Replicability

This page records notes on replicability that would otherwise be buried in technical content.

Note that if you use them in an assessment, make sure that any activation instructions are given in either README.md or the appropriate project file.

Replication in R

To install a package if it is not already installed, you can use (for the example fs):

if (!require("fs")) install.packages("fs")

If you would prefer to use a requirements.txt like solution, instead create the file requirementR.txt file with the package information:

fs 1.4.1
data.table 1.11.4

Create an R script called requirementsR.sh containing the content:

#!/usr/bin/bash
while IFS=" " read -r package version; 
do 
  Rscript -e "devtools::install_version('"$package"', version='"$version"')"; 
done < "requirementsR.txt"

Make it executable, and then install via:

./requirementsR.sh

See stack overflow for more details.

Replication in Python

Python modules are often changing. I had to make several changes to get this code from 2018 to work in 2020. This highlights the importance of documenting the libraries and python version that were used to generate a particular output.

Python handles versioning with something called requirements.txt. I have done this for this project, using the recommended way to record the status of every package:

pip3 freeze > requirements.txt

This can then in theory be exactly replicated on your system using

pip install -r requirements.txt

However, if we just do this, we may have all sorts of versioning problems if you and I have slightly different verisons of python. Instead, we create a Virtual Environment, in which we can install a specified set of packages and which we can do what we like with without messing up the rest of the system.

To do this, I followed the instructions to do this with venv

python3 -m venv project0

This creates a new directory called project0 in your working directory

source project0/bin/activate
jupyter-notebook # Run Jupyter-notebook in the virtual environment

When you are done, you tear-down the virtual environment with

deactivate

Requirements management can be done many ways and there are several tools to do it “better”. The pip freeze approach is considered harmful. Alternatives include, for example, pip-tools.