SuperSelector

Overview

The SuperSelector package extends Cross-Validated SuperLearner (SuperLearner::CV.SuperLearner()) for ensembled feature selection. This package also contains utility functions for simulating input data for, and visualizing outputs from, its feature selection algorithm.

Installation

The SuperSelector package is currently only available via GitHub. To install:

remotes::install_github("saraemoore/SuperSelector")

Examples

Simple

From ?sim_toy_data:

Simulate data for a 3 continuous covariate, 1 binary outcome example where the outcome (Y) is unrelated to the first covariate (normally-distributed W1) but is related to the second and third covariates (lognormally-distributed W2 and W3, which are also related to one another).

library(SuperSelector)
library(SuperLearner)  # for SL.mean, method.NNloglik
library(FSelector)     # for cutoff.biggest.diff

dat <- sim_toy_data(n_obs = 1000, rnd_seed = 54321)
res <- cvSLFeatureSelector(dat$Y,
                           dat[,-which(colnames(dat) %in% c("ID", "Y"))],
                           family = binomial())

knitr::kable(res$whichVariable[,"keep_names", drop = FALSE])

	keep_names
cutoff.biggest.diff_kNA	W2

res_sum <- summarizeScreen(res$summary,
                           groupCols = "method")
p1 <- cvSLVarImpPlot(res_sum)
p1

res_sum2 <- summarizeScreen(res$summary,
                            groupCols = c("method", "screener"))
p2 <- cvSLVarImpPlot2(res_sum2)
p2

Advanced

From ?sim_proppr_data:

Simulate data for a 9 covariate, 1 binary outcome example where the outcome (Y) is unrelated to the first four covariates (normally-distributed W1:4) but is related to the remaining five covariates (some of which are also related to one another).

library(SuperSelector)
library(SuperLearner)  # for SL.mean, SL.glm, method.NNloglik
library(FSelector)     # for cutoff.biggest.diff, cutoff.k
# remotes::install_github("saraemoore/SLScreenExtra")
library(SLScreenExtra) # for screen.wgtd.lasso, screen.randomForest.imp, screen.earth.backwardprune

dat <- sim_proppr_data(n_obs = 500, rnd_seed = 54321)
dat_x <- dplyr::bind_cols(dat[, -which(colnames(dat) %in% c("GCS", "ID", "Y"))],
                          factor_to_indicator("GCS", dat))

libraryCVSLFeatSel <- list(
    `lasso mean` = c("SL.mean", "screen.wgtd.lasso"),
    `random forest biggest diff mean` = c("SL.mean", "screen.randomForest.imp"),
    `splines biggest diff mean` = c("SL.mean", "screen.earth.backwardprune"),
    `lasso glm` = c("SL.glm", "screen.wgtd.lasso"),
    `random forest biggest diff glm` = c("SL.glm", "screen.randomForest.imp"),
    `splines biggest diff glm` = c("SL.glm", "screen.earth.backwardprune")
)

libraryMetaFeatSel <- data.frame(selector = c("cutoff.biggest.diff", "cutoff.k", "cutoff.k"),
                                k = c(NA, 3, 6),
                                stringsAsFactors = FALSE)
rownames(libraryMetaFeatSel) <- c("biggest diff", "top3", "top6")

res <- cvSLFeatureSelector(as.numeric(dat$Y),
                           dat_x,
                           family = binomial(),
                           SL.library = libraryCVSLFeatSel,
                           selector.library = libraryMetaFeatSel,
                           nFolds = 5)

knitr::kable(res$whichVariable[,"keep_names", drop = FALSE])

	keep_names
biggest diff	GCS.15
top3	GCS.15 , PrehospOCY , GCS.4_to_14
top6	GCS.15 , PrehospOCY , GCS.4_to_14, Age , W1 , BMI

res_sum <- summarizeScreen(res$summary,
                           groupCols = "method")
p1 <- cvSLVarImpPlot(res_sum)
p1

res_sum2 <- summarizeScreen(res$summary,
                            groupCols = c("method", "screener"))
p2 <- cvSLVarImpPlot2(res_sum2)
p2