recursive_feature_elimination: Recursive Feature Elimination

In elpidiofilho/labgeo: Collection of functions to fit models with emphasis in land use and soil mapping

Description

This function performs an recursive feature elimination using the rfe function from caret package. This implentation of RFE uses parallel processing

Usage

recursive_feature_elimination(df, sizes = c(2:5, 10), index = NULL,
  nfolds = 5, repeats = 1, fun = caret::rfFuncs, method_rfe = NULL,
  cpu_cores = 6, metric = ifelse(is.factor(df[, 1]), "Kappa",
  "Rsquared"), seeds = NULL, verbose = TRUE)

Arguments

df	dataframe, with income variable in the first column
sizes	A numeric vector of integers corresponding to the number of features that should be retained. Default = c(2:5,10)
index	Users cross validation folds. Default = NULL
nfolds	Number of folds to be build in cross-validation. Default = 10
repeats	repeats
fun	Default = rfFuncs , get importance values from Random Forest model.
method_rfe	regression/classificiatio method to be used when fun = caretFuncs
cpu_cores	Number of CPU cores to be used in parallel processing. Default = 6. For avoid parallel execution set this parameter to zero.
metric	metric used to evaluate model fit. For numeric outcome possible values are ("RMSE", "Rsquared) and for categorical outcome ("Accuracy","Kappa")
seeds	seeds
verbose	print results and execution time of function

Details

This function implements backwards selection of predictors based on predictor importance ranking. The predictors are ranked and the less important ones are sequentially eliminated prior to modeling. The goal is to find a subset of predictors that can be used to produce an accurate model. The web page http://topepo.github.io/caret/recursive-feature-elimination.html#rfe has more details and examples related to this function.

details

Author(s)

RFE by Max Kuhn

Elpidio Filho, [e-mail]

Examples

## Not run: 
rfe(df,sizes = c(2:10,12,15,20), cpu_cores = 4)

## End(Not run)

elpidiofilho/labgeo documentation built on May 14, 2019, 9:35 a.m.