Using requirements or env.yml

Using requirements or env.yml

This comes up often when moving between Python projects managed with pip and those managed with Conda. Let’s clarify:

requirements.txt

• Standard in Python projects.

• Used by pip.

• Lists Python packages (and optionally versions).

• Example:

  numpy==1.26.0
  pandas>=2.0.0
  scikit-learn

Install with:

pip install -r requirements.txt

env.yml (or environment.yml)

• Used by Conda.

• YAML format describing the full environment:

  • Python version
  • Conda channels
  • Conda packages
  • Pip dependencies

• Example:

  name: myproject
  channels:
    - conda-forge
    - defaults
  dependencies:
    - python=3.11
    - numpy=1.26
    - pip
    - pip:
      - some_pip_only_package==0.3.1

Create environment from it:

conda env create -f env.yml

Update an existing environment:

conda env update -f env.yml --prune

Export an existing environment:

conda env export > env.yml

Key differences

• requirements.txt → lightweight, only lists pip packages.
• env.yml → full specification of environment, includes Python version, Conda channels, and both Conda + pip packages.

When to use

• Use requirements.txt if:

  • Project is pip/venv-based.
  • You want compatibility with tools like pip install or deployment on services that expect pip.

• Use env.yml if:

  • You want to reproduce a full Conda-managed environment (exact Python version, Conda dependencies).
  • You work in data science projects where binary compatibility matters (e.g., numpy, pandas, pytorch).