mescla.uncertainty.propagation.delta_method

mescla.uncertainty.propagation.delta_method(function, values, sigma, step=1e-06)

First-order propagation through an arbitrary function, by finite differences.

The general case of genereux_sigma(), for when the mixing model has no closed form – more than three end-members, or weights that depend on the data. Assumes independent inputs and local linearity: with strongly nonlinear or constraint-active problems, prefer mescla.uncertainty.montecarlo.monte_carlo_ratios(), which respects the simplex constraints exactly.

Parameters:

• function (callable) – Maps a parameter vector to an output vector.

• values (ndarray, shape (n_params,))

• sigma (ndarray, shape (n_params,))

• step (float, default 1e-6) – Relative finite-difference step.

Returns:

ndarray – Standard deviation of each output.

Return type:

ndarray