Is Your Research Software Correct?

Slides

The slides are in presentation/index.html. On your local machine, Navigate to that folder and open index.html in a browser. See the online version at https://mikecroucher.github.io/reproducible_ML/

External resources

GitHub Repos

• PCA_demo The PCA demonstration used to motivate sharing of environments
• Bob's repo Repo created from the slide motivating the use of version control

NAG Numerical Library

• NAG Library for Python

Academic Papers

• The (black) art of runtime evaluation: Are we comparing algorithms or implementations? - Demonstrating that the difference between the best and worst implementations of ML algorithms can be more than 10,000 times. Even when only considering mature frameworks
• Good enough practices in scientific computing

Version Control - Learn git

• XKCD comic discusses git
• Software Carpentry lesson on git
• Download and install git
• GitLab - GitLab. If you need to host your own in-house replacement for GitHub.
• Oh Shit! Git - Hardly anyone knows how to use all of git. What we do is use the same few commands and google our way out of trouble when things go wrong. This site covers many cases that you'll run into in practice

Environments - conda cheatsheet

Download miniconda at https://docs.conda.io/en/latest/miniconda.html

Listing and using installed environments

• conda env list - List all environments on your machine
• conda activate foo - Activate the environment called foo
• conda deactivate - Deactivate the environment you are currently using

Creating and deleting environments

• conda create --name foo Python=3.6 scipy=1.9 jupyter - Create an evironment called foo using Python 3.6, scipy 1.9 and the latest version of jupyter
• conda create --name bar --clone foo - Create an enviroment called bar that's a clone of the foo environment
• conda remove --name foo --all - Removes the environment called foo

Package management

• conda install scipy - Installs the latest version of scipy into the active environment
• conda remove scipy - Uninstalls scipy from the active environment

Creating and using environment files

Environment files contain full descriptions of environments and can be shared to allow others to use the same environment as you.

• conda env export > environment.yml - Create environment.yml describing the currently active environment
• conda env create -f environment.yml - Create an environment from the description in environment.yml

Note to self

The way you push the presentation folder to gh-pages to make the slide deck live is

git subtree presentation --prefix dist origin gh-pages