Re: Dimini****ng Returns in Game Engineering

"Anton" <anton.txt@[EMAIL PROTECTED]
> wrote in message 
news:Xns9AC4CB7BE3B12antontxtgmailcom@[EMAIL PROTECTED]
 Anton,

> Skill = log(Experience)
>  It is neither multi-dimensional  nor  consists of several variables...

The term 'log' in the above example is a bit ambiguous.
Are we taking the log with a base of e, 2, 10, or some other value?
Log_10 (1000) is 3. This means 10*10*10 = 1000.
This is just the degenerate case of X*Y*Z.
A square is still a rectangle, and a cube is still a box.

If we assign a base (lets just use b and fill in the value later):
Skill =  log_b (Experience) means b^Skill = Experience.
Or b*b*b*...(Skill times) = Experience.
And perhaps this is the correlation you speak of,
where the same dimension is being used multiple times.

Instead of using b*b*b -- a cube,
wouldn't it be more interesting to use b*c*d -- a box?

>  Another assumtion of yours that I don't like is the  existence
>  of a  single  fitness  function  that  summarizes  a  unit's
>  effectiveness.
Such as Character Level, Armor class, Damage output, etc.
Perhaps you'll invent a replacement for these.

>  In a good model the  variables  are  not  correlated  and  you
>  graphs loose all sense.
Here is where I suggest a small change:
"and your graphs lose all _common_ sense".
True: Common sense does not work for this problem space.

>  And replacing naturally analogous values by  booleans  is  bad
Unless you multiply many of them together.
Because when multiplying many of them together,
the result is that max dwarfs everything else.
In other words, the game winds up having only one good armor/gun/whatever.

> P.S.: Deciding about the number of stats before designing a game
Putting the cart before the horse are we?
Deciding about the number of stats _is_ designing the game,
and in my opinion it is a step up from pound-and-observe.