README.md

In wtcooper/modUtils: Collection of utils for model building and tuning.

modUtils

Collection of utils for model building and tuning.

Overview

A small collection of some utility functions I use frequently for data pre-processing and model evaluations.

Preprossing functions include:

getBasicCleanData(): Wrapper function to do many of the functions below sequentially. removeHighVIF(): Removes columns with high multicollinearity automatically based on VIF. removeZeroVar(): Rip of caret::nearZeroVar that just looks at % of most frequent, not relative to the second most frequent. convToBinary(): Converts select columns to binary. convToDummy(): Converts select factor/character columns to dummy coded columns using a data.table approach. balanceClasses(): Simple random class balancer for multinomial targets that does up- or down-sampling - you supply how many observations per class you want.

Model evaluation functions include:

getBalancedProbCut(): Returns a probability threshold that minimizes difference between sensitivity and specificity (balanced accuracy). getBalancedAcc(): Returns the balanced accuracy by choosing a probability threshold value where Specificity == Sensitivity. getAUC(): Returns AUC. getGINI(): Returns Gini. multiClassMetrics(): Custom caret:summaryFunction for multiclass targets (I think same as caret's multiClassSummary() function now). multiClassMetricsMeans(): Another custom caret:summaryFunction for multiclass targets. This one returns one-vs-all calcs for AUC and log loss (mean across all one-vs-all). getCMVals(): Returns a subset of classification metrics for binary and multiclass classification.

Installation

library(devtools) 
devtools::install_github("wtcooper/modUtils")

wtcooper/modUtils documentation built on May 4, 2019, 11:59 a.m.