Python Dependency Management

Dependency management can make or break your progress. You need a good workflow for

• Efficient Progress instead of waiting for installs or dealing with issues
• Reproducible Results due to locked environments
• Impactful Research that other people can use and build upon with ease

Part of our job as academics is to write awful code. Often it's just a one-off thing, for a paper or specific research question, so our code needs to be rigorous in terms of scientific standards, but can (and often is) quite shoddy by SWE standards. That's fair enough, no one expects production-grade code in the supplemental material of a paper (i hope). What can become an issue, though, is reproducibility. Maybe your code on your machine gave you the right results, but if it does something different on my machine, maybe due to changing package versions, our SWE problems become serious scientific problems.

Unless your name is George Hotz or you work at a company called tinygrad, you'll probably use other people's packages for your project. And that's great, why reinvent the wheel? The issue is that these packages change, and this can outright block your scientific work.

if it does something different on my machine, maybe due to changing package versions, our SWE problems become scientific problems

So here's a quick workflow, that'll make sure your projects are

• Reproducible for others and your future self
• Install seriously fast
• Don't randomly break when something updates

1. Get uv
2. Create a new folder for your project
3. In that folder, initialize a project using uv init
4. To add a dependency, e.g., seaborn or bokeh for plotting (if you only use matplotlib, definitely give them a try!), use uv add seaborn
5. Run your code either by
   1. Running it through uv uv run some_script.py
   2. Running it manually in the virtual environment
      1. Create a venv using uv venv
      2. Making sure your packages are installed uv sync
      3. Activating the venv with source .venv/bin/activate
      4. Running the script python some_script.py
6. When sharing your code, always include the pyproject.toml and uv.lock. These are essential to reproduce the exact environment and get your project running as it did on your machine.