mescla.uncertainty.resampling.jackknife_endmembers

mescla.uncertainty.resampling.jackknife_endmembers(endmember_samples, samples, sd_samples=None, estimator=None)[source]

Leave-one-out uncertainty from the replicates defining each end-member.

Drops one replicate of one end-member at a time, recomputes that end-member’s mean composition, and re-runs the whole mixing calculation. The spread of the resulting fractions is uncertainty that comes from the definition of the end-members, which is the part that usually dominates.

Parameters:

  • endmember_samples (dict of {str: array_like}) – One entry per end-member, holding its replicate analyses (n_replicates, n_species). Every entry must use the same species, in the same order as samples.

  • samples (array_like, DataFrame or ChemTable)

  • sd_samples (array_like, optional)

  • estimator (callable, optional) – estimator(endmembers, samples, sd_samples) -> ratios. Defaults to weighted constrained least squares. Pass a lambda wrapping mix_ml() to jackknife the maximum-likelihood fit.

Returns:

dict – ratios (the full-data estimate), sd, lower, upper (mean +/- 1 sd), replicates (all leave-one-out estimates) and n_replicates.