RE: the AD in VisAD

Hi Andrew,

On Tue, 25 Mar 2003, Wilson, Andrew, J wrote:
> OK, so for the purposes of my education, lets start with a
> very trivial example. What would be the best, most generic,
> most re-usable way for me to extend the existing
> functionality to enable average and variance to be computed?

I am CC'ing to visad-list, since this is the sort of technical
question that should go to the list.

I wouldn't extend FieldImpl or FlatField, but would write a
static method in some class to do this, much like the methods
in visad.math.Histogram and visad.math.FFT. Given a Field
argument to the method, you could call its getValues() or
getFloats(), then add up the double or float values and
compute the means for each column (i.e., each RealType in
the Field range). Similarly for the variances. You'd probably
want to test values for missing (represented by NaNs, which
fail a 'values[i][j] == values[i][j]' test), and exclude them
from your means and variances. Such a method wouldn't be much
code. You could return the means and variances as simple
double or float arrays, or as a RealTuple.

Note that different samplings of the same function would give
different means and variances, so you might want to precede
the computation by a resample() call to a uniform gridded
sampling.

Good luck,
Bill