negentropy

esis.data.inversion.moses_mart.sampling_issue.moses.negentropy(G)

Return the negative entropy of sqrt(G)

I use the negative entropy because I want a function to maximize, and I actually want to minimize the entropy of spectra to concentrate the intensity into as few features (spectral lines) as possible. I view this as a rough moral equivalent to minimizing the “L0 norm”.

The rationale behind the square root is that if G is in counts, then the unit bins of the image histogram in sqrt(G) will be spaced at the 1-sigma uncertainty across all signal levels.