LoginSign Up
← Back to all sparks
M

mev

ANALYTICS
Velocity0.0

R package mev by lbelzile — release notes from GitHub.

github.com/lbelzile/mev ↗

An extreme-value toolkit reorganised its whole API into prefixed families and tripled its estimator count.

extreme-value-theorythreshold-selectionstatistical-estimationapi-redesignr-package
◆Current state
mev provides likelihood-based inference for univariate and multivariate extreme value models — threshold selection, shape estimation, tail dependence and max-stable simulation. Version 2.0 was a deliberate reorganisation: every threshold-selection routine now carries a thselect. prefix, every stability plot a tstab. prefix, and every extremal-dependence measure an xdep. prefix, with the old names deprecated but mostly still working. The same release added a large batch of estimators — Stein-weighted GPD, roughly a dozen shape estimators, second-order regular variation, L-moment GPD and Weissman quantiles.
◆Where it's heading
The package is consolidating into a reference implementation of the extreme-value literature rather than a collection of one-off routines. Sixteen threshold-selection methods now share standardised arguments and their own plot and print methods with automatic selection, which is the tell: the goal is comparability across methods, not just availability. Dependency reduction runs alongside, with distribution functions written in-package to drop evd and Rsolnp replacing nloptr in earlier releases.
◆Prediction
Version 2.1 continued adding threshold-selection routines within the new naming scheme, so the next release most likely follows the same pattern — more estimators fitted to the established prefixes, plus fixes to the 2.0 renaming. The entries give no sign of a further structural change.

◆Recent moves

  1. 9mo ago

    Two more threshold-selection routines slot into the new scheme

    The first release built on 2.0's reorganised API, and it behaves exactly as that structure invites: thselect.egp and thselect.cbm are new methods dropped into an existing family rather than new machinery. The fixes clean up 2.0 fallout, including thselect.mdps returning the shape parameter rather than its reciprocal so it agrees with its siblings.

    View source ↗
  2. 9mo ago

    Threshold, stability and dependence functions regrouped under prefixes

    ⚡ SPARK

    The release that gives the package its current shape. Everything after it is an addition within the naming scheme this release imposed, which is why 2.1 could add threshold methods without any structural work.

    View source ↗
  3. 2y ago

    Boundary-case likelihood fixes, bundled with the prior release's notes

    This entry carries two releases in one body — 1.17's fixes plus 1.16's Pickands U-statistic shape estimator and new wave-height, pandemic and nutrient datasets. The fixes concentrate on boundary behaviour at xi = -1 and on profile likelihood intervals being truncated to admissible values, the kind of correctness work that precedes a major reorganisation.

    View source ↗
  4. 3y ago

    GEV and GP distribution functions brought in-house to drop evd

    Implementing the d/p/q/r functions directly removed the evd dependency, and swapping nloptr for Rsolnp did the same on the optimization side. Dependency trimming that 2.0 would later extend by vectorising those same distribution functions over all their arguments.

    View source ↗
  5. 4y ago

    Four max-stable families, fixed parameters and threshold diagnostics

    A broad feature release covering multivariate and univariate work at once: four new max-stable model families for simulation, fixed-parameter support in the maximum likelihood fitters, and several threshold-selection diagnostics. The threshold routines added here are the ones 2.0 would later fold into the thselect. family.

    View source ↗