[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

950328: data representation

Christoph,

> Subject: data representation
> >To:       address@hidden
> >From: address@hidden (Christoph Droste)
> >Keywords: 199503281547.AA00582

In the above message you wrote:

> My data have the folloing form. There is a sole variable (the fitness)
> depending on many dimensions (from 1 to perhaps 100) which build my
> parameter space. I am searching for an optimum of the fitness in this
> parameter space. During the procedure I get a lot of points in this
> space, where the fitness is evaluated. But these points lie by no
> means on a grid, i.e. every point has its own location in every(!)
> dimension.

> Thus I have a list of n-touples
> (value of dimension1, value of dimension2, .... , fitness)
> (value of dimension1, value of dimension2, .... , fitness)
> (value of dimension1, value of dimension2, .... , fitness)
> ...
> Now my question:
> Is there any chance to store this efficiently by netCDF? Or (if not)
> do you know any other tool or generic data structure to support
> visualization of such multidimensional data?

Typically, such `sparse' data is stored as a set of independent
variables that are indexed along one (record) dimension and separate set
of dependent variables (but only one in your case) that are indexed
along the same dimension.  For example, in the following CDL:

    dimensions:
        i = unlimited ;

    variables:
        double  dim1(i);
        double  dim2(i);
        double  dim3(i);
        ...
        double  fitness(i);

the independent variables are `dim1', `dim2', etc., and the dependent
variable is `fitness'.  The `i' dimension is more of a parametric index
used to associate the individual observations.

--------
Steve Emmerson   <address@hidden>