Re: [ChurchBass] [NBC] Numerical methods (Reply)
Posted by: famkan <famkan@...>
> >They want a system to determine the trends in orders over the last few
> weeks, and predict what they will require in the coming weeks.
and peter replied...
> An interesting problem indeed. I'm no statistician, but I'm thinking some
> sort of averaging scheme is needed. I picture an imaginary window over
the
> time scale that contains past order quantities. The value you would plot
> for the centre of the window is a simple average of all the values in that
> window. You may want to vary the size of the 'window' depending on how
> responsive you want your trend analysis to be: wider window for a
smoother
> trend curve, narrower window for more of a roller coaster feel. Just some
> thoughts from a non-statistician geek type wannabee programmer.
Aaah. simple averages don't take into account decling trends until its too
late, and you have too much stock to have to store, and wait for orders to
come in to clear up the over manufacturing. However, applying a polynomial
of can also have problems, particularly at high orders (>3??), where they
just go off in their own sweet world of Very Large Numbers.
mathworld.wolfram.com has been my friend. (BTW, for a geeky, yet very
funny articles check out The Strong Law of Small Numbers, and the Frivolous
Theroem of Arithmetic)
What I'm currently thinking is kind of like what you're thinking, I'm
thinking... Its a Lagrange polynomial fit weighting the points in bias of
more current results. Where's Magnus when you need him!?!
p "lostinmathematicsanditswaytoolateforthat" k