How to Use GitHub Codespaces for Cloud-Based Development of the AI-For-Beginners Project

GitHub Codespaces provides a fully managed, container-based development environment that runs directly in your browser, allowing you to work through the AI-For-Beginners curriculum without installing Python, Conda, or any dependencies locally.

The AI-For-Beginners repository by Microsoft contains a complete 12-week curriculum covering neural networks, computer vision, natural language processing, and reinforcement learning. Because the project includes a Conda environment definition and Jupyter notebook-based lessons, GitHub Codespaces for cloud-based development is the fastest way to start learning—no local setup required.

Why GitHub Codespaces Works for AI-For-Beginners

The repository is purpose-built for cloud development through several key design decisions:

• environment.yml defines all required Python packages including TensorFlow, PyTorch, Keras, OpenCV, scikit-learn, and Jupyter—ensuring reproducible environments across all learners
• Jupyter notebooks in the lessons/ folder are automatically recognized by VS Code's built-in notebook renderer
• Docker-based containerization allows GPU-enabled runtimes for performance-intensive computer vision and NLP lessons (when available on your plan)
• .devcontainer metadata tells GitHub exactly how to build the development image

According to the official setup guide in lessons/0-course-setup/how-to-run.md, cloud execution is the recommended path for learners who want to skip local installation.

Step-by-Step: Launching Your Codespace

1. Create a New Codespace

Navigate to the repository at https://github.com/microsoft/AI-For-Beginners, then:

1. Click the green "Code" button
2. Select the "Codespaces" tab
3. Click "Create codespace on main"

GitHub provisions a virtual machine, clones the repository, and starts building a Docker container based on the project's configuration.

2. Wait for Container Build

The build process reads .devcontainer/environment.yml and executes conda env create. First-time builds typically take 3–5 minutes. The terminal panel shows real-time progress.

3. Open and Run Notebooks

Once the Codespace loads:

• Browse to lessons/ in the Explorer panel
• Open any .ipynb file (example: lessons/3-NeuralNetworks/03-Perceptron/Perceptron.ipynb)
• VS Code automatically starts a Jupyter server and renders the notebook

Execute cells sequentially—imports like import torch and import tensorflow as tf work immediately because the Conda environment ai4beg is pre-activated.

Verifying Your Environment

Run these commands in the integrated terminal to confirm everything is configured:

# Verify the Conda environment exists and is activated

conda activate ai4beg

# Check installed ML framework versions

python -c "import tensorflow as tf, torch, keras; print(f'TF: {tf.__version__}, PyTorch: {torch.__version__}, Keras: {keras.__version__}')"

# In any notebook cell, verify GPU availability

import torch
print("CUDA GPU available:", torch.cuda.is_available())

import tensorflow as tf
print("TensorFlow GPU devices:", tf.config.list_physical_devices('GPU'))

Understanding the Codespaces Architecture

Component	Implementation in AI-For-Beginners
Base image	Ubuntu with Miniconda (via Microsoft's dev container images)
Environment definition	.devcontainer/environment.yml pins exact package versions
Post-create setup	conda env create -f environment.yml followed by activation
IDE connection	VS Code web interface over secure WebSocket
Optional GPU tier	NVIDIA drivers pre-installed for CUDA-enabled frameworks

The environment.yml file referenced in the container build specifies channels (conda-forge) and dependencies including pytorch, tensorflow, opencv, matplotlib, numpy, and jupyterlab.

Running Code from the Terminal

Beyond interactive notebooks, you can execute lessons programmatically:

# Convert and execute a specific notebook to HTML

jupyter nbconvert --to html --execute lessons/3-NeuralNetworks/03-Perceptron/Perceptron.ipynb

# Run Python scripts directly

python lessons/4-ComputerVision/06-ImageClassification/mnist_example.py

# Start a standalone JupyterLab server (alternative to VS Code's renderer)

jupyter lab --ip=0.0.0.0 --port=8888 --no-browser

Persisting and Sharing Your Work

All file modifications auto-save to the Codespace's persistent storage. To commit changes:

1. Use the Source Control panel in VS Code (left sidebar)
2. Stage changes, write a commit message, and push to your fork
3. Alternatively, use the terminal: git add . && git commit -m "progress" && git push

Each Codespace includes a full Linux terminal with standard Git tools, so contributions follow the standard workflow documented in etc/CONTRIBUTING.md.

Customizing Your Development Container

For advanced use cases, modify .devcontainer/devcontainer.json (create if absent):

{
  "name": "AI-For-Beginners",
  "image": "mcr.microsoft.com/devcontainers/python:3.10",
  "features": {
    "ghcr.io/devcontainers/features/nvidia-cuda:1": {}
  },
  "customizations": {
    "vscode": {
      "extensions": [
        "ms-python.python",
        "ms-toolsai.jupyter"
      ]
    }
  },
  "postCreateCommand": "conda env create -f environment.yml && echo 'conda activate ai4beg' >> ~/.bashrc",
  "remoteUser": "vscode"
}

This configuration adds CUDA support, pre-installs VS Code extensions, and ensures the Conda environment activates automatically in new terminal sessions.

Troubleshooting Common Issues

• "Module not found" errors: Ensure the ai4beg environment is active (conda activate ai4beg)
• Notebook kernel not found: Select "Python Environments" → "ai4beg" from the kernel picker in the top-right of any notebook
• Build timeouts: Large Conda environments may exceed default build limits; restart the Codespace or upgrade your GitHub plan

Summary

• GitHub Codespaces eliminates local setup by building a container with all AI-For-Beginners dependencies automatically
• The environment.yml file ensures TensorFlow, PyTorch, and supporting libraries install reproducibly
• Jupyter notebooks run natively in VS Code's web interface without additional configuration
• GPU acceleration is available on eligible Codespace plans for faster model training
• All changes persist to Git, enabling seamless collaboration and progress tracking

Frequently Asked Questions

How much does GitHub Codespaces cost for AI-For-Beginners?

GitHub provides 120 core-hours per month free on the Pro plan (60 hours on Free accounts). The AI-For-Beginners curriculum typically uses 2-core machines, giving you 60 hours of free usage monthly. Standard Linux instances cost approximately $0.18 per hour; GPU-enabled instances start at $0.99 per hour.

Can I use my local VS Code instead of the browser?

Yes. Click the hamburger menu (≡) in the top-left of any Codespace, select "Open in VS Code Desktop", and the remote extension connects your local editor to the cloud container. All files, terminals, and debug sessions remain synchronized.

What if the Conda environment fails to build?

Check the "Creation Log" from the Codespaces management page. Common fixes include: clearing the Codespace cache and rebuilding, updating the channels priority in environment.yml, or removing version pins for conflicting packages. The maintainers test builds against the current environment.yml, so filing an issue at microsoft/AI-For-Beginners is recommended for persistent failures.

Is internet access required after the Codespace loads?

Yes, for downloading datasets and model weights referenced in lessons. However, core library imports and local computations work offline once the container is running. Pre-download large datasets to /workspaces/AI-For-Beginners/data/ for offline resilience.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}