Skip to contents

PEVmax

Source: vignettes/articles/criterion-pevmax.Rmd
criterion-pevmax.Rmd

PEVmax is the largest prediction error variance among the test individuals, so it protects the worst-predicted one.

Type Model-based
Direction Smaller is better
Uses a test set yes
Score a given set pevmax_score()
Search with it optTrain(method = "PEVmax")

Definition

PEVmax=max⁡ix0i′Bx0i\mathrm{PEVmax} = \max_i x_{0i}' B x_{0i} with an intercept column and B=(X′X+λI)−1B = (X'X + \lambda I)^{-1}. Without a test set the candidates left out of the training set are the target.

Use in TSDFGS 

library(TSDFGS)
data(geno)
pevmax_score(geno[1:50, ], geno[51:100, ])
## [1] 12487101
fit <- optTrain(geno, cand = 1:404, n.train = 50, method = "PEVmax",
                 min.iter = 300, console = FALSE, seed = 1)
fit$OPTtrain
##  [1] 324 310 171 218 283  73 150  81 226 354 337  88 314 313 165 372 180 243 220
## [20] 339 227 373  68 335 292 239 262  78 371 247 268 189 298 312  70 194  79 208
## [39]  75  11  56 118  42  82 156 321  30 237  33 230
# targeted: the test set is known
fit2 <- optTrain(geno, cand = 1:300, n.train = 50, test = 301:404,
                  method = "PEVmax", min.iter = 300, console = FALSE, seed = 1)
fit2$OPTtrain
##  [1]  78 129  96  28 216 277  35 293 179  43 186  27 152 296 161  68 103 142  70
## [20]  32 137 110 113 149 290 201 164  14  80  40 189 283 260 262 101  74 202  38
## [39]  61  18  33 230 157 246 141  31 197 256 225 108

The search is a random-start exchange (hill-climbing) search; see How the scores are computed. Use min.iter larger than 300 on real data and check fit$converged.

Original work

Akdemir, D., Sanchez, J. I., & Jannink, J.-L. (2015). Optimization of genomic selection training populations with a genetic algorithm. Genetics Selection Evolution, 47, 38. https://doi.org/10.1186/s12711-015-0116-6

Isidro y Sánchez, J., & Akdemir, D. (2021). Training set optimization for sparse phenotyping in genomic selection: a conceptual overview. Frontiers in Plant Science, 12, 715910. https://doi.org/10.3389/fpls.2021.715910

Repository or website of the original authors: R package STPGA by Akdemir and colleagues: CRAN.R-project.org/package=STPGA (GPL-3).

License review

Formulas and algorithms are ideas and methods, which copyright does not protect. Where an original package is under a GPL license, we read it to check our results and credit it; TSDFGS is GPL-3, which is compatible with GPL-3 and GPL (>= 3) code. This is our reading of the public record, not legal advice; see the license review.

  • Original software: STPGA, GPL-3 (PEVMAX); compared, not copied.
  • Article license: Akdemir et al. (2015) CC BY 4.0; Isidro y Sánchez and Akdemir (2021) open access.
  • Verdict: included.

Reviewed 2026-11-04. All criteria and the ones left out.