On Mon, 11 Nov 2013, Reginald Beardsley wrote:
Marc's suggestion fits very nicely with what I'm trying to do at the moment.
I sometimes get terms which constitute much less than 1% of the total result.
Those are what I want to drop. Doing so requires recalculating the other terms
to minimize the error.
Of course you will want to recalculate.
If you have fifty one-percenters,
you will want to recalculate your solution.
If you have a thousand one-percenters,
you will want to recalculate your threshold.
Marc's works if you already know your threshold.
I may want to do something fancier eventually, but right now I'm still trying
to develop an appropriate mathematical model for the physics.
That said, there's another problem for which what is suggested here looks very
attractive. I've attempted to solve that problem several times over the course of
20+ years without ever finding a satisfactory solution. It involves finding the
derivative of a series irregularly sampled in x with truncation errors in both x
& y. When the data are finely sampled in x the truncation to 3-4 digits in y
leads to wild swings in the derivative. To date the only solution has been
smoothing based on ersatz heuristics. Good enough, but not aesthetically
satisfying.
You might want to consider doing something related to splines.
The values at the knots would have ranges rather than fixed values.
In the case of a cubic spline, you might want to minimize the
sum of the squares of the third derivitives or something.
--
Michael [email protected]
"On Monday, I'm gonna have to tell my kindergarten class,
whom I teach not to run with scissors,
that my fiance ran me through with a broadsword." -- Lily
_______________________________________________
Help-glpk mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/help-glpk
