In UBC-MDS/rb4model: rb4model

rb4model R Package

Summary

This project aims to build an R package that elegantly performs data pre-processing in a fast and easy manner. With four separate functions that will come along with the rb4model package, users will have greater flexibility in handling many different types of datasets in the wild or those collected by them. With the rb4model package, users will be able to smoothly pre-process their data and have it ready for the machine learning model of their choice.

Installation

It is recommended that rb4model be installed via devtools.

devtools::install_github("UBC-MDS/rb4model")

Functions

missing_val

This function will take in a dataframe and handle any missing values by either deleting the row, filling in the value with the average, or filling in the value with the last observation (the user will specify which method to use in the function argument).
This function will return a dataframe without missing values.

feature_splitter

This function will take in a dataframe and split the data into numerical and categorical features.
This function will return two lists, one list containing the names of the numerical features and one list containing the names of the categorical features.

fit_and_report

This function will take in data, fit a model, and calculate its training and validation scores.
This function will return the model’s training and validation scores.

ForwardSelection

This function will take in data, fit a model, and perform forward feature selection.
This function will return a dataframe with only the selected features.

Usage

This is a basic example which shows you how to solve a common machine learning problem.

library(rb4model)

Let's work with the iris dataset.

iris_copy <- iris
iris_copy[1,1]<-NA
head(iris_copy)

Use the missing_val function to fill in any missing values in your dataframe.

head(missing_val(df=iris_copy, method='mean'))

Use the feature_splitter function to split your data into categorical and numerical features.

feature_splitter(iris)

Use the fit_and_report function to fit a machine learning model of choice and report its training and validation score.

x1<- iris[1:2][1:100,]
x2<-iris[1:2][100:150,]
y1<- iris$Petal.Length[1:100]
y2<-iris$Petal.Length[100:150]
fit_and_report(x1,y1,x2,y2,'glm','regression')

Use the ForwardSelection function to perform forward feature selection on your data. Then subset your original dataframe with the selected features.

y <- iris$Species
x <- iris[c(1,2,3,4)]
ffs <- ForwardSelection(feature=x, label=y, my_mod="rf")
head(x[ffs])

Dependencies

| Package | Version | | ------------ | ------- | | mice | 3.7.0 | | stats | 4.0.0 | | testthat | 2.1.0 | | caret | 6.0-85 | | datasets | 4.0.0 | | mlbench | 2.1-1 | | randomForest | 4.6-14 | | e1071 | 1.7-3 | | plyr | 1.8.6 |

R Ecosystem

caret is probably the most widely used package for supervised learning problems in R. Although the library provides various model fitting and preprocessing features, programmers end up with writing the same line of code over and over again. Our rb4model library provides a simple solution to this pain point: wrapper functions of caret and other primary libraries used for supervised learning to reduce lines of code and promote efficiency.