collinear

R package to manage multicollinearity in modeling data frames.

https://github.com/blasbenito/collinear

Keywords

Repository

R package to manage multicollinearity in modeling data frames.

README.Rmd

---
output: github_document
---



```{r, include = FALSE}
knitr::opts_chunk$set(
  eval = TRUE,
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)
# options(tibble.print_min = 5, tibble.print_max = 5)
```


# `collinear` \n Seamless Multicollinearity Management 





[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.10039489.svg)](https://doi.org/10.5281/zenodo.10039489)
[![CRAN status](https://www.r-pkg.org/badges/version/collinear)](https://cran.r-project.org/package=collinear)
[![CRAN\_Download\_Badge](http://cranlogs.r-pkg.org/badges/grand-total/collinear)](https://CRAN.R-project.org/package=collinear)
[![R-CMD-check](https://github.com/BlasBenito/collinear/actions/workflows/R-CMD-check.yaml/badge.svg)](https://github.com/BlasBenito/collinear/actions/workflows/R-CMD-check.yaml)



## Warning

Version 2.0.0 of `collinear` includes changes that may disrupt existing workflows, and results from previous versions may not be reproducible due to enhancements in the automated selection algorithms. Please refer to the Changelog for details.

## Summary

[Multicollinearity hinders the interpretability](https://www.blasbenito.com/post/multicollinearity-model-interpretability/) of linear and machine learning models.

The `collinear` package combines four methods for easy management of multicollinearity in modelling data frames with numeric and categorical variables:

- **Target Encoding**: Transforms categorical predictors to numeric using a numeric response as reference.
- **Preference Order**: Ranks predictors by their association with a response variable to preserve important ones in multicollinearity filtering.
- **Pairwise Correlation Filtering**: Automated multicollinearity filtering of numeric and categorical predictors based on pairwise correlations.
- **Variance Inflation Factor Filtering**: Automated multicollinearity filtering of numeric predictors based on Variance Inflation Factors.

These methods are combined in the function `collinear()`, which serves as single entry point for most of the functionalities in the package. The article [How It Works](https://blasbenito.github.io/collinear/articles/how_it_works.html) explains how `collinear()` works in detail.

## Citation

If you find this package useful, please cite it as:

*Blas M. Benito (2024). collinear: R Package for Seamless Multicollinearity Management. Version 2.0.0. doi: 10.5281/zenodo.10039489*

## Main Improvements in Version 2.0.0

1. **Expanded Functionality**: Functions `collinear()` and `preference_order()` support both categorical and numeric responses and predictors, and can handle several responses at once.
2. **Robust Selection Algorithms**: Enhanced selection in `vif_select()` and `cor_select()`.
3. **Enhanced Functionality to Rank Predictors**: New functions to compute association between response and predictors covering most use-cases, and automated function selection depending on data features.
4. **Simplified Target Encoding**: Streamlined and parallelized for better efficiency, and new default is "loo" (leave-one-out).
5. **Parallelization and Progress Bars**: Utilizes `future` and `progressr` for enhanced performance and user experience.


## Install

The package `collinear` can be installed from CRAN.

```{r, eval = FALSE}
install.packages("collinear")
```


The development version can be installed from GitHub.

```{r, eval = FALSE}
remotes::install_github(
  repo = "blasbenito/collinear", 
  ref = "development"
  )
```


Previous versions are in the “archive_xxx” branches of the GitHub repository.

```{r, eval = FALSE}
remotes::install_github(
  repo = "blasbenito/collinear", 
  ref = "archive_v1.1.1"
  )
```


```{r packages, message = FALSE, warning = FALSE, include = FALSE}
library(collinear)
library(future)
library(parallelly)
```


## Getting Started

The function `collinear()` provides all tools required for a fully fledged multicollinearity filtering workflow. The code below shows a small example workflow.

```{r}
#parallelization setup
future::plan(
  future::multisession,
  workers = parallelly::availableCores() - 1
  )

#progress bar (does not work in Rmarkdown)
#progressr::handlers(global = TRUE)

#example data frame
df <- collinear::vi[1:5000, ]

#there are many NA cases in this data frame
sum(is.na(df))
```

```{r}
#numeric and categorical predictors
predictors <- collinear::vi_predictors

collinear::identify_predictors(
  df = df,
  predictors = predictors
)
```

```{r}
#multicollinearity filtering
selection <- collinear::collinear(
  df = df,
  response = c(
    "vi_numeric",    #numeric response
    "vi_categorical" #categorical response
    ),
  predictors = predictors,
  max_cor = 0.75,
  max_vif = 5,
  quiet = TRUE
)
```

The output is a named list of vectors with selected predictor names when more than one response is provided, and a character vector otherwise.  

```{r}
selection
```

The output of `collinear()` can be easily converted into model formulas.

```{r}
formulas <- collinear::model_formula(
  predictors = selection
)

formulas
```

These formulas can be used to fit models right away.

```{r, eval = FALSE}
#linear model
m_vi_numeric <- stats::glm(
  formula = formulas[["vi_numeric"]], 
  data = df,
  na.action = na.omit
  )

#random forest model
m_vi_categorical <- ranger::ranger(
  formula = formulas[["vi_categorical"]],
  data = na.omit(df)
)
```

## Getting help

If you encounter bugs or issues with the documentation, please [file a issue on GitHub](https://github.com/BlasBenito/collinear/issues).

Owner

GitHub Events

Issues and Pull Requests

Last synced: 10 months ago

Packages