Virtual Environments for Data Science

What is a virtual environment?

A virtual environment is an isolated Python setup for a specific project.

It keeps:

• Python version (sometimes)
• Installed libraries
• Tooling (Jupyter, linters, etc.)

separate from other projects.

Why it’s essential in data analytics

Data analytics projects often depend on:

• Specific versions of NumPy/Pandas/Matplotlib
• Jupyter
• Database drivers
• Visualization libraries

If you install everything globally, you will eventually face:

• Version conflicts
• “It worked yesterday” problems
• Broken notebooks after updates

Virtual environments prevent most of those issues.

Two popular choices

Option 1: venv (built-in)

• Comes with Python
• Lightweight
• Uses pip for packages

Option 2: conda environments

• Great for data science libraries
• Handles compiled packages easily
• Works with both conda install and pip install

Using venv (recommended for pure pip projects)

Create a new environment

From your project folder:

command

python -m venv .venv

Activate the environment

• macOS/Linux:

command

source .venv/bin/activate

• Windows (PowerShell):

command

.\.venv\Scripts\Activate.ps1

Install packages

command

pip install numpy pandas matplotlib seaborn jupyter

Freeze requirements

This creates a reproducible spec:

command

pip freeze > requirements.txt

Later someone can recreate the same installs:

command

pip install -r requirements.txt

Using conda environments (recommended for analytics stacks)

Create and activate

command

conda create -n analytics python=3.12

command

conda activate analytics

Install packages

command

conda install numpy pandas matplotlib seaborn jupyter

Mixing conda + pip safely

Sometimes a package is not available in conda.

Recommended approach:

1. Install as much as possible with conda
2. Then install remaining packages with pip

Example:

command

conda install numpy pandas
pip install yfinance

Caution

Avoid mixing randomly If you install many packages with both conda and pip in random order, you can break the environment. Prefer conda first.

Visualize it

Each project gets its own isolated Python + package set, so changes in one never break another.

diagram Environment isolation mermaid

  flowchart LR
  subgraph EnvA["Env: titanic-eda"]
    A1["Python 3.10"] --> A2["pandas 1.5, numpy 1.24"]
  end
  subgraph EnvB["Env: forecasting"]
    B1["Python 3.12"] --> B2["pandas 2.2, statsmodels 0.14"]
  end
  Global["Global / base Python"] -.->|"kept untouched"| EnvA
  Global -.->|"kept untouched"| EnvB

Two separate projects each keep their own Python version and installed packages

From the book

McKinney treats this as standard practice, not an optional extra: he creates a single dedicated environment for the whole book (conda create -y -n pydata-book python=3.10) rather than installing packages into the base environment. The same principle applies to every real project you build — one project, one environment — so upgrading a library for a new project never silently breaks an older one.

Environment naming conventions

Good names:

• analytics
• titanic-eda
• viz

Avoid generic names like test or newenv.

Best practices for data analytics projects

• Create one environment per project
• Pin versions for important packages (especially for long projects)
• Keep a requirements.txt (pip) or environment.yml (conda)
• Store notebooks inside a project folder

Example conda environment file (environment.yml)

This is a common way to share environment configuration:

environment.yml

name: analytics
channels:
  - conda-forge
dependencies:
  - python=3.12
  - numpy
  - pandas
  - matplotlib
  - seaborn
  - jupyter
  - pip
  - pip:
      - yfinance

Then create it with:

command

conda env create -f environment.yml

Next

Continue to: Installing Data Science Libraries (pip & conda) to learn the best ways to install and verify common analytics libraries.

🧪 Try It Yourself

Exercise 1 – Model Two Isolated Environments

pch.exerciseTag pch.exerciseTitle

Exercise 2 – Validate an Environment Name

pch.exerciseTag pch.exerciseTitle

Exercise 3 – Build an environment.yml Dependency List

pch.exerciseTag pch.exerciseTitle