Skip to content

← Build an app

Generate the requirements of your notebook

You can get your Jupyter notebook’s requirements automatically with Option 1 or manually with Option 2.

  • Option 1 (Automated): Recommended if you need to generate requirements from scratch.
  • Option 2 (Manual): Recommended if you already have a versioned requirements.txt file, as it will be faster.

Option 1: Automated Extraction of Requirements

1. Open the Requirements Generator Notebook

⚠️ IMPORTANT: You will need to choose the option based on the environment you used to run the target notebook whose requirements you want to extract.

  • Did you ran your notebook locally on your machine (with Conda/VirtualEnv)? ➡️ Locally
  • Did you ran your notebook on Google Colab? ➡️ Online

In both cases the Python environment that you use for the notebook, needs to run without errors. Knowing this please choose the option that best suits your case:

  • Locally: Go to the Requirements_Generator.ipynb notebook, download it (clicking on the button indicated on the screenshot below ⬇️) and open it in your local Python environment (conda or virtualenv).

    Note: When using it locally, the Python environment where you open the requirements generator notebook needs to be exactly the one you use to run the target notebook whose requirements you want to extract. For example, if you have multiple conda environments, you need to make sure that you use the same conda environment that you use to run the target notebook.

Download Notebook

2. Run the Requirements Generator

Follow the instructions provided inside the notebook to generate the requirements.yaml file. You will be prompted to provide the path to the notebook file you wish to analyse.

Option 2: Manual Extraction of Requirements

1. Create a requirements.yaml File

You will need to create a requirements.yaml file similar to the example requirements.yaml file. This file must contain the following three fields:

  • dependencies: A list of all packages used in the notebook along with their versions. The list needs to be indexed with a hyphen (-) followed by a space before each package.
  • python_version: The version of Python used in the notebook (e.g., 3.11.1).
  • description: A short description of the notebook and its purpose.

Example File Content:

dependencies:
  - pandas==2.2.2
  - ipython==7.34.0
  - tqdm==4.67.1
description: This notebook is for analyzing cell migration data...
python_version: 3.11.1

2. Open the Requirements Validator Notebook

We recommend using Google Colab for this step, as you do not need a specific Python environment to run the validator.

3. Run the Requirements Validator

Follow the instructions provided inside the notebook to validate your file. You will need to provide the path to your newly created requirements.yaml file when prompted.

What happens next: the lock file

When you push a notebook with its requirements.yaml, the workflow Updates on notebook change merges the requirements of every notebook into requirements.txt (plus requirements_gpu.txt and, if a notebook needs them, requirements-windows.txt, requirements-linux.txt or requirements-macos.txt).

The installer uses requirements_gpu.txt when it finds an NVIDIA GPU on Windows or Linux. In that file tensorflow becomes tensorflow[and-cuda] on Linux only: TensorFlow has had no Windows GPU build since version 2.10, and its CUDA packages exist only for Linux, so Windows keeps plain tensorflow (CPU).

It then writes a lock file next to each of them (requirements.lock.txt, ...). A lock file lists every package the installer installs, including the packages your packages depend on, each with its exact version, for Windows, macOS and Linux. The installer installs the lock file, so every installation of a release gets the same versions.

  • A lock file changes only when its requirements file, the Python version or the JupyterLab version changes. Pushing a notebook without new requirements does not move any version.
  • To move every package to its newest version, delete the requirements*.lock.txt files and run Updates on notebook change by hand (Actions > Updates on notebook change > Run workflow).
  • If the requirements cannot be locked, the workflow shows a warning and the installer installs the plain requirements.txt instead.

Edit this page on GitHub