[argyllcms] Re: rspl weights

Gerhard Fuernkranz wrote:
could you please tell me, how must I interpret the weights of the data points for RSPL fitting? Are these the weights for the squared errors, as typically used in a regression? E.g. minimize the sum of error[i]^2 * weight[i] ? 

Good question. I'm not sure I have a clear answer. It is a weight
applied to a sum of error squared. The core maths of the regular spline
stuff is:

    Minimize penalty function P:

        P = Pe + Pxx

    Where:

        Pe = sum(interpolation_errors^2 * ken) [over data points n]

        Pxx = sum(grid_second_difference_errors^2 * kxxi) [over grid points i]
        
The kxxi values are set to be equal, in such a way as to make the
smoothness factor independent of the grid resolution.
kxxi is also multiplied by the overall smoothness factor times
a normalizing #define that sets the behaviour for a smoothness of 1.0

The ken value is normally 1.0, but can be overridden by the
per data point weight.

The ken factor is also altered when the "extra fitting" flag is set.
It increases data point weight where the interpolation error is greatest.

Hope this helps answer your question.

Graeme Gill.

Other related posts: