mc_train_val_multienv: Monte Carlo training/validation splits for multi-environment...

In etnite/bwardr: Brian Ward personal R package

Monte Carlo training/validation splits for multi-environment data

Description

Monte Carlo training/validation splits for multi-environment data

Usage

mc_train_val_multienv(pheno, prop_val = 0.2, cv_scheme = "CV1")

Arguments

pheno	Dataframe containing phenotypic data. The dataframe must contain genotype IDs in a column named "IID", and environment designators in a column named "ENV". The data can be replicated within environment, and can have any additional number of columns.
prop_val	Number between 0 and 1 indicating the proportion of phenotypic data that should be set to the validation set.
cv_scheme	Character string consisting of either "CV1" or "CV2". "CV1" assigns validation data by genotype (i.e. simulates introducing new genotypes which have not been tested in any environment). CV2 assigns validation data by genotype-environment combination (i.e. simulates introducing genotypes into new environments).

Details

This function only performs Monte Carlo (i.e. random subsampling) cross-validation training/validation assignment. Note that k-fold cross validation becomes difficult to perform for a CV2 scheme. If the CV2 cross-validation scheme is selected, the function will attempt to choose genotype-environment combinations to assign to the validation set such that any given genotype will only be assigned to the validation set in at most one environment. However, if the user sets a prop_val value higher than (1 / # envs), then some genotypes will be assigned to the validation set in multiple environments. Note that the value of prop_val at which this occurs will be lower for data that is unbalanced across environments. As prop_val is increased, the CV2 cross-validation scheme will begin to more closely resemble CV1, as a higher proportion of genotypes will be assigned to the validation set across all environments.

Value

A dataframe identical to the input phenotypic dataframe, except with a VAL_SET column added (if not previously present) or else re-randomized (if the column was already present) to indicate lines assigned to the validation set

etnite/bwardr documentation built on Jan. 6, 2023, 7:12 a.m.