mixle.utils.automatic.detectors.generalized_gaussian module¶
Generalized-Gaussian (exponential-power) continuous candidate.
Symmetric family with a tunable shape beta: beta=2 is Gaussian, beta=1 is Laplace,
beta<1 is peakier/heavier-tailed and beta>1 flatter-topped. The candidate only fires on
plausibly symmetric real-valued data (a strong skew is a clear out-of-family signature – e.g. an
exponential or other one-sided law – which the symmetric exponential-power model would otherwise
“steal” with a poorly-justified sub-Gaussian shape).