docs/10-ranking.md

In ppernot/ErrView: Analysis of error sets

Ranking module

Plot ranking probability matrices [1].

Controls

• Plot type:

  • Levels: plot a color-coded matrix

  • CI: plot 95% confidence intervals of the ranks

• Statistic: select ranking statistic (MUE, Q95 or MSIP)

• Bootstrap M/N: test of M-out of-N bootstrap sampling to avoid uncertainty underestimation [2]. The value 1 corresponds to M = N.

• Label size: adjust methods label size in plot

Plot

Plot of the ranking probability matrix.

References

1. P. Pernot and A. Savin (2020) Probabilistic performance estimators for computational chemistry methods: Systematic Improvement Probability and Ranking Probability Matrix. I. Theory. J. Chem. Phys. 152:164108. http://dx.doi.org/10.1063/5.0006202

2. P. Hall, H. Miller et al. (2009) Using the bootstrap to quantify the authority of an empirical ranking. Ann. Stat. 37:3929–3959. https://doi.org/10.1214/09-AOS699

ppernot/ErrView documentation built on Jan. 30, 2022, 6:59 a.m.