RF: CV-optimized Random Forest

In nad7wf/RF: Run Random Forest with Different Cross Validation Strategies

Description

In plant breeding, it is common to leave one population out when training a machine learning model. This package allows the user to run random forest using one population as the test set, as well as with other common CV strategies (e.g. 5-fold, 10-fold, etc.).

Usage

RF(dat, cv_method)

Arguments

dat	a data.frame with columns "Y", "GE_ID", "Parent_A", "Parent_B", followed by one column per genetic marker.
cv_method	Either an integer, specifying the number of folds, or "family" to fold according to population.

Value

A list of two data.frames, one containing the predictions for each GE, and one containing the correlation coefficients between observed and predicted values for each fold.

Examples

library(magrittr)

set.seed(76123)
row_num <- 1000

ex <- data.frame(
    Y = runif(row_num, 30, 150),
    GE_ID = runif(row_num, 1e8, 2e8),
    Parent_A = sample(LETTERS[1:4], row_num, replace = TRUE),
    Parent_B = sample(LETTERS[5:8], row_num, replace = TRUE)
) %>%
    cbind(replicate(10, runif(row_num, 0, 1)) %>%
              as.data.frame() %>%
              magrittr::set_names(paste0("Marker", seq(10))))

result <- RF(ex, "family")

nad7wf/RF documentation built on Sept. 30, 2020, 11:17 a.m.