adelie.diagnostic.plot_kkt#

adelie.diagnostic.plot_kkt(lmdas: ndarray, scores: ndarray, *, idx: int | None = None, relative: bool = False)[source]#

Plots KKT failures.

This function only plots a subset of the scores. Specifically (assuming idx is not None), if there are n number of scores in scores[idx] above the threshold lmdas[idx], then these n scores as well as the n largest scores below the threshold are shown. If idx is None, then the same plot is drawn at every index as an animation.

Parameters:

lmdas(L,) ndarray: Regularization parameters \(\lambda\).
scores(L, G) ndarray: Gradient scores.
idxint, optional: Index of lmdas and scores at which to plot the KKT failures. If None, then an animation of the plots at every index is shown. Default is None.
relativebool, optional: If True, then plots the relative error score / lmda[:, None] - 1. Otherwise, the absolute values are used. Default is False.

Returns:

fig, ax: If idx is not None, then a figure and axes is returned.
anim: If idx is None, then an animation of the plots is returned.