play a
> particular move, just an assortment of good moves. Without it, you can see
> the chess program tending to play the same moves over and over very
> mechanicly.
Yes, this is the _idea_. I hope that more
recent versions of Chessmaster have success-
fully implemented it for real and not as a
bad handicapping joke...
Regards,
Major Cat
Magnulus
"Major Cat" <epik...@istar.ca> wrote in message
news:4395F4...@istar.ca...
> I am not familiar with the particulars
> pertaining to Chessmaster 8, 9 or 10. Do
> these newer versions just make somewhat
> weaker moves or do they actually make bad
> mistakes like in earlier versions?
They won't blunder with the randomness turned up, but I suppose some of
the moves they (the personalities) pick could be considered weaker. The
propensity to out-and-out blunder (leave a piece hanging) seems more related
to the "strength" setting, first and foremost. The randomness doesn't
cripple the playing strength nearly as much as reducing the search depth and
play strength (which I suspect includes a combination of evaluation
functions and also deliberate blundering).
> You mean programs playing at their strongest
> settings, right?
Yes. The stronger personalities in Chessmaster see deep enough ahead that
they don't need to pick the same bad move over and over, which is what
randomness is trying to hide (the fact that a weaker player doesn't know the
best move as deeply as a stronger players, so they pick from a collection of
weak moves depending on what they see ahead- I'm speaking from experience
here!).
Some sort of learning function in Chessmaster would be good, though. Then
it could prune out the weak moves next time it plays. Alot of the Chessbase
engines already do this.
> However, the _simulation_ of human play may
> _still_ be far off... 8>)
Do we tend to be critical of computer programs for playing "inhuman" chess
simply because we are aware they are machines. Why not have a computer
Turing test and see?
Major Cat
>
> "Major Cat" <epik...@istar.ca> wrote in message
> news:4395F4...@istar.ca...
> > I am not familiar with the particulars
> > pertaining to Chessmaster 8, 9 or 10. Do
> > these newer versions just make somewhat
> > weaker moves or do they actually make bad
> > mistakes like in earlier versions?
>
> They won't blunder with the randomness turned up, but I suppose some of
> the moves they (the personalities) pick could be considered weaker. The
> propensity to out-and-out blunder (leave a piece hanging) seems more related
> to the "strength" setting, first and foremost. The randomness doesn't
> cripple the playing strength nearly as much as reducing the search depth and
> play strength (which I suspect includes a combination of evaluation
> functions and also deliberate blundering).
This is good news then. Thanks for the heads
up.
>
> > You mean programs playing at their strongest
> > settings, right?
>
> Yes. The stronger personalities in Chessmaster see deep enough ahead that
> they don't need to pick the same bad move over and over, which is what
> randomness is trying to hide
I wish JM were around for this... The plot
thickens again! Are you saying that, random-
ness works more like an insurance policy for
weaker personalities which, otherwise, may
blunder again and again due to their impaired
position evaluation capabilities?
> (the fact that a weaker player doesn't know the
> best move as deeply as a stronger players, so they pick from a collection of
> weak moves depending on what they see ahead- I'm speaking from experience
> here!).
>
> Some sort of learning function in Chessmaster would be good, though. Then
> it could prune out the weak moves next time it plays. Alot of the Chessbase
> engines already do this.
For the unimpaired personalities, yes, without
a doubt. However, one may not want the impaired
personalities to learn that much! 8>)
>
> > However, the _simulation_ of human play may
> > _still_ be far off... 8>)
>
> Do we tend to be critical of computer programs for playing "inhuman" chess
> simply because we are aware they are machines. Why not have a computer
> Turing test and see?
This is an important, almost philosophical,
and, yet, practical issue as well. It depends
on what it is that a chess enthusiast desires
chess computer programs to do _for_ him as op-
posed to _to_ him. 8>) For example, to the
extent that I play OTB chess with other humans,
it is always in the context of some political/
artistic cafe in Montreal. Such games, of course,
lie outside the world of rated play. Now, because
of my particular interest, I am always on the
lookout for a computer program that would
_simulate_ a cafe opponent with some degree of
fidelity. My interest, therefore, lies definitely
not in training, nor even in practicing. It just
revolves around some "make believe" experience.
Goes without saying that, if , say, I were to become
fired up to try and become an IM, well, I would be
looking at chess programs through quite a different
prism.
Regards,
Major Cat
Wilma
created "Toni 2" and had them play each other several times with
alternating colors. Toni with randomness won the match hands down.
Surprised me.
Wilma
Major Cat <epik...@istar.ca> wrote in news:4395F4...@istar.ca:
Wilma
news:Xns9724B5CECA2DBwi...@207.217.125.201:
That is, Toni 2 had zero randomness.
JVMe...@aol.com
go again....
The randomness value basically adds or subtracts a number range to each
legal move's evaluation. So, let's say that this number is 10; then
each legal move's evaluation can be adjusted by any number between -10
and +10.
However, I DON'T think that it's a 1-to-1 equivalent between the actual
randomness value in the personality and the number that is used to
define the potential range of adjustment. In other words, if you have a
randomness value of 10, I don't think that means that each score will
be modified by a range of -10 to +10, or even -0.10 to +0.10. I think
it's a bit more complex than that.
But that's basically how it works. And, of course, as Johan told me,
any randomness value greater than zero will result in overall lower
quality play.
jm
Major Cat
Thank you for the explanation. It seems
to me that, over the years, this feature
got significantly fine-tuned which is a
very good thing for the program's evolution.
>
> jm
Regards,
Major Cat
Wilma
best move if randomness is zero may be given a -10 if randomness is 10,
and the sixth best move if randomness is zero may be given a +6, say.
I like it. Given that we can design our own personalities, Chessmaster
has given us tremendous flexibility in creating new personalities that
don't always play the same move in a position, and we can't easily
predict what move it will make if we give a randomness above zero.
God, I love this program. Creating opponents is my favorite pastime.
BTW, when I said that Toni with randomness beat Toni without randomness,
I had gotten confused. It wasn't randomness I adjusted, it was the
evaluation of pawns. So disregard.
Wilma
JVMe...@aol.com wrote in
news:1133921223.1...@o13g2000cwo.googlegroups.com:
Magnulus
> I wish JM were around for this... The plot
> thickens again! Are you saying that, random-
> ness works more like an insurance policy for
> weaker personalities
No, I'm saying randomness is trying to imitate that sometimes a human
player will see several good moves, and just pick one seemingly randomly,
without fully understanding. Without the randomness, the weaker engines
tend to play the same weak moves over and over.
The blunders seems random in the weaker engines, but I suspect there's a
level of randomness to when it decides to commit a major blunder, and it's a
seperate from the "randomness" setting. I suspect it's similar to how the
"friend" mode in Fritz works. The point at which the computer stops making
major blunders seems to be on the "strength" setting between 35-45, any
lower and it will commit major blunders regardless of how deep the search
depth is.
>
> For the unimpaired personalities, yes, without
> a doubt. However, one may not want the impaired
> personalities to learn that much! 8>)
All the personalities except the Chessmaster are "impaired" (and has been
shown, there are actually better "tweaks" for the King engine than the
default Chessmaster). And even most of the better "impaired" personalities
would beat 95 percent of all chessplayers. Strength is relative.
Major Cat
>
> "Major Cat" <epik...@istar.ca> wrote in message
> news:43964D...@istar.ca...
> > I wish JM were around for this... The plot
> > thickens again! Are you saying that, random-
> > ness works more like an insurance policy for
> > weaker personalities
>
> No, I'm saying randomness is trying to imitate that sometimes a human
> player will see several good moves, and just pick one seemingly randomly,
> without fully understanding.
According to JM's response regarding the
implementation of the randomness feature
in the Chessmaster family of programs, the
program does not choose one of several
"good" moves at random where "goodness"
is determined on the basis of "unimpaired"
evaluation function calculations. Rather,
it selects the "best" among a number of moves
where each one has been subjected to an
"impaired" evaluation function calculation.
In other words, the randomness pertains to
the evaluation of each move under consideration
and not to the ultimate choice among various
candidate moves.
However, the above notwithstanding, your
focusing on the _simulation_ aspects of this
feature is very interesting to me. Here is
what I see. Namely, a feature that simulates
a human OTB player who always _believes_
that each and every move of his is the "best"
while the "objective" reality may not sup-
port such....superiority beliefs! 8>)
>From time to time, posts appear rhetorically
asking why programmers have not coded things
so as to reflect the way you thought things
work... I agree with all of you gentlemen and
ask the question again. How difficult is it
to implement a randomization algorithm that
chooses at random among a number of moves that
are "near-optimal" ("eminently playable" if you
like)?
> Without the randomness, the weaker engines
> tend to play the same weak moves over and over.
And the good ones as well...
>
> The blunders seems random in the weaker engines, but I suspect there's a
> level of randomness to when it decides to commit a major blunder, and it's a
> seperate from the "randomness" setting.
It appears that the bad moves that the
program makes when the randomness feature
is on are completely "unintended".
> I suspect it's similar to how the
> "friend" mode in Fritz works. The point at which the computer stops making
> major blunders seems to be on the "strength" setting between 35-45, any
> lower and it will commit major blunders regardless of how deep the search
> depth is.
Yes, the strength feature seems to
be _the_ force behind "intended"
mistakes.
>
> >
> > For the unimpaired personalities, yes, without
> > a doubt. However, one may not want the impaired
> > personalities to learn that much! 8>)
>
> All the personalities except the Chessmaster are "impaired" (and has been
> shown, there are actually better "tweaks" for the King engine than the
> default Chessmaster). And even most of the better "impaired" personalities
> would beat 95 percent of all chessplayers. Strength is relative.
In the earlier versions of the Chessmaster
program, GM personalities exhibited some
"pet" preferences or peculiarities that
defied the norm (i.e., the default settings).
I would not consider such deviations as
strength "impairment". Rather, they are
matters of playing style very much like
the 5 generic personality styles.
Regards,
Major Cat
JVMe...@aol.com
Wilma wrote:
> I assume that all moves do not receive the same adjustment. That is, the
> best move if randomness is zero may be given a -10 if randomness is 10,
> and the sixth best move if randomness is zero may be given a +6, say.
>
> I like it. Given that we can design our own personalities, Chessmaster
> has given us tremendous flexibility in creating new personalities that
> don't always play the same move in a position, and we can't easily
> predict what move it will make if we give a randomness above zero.
>
> God, I love this program. Creating opponents is my favorite pastime.
>
> BTW, when I said that Toni with randomness beat Toni without randomness,
> I had gotten confused. It wasn't randomness I adjusted, it was the
> evaluation of pawns. So disregard.
>
> Wilma
Yes, each move is adjusted by some random value within the range of -X
to +X. This is how a move that isn't "best" can get a higher score than
the "best" move.
You've found the reason for randomness -- to create opponents that
don't always play the same move when all else is equal. However, you
should obviously never use a personality that has a non-zero randomness
for anything other than playing against, or for testing the
effectiveness of the randomness value itself; CERTAINLY not for
analysis. :-)
jm
Magnulus
I really wish Ubisoft would concentrate on developing the
non-graphical/interface aspects of Chessmaster. They would have a really
good chess program if they did so. The analysis feature just is not all
that useful in there. I should be able to set a threshold so that it will
not make comments about every single move I make- a bit of symbols or
natural language annotation would be good too.
I end up playing alot in Chessmaster and using Fritz for analysis. I do
think that the personality settings in Chessmaster produce a very fun
opponent, whereas the "Friend" mode in the Chessbase games produces a
somewhat uneven opponent. It tries to come down to your level somewhat but
it also seems to occasionally have way too much tactical brilliance, even if
it occasionally miscalculates (purposefully) exchanges.
JVMe...@aol.com
analysis. However, there is a way by which you can set any personality
to be the "mentor" personality which will be used for all analysis
features. Or, at least there is in CM9000. I don't know if the
following will work in CMX000, but here it is:
----------
Q: I would like to use a personality other than the Chessmaster
personality to analyze my games and get advice. Is this possible?
A: Yes. You can do this by editing your user's INI file, which will
likely be found in "My Documents\Chessmaster
9000\Users\USERNAME\USERNAME.INI", where USERNAME is your user name in
Chessmaster 9000. Open this file with any text editor and, in the
[mentor] group, add the line:
personality=PERS_NAME
where PERS_NAME is the name of the personality that you want to use for
all mentor features (Game Analysis, Advice, Solve for Mate and the
Mentor Lines window).
WARNING! Changing the mentor personality can very likely result in
lower quality analysis, so use this setting with care!
----------
As for a couple of things you want Chessmaster to do, can you clarify
what you are talking about when you say:
-- "I should be able to set a threshold so that it will not make
comments about every single move I make"
-- "A bit of symbols or natural language annotation would be good too."
Before I comment on those statements I need to know exactly what you
are asking for....
jm
Magnulus
<JVMe...@aol.com> wrote in message
news:1134363517.5...@g49g2000cwa.googlegroups.com...
>
> -- "I should be able to set a threshold so that it will not make
> comments about every single move I make"
> -- "A bit of symbols or natural language annotation would be good too."
>
> Before I comment on those statements I need to know exactly what you
> are asking for....
In Fritz you can change the threshold for checking the move list for
what the engine considers a blunder or less-tha-ideal move. If it falls
outside the threshold, the engine doesn't comment on it.
JVMe...@aol.com
10th, but it might be the same).
In the Game Room, just go to "Preferences -> Engine", and set the
"Blunder Alert Threshhold" to a higher value, so alerts will happen
less often.
jm