Skip to main content

EDAV Fall 2021 Mon/Wed Community Contributions

54 Linear Association R Studio Plugin

Ruilin Liu and Zhifeng Zhang

54.0.1 Introduction

In this community contribution project, we write a simple R addin project that helps us find out the linear association between feature columns and a target column in a data frame. The idea actually comes from problem sets 2 and 3 where we need to find if there is correlation between two variables. There are indeed several ways to do so like pair scatter plots and mosaic plots.

However, we want a simpler and more efficient way to do so. As learnt from statistic class, p-value in linear regression tests the null hypothesis that the coefficient is equal to zero. The lower the p-value, the stronger contribution of a feature column to the target column.

Thus, we develop such idea that we can write a simple R addin that fits a linear model onto the target column and return top n feature columns that has the lowest p-values, or highest possibilities of having correlation with target column.

54.0.2 Installation

Link: https://github.com/gzmason/linear_association_rplugin/tree/main

  1. Download the files in the above repo, or download the the zip file LinearAssociation.zip and unzip it.
  2. Open the project file LinearAssociation.Rproj in Rstudio.
  3. On the top right panel under Build sub panel, click Install and Restart.
  4. Check the Addins menu if there is an addin called Find Linear Association under LINEARASSOCIATION. If so, you are done installing our addin.

54.0.3 Usage

54.0.3.0.1 Note: This addin runs on library ‘tidyverse’. Please run the following command first before running into next step!

library(tidyverse)

This addin will take three arguments in a line, separated by commas:

  • name of a data frame
  • name of a target column
  • number of columns, referred as n

and returns n number of feature columns that have highest probability of having linear correlation with target column.

Tips:

  • This addin tolerates any white-space between and after arguments.
  • The first two arguments are required, whereas the third argument n is optional. If not provided, n is 5 by default.
  • The data frame has to be stored in the environment first, as this addin reads data frame from current environment variables.
  • n will be reset to 5 if any non-integer or negative value is passed.
  • n will be capped at maximum number of feature columns if a greater number is passed.
  • Currently, the target column can only contain continuous and numeric data, since we are using linear regression. Categorical target column is not supported!
54.0.3.0.2 Example

I am using SleepStudy example from PSet3 as an example.

And the console output would be:

54.0.4 Troubleshooting

  1. “str_split” not found

Solution: load tidyverse package library(tidyverse)

  1. Only showing one row of “NA”

Solution: change the target column to a continuous one. This addin does not support categorical target column as we are not doing logistic regression here.

54.0.5 Evaluation and Improvement

We have learnt a lot during the process of creating a small R addin project. At first it might seem a bit scary to write an addin since both of us have no clues what is going on here. But after reading some tutorials on how write an R addin, we find it rather simple and enjoying. It is needless to say that creating an R package of our own is of huge sense of accomplishment, though we have been through lots of trials and errors along the way.

There are a number of ways that this little project can be improved. For example, adding a logistic regression feature to categorical target column. Evaluating more metrics regarding linear association rather than a simple p-value to achieve more comprehensive results.