For those comfortable using the command line, we offer basic instructions for:

• Mac OS X
• Windows 10

Not recommended local JupyterLab installation methods

We do not recommend Anaconda because it can be difficult to install, troubleshoot, and uninstall. Anaconda installs a large suite of data science tools; many are not necessary for using these notebooks. Moreover, Anaconda can be difficult to properly un-install which makes it hard to switch if you would like to use another solution.

A better solution is Miniconda, which installs far fewer applications on the machine. However, the difficulty of installing Miniconda is similar to JupyterLab by itself.

Using a local IDE for Jupyter notebooks

If you are familiar with using the VS Code or Pycharm IDEs, they offer tooling for working with Jupyter notebooks. VS Code has native support for Jupyter notebooks. PyCharm is also a good solution for Jupyter notebooks. PyCharm has free version with basic Jupyter notebook support and a Pro version with advanced Jupyter notebook functionality that is free for teachers and students.

Table of recommended local solutions

Recommended cloud-based solutions for using Jupyter notebooks

There is a diverse market for Jupyter notebook cloud-based solutions including free, paid, and hosted solutions. The tradeoffs generally come in terms of monthly costs and environment management. It is a good idea to see what resources your campus offers; you may already have access to free, managed solutions with IT support.

GPU compute is not necessary

There is no need to prioritize a cloud environment with GPU access for using these notebooks. These notebooks do not use a GPU directly, preferring to use GPU computation through APIs which are usually easier, faster, and cheaper. If you plan to train models, then an environment with GPU access may make sense.

JupyterHub (the best solution)

Project Jupyter offers a cloud-based solution with integrated JupyterLab called JupyterHub. While this is the best solution, it is impractical for most users to run their own JupyterHub since it requires server management expertise. However, many higher education institutions run their own JupyterHub servers, so it is worth investigating whether a JupyterHub server is available on your campus. Check with your local IT and ask colleagues working in data science to see if JupyterHub is available.

CoCalc

CoCalc has a unique user interface designed for collaboration and teaching. It is especially useful for mathematics and computer science education with support for grading, LaTex, and course management. CoCalc free tier is very limited. An instructor can expect to pay about $30/month for access.

Anaconda Cloud

Anaconda Cloud may be a good solution if you are comfortable with Anaconda and/or conda package management. It is easy to share environments for reproducibility. There are collaboration features and version control available. A basic starter account is $15/month, but it is possible to get free access with an academic account.

JupyterHub hosting through Hugging Face, Digital Ocean

If you are technically savvy, Hugging Face Spaces and Digital Ocean offer relatively easy ways to get JupyterHub running on a server. Plan on spending about $10/month for persistent storage and basic compute. The downside to this approach is you will need to be prepared to troubleshoot any platform issues. It is always a good practice to back up your notebooks to your local machine or a git repository, but this is especially important if you are new to server management.

GitHub Codespaces IDE

If you would like a full, cloud-based IDE, then GitHub Codespaces offers cloud access to VS Code which has support for working with Jupyter notebooks. Choose this option if you are familiar with VS Code and you would benefit from a tight integration with GitHub.

Not recommended cloud-based solutions

Google Colab is a popular platform for using Jupyter notebooks because Google offers limited-access to GPU acceleration for model training. The Constellate notebooks do not include model training and do not require GPU acceleration. Instead, they use APIs to access GPU acceleration, which is often faster, easier, and cheaper.

Outside of its limited, free GPU access, Google Colab is inferior to other cloud-based Jupyter notebook solutions. The user interface and functionality is closed, non-standard, and difficult to use for file management. Moreover, it lacks support for basic notebook features, including:

• Markdown local file access (no local notebook linking, no local image rendering)
• Only one notebook at a time (no tabs for opening multiple notebooks)
• No command line access (in the free tier)

Notebooks developed for Colab often only work in Colab, locking users into the environment. Google Colab's architecture violates the FAIR principles of scientific data management.

Deepnote is a useful platform for data science professionals with some advanced functionalities for working collaboratively and accessing data repositories. The additional features make it better for working data scientists than teachers and learners.

Binder is a free platform useful for demonstrating notebooks, but it has no persistent storage. The speed and availability are not reliable.

Table of recommended cloud-based solutions

Package management

Regardless of the solution you choose, you will need to become familiar with installing Python packages that these notebooks use. If you get an error that a package could not be imported when running a cell, that probably means it is not installed. You will need to install it (and may also need to restart the notebook kernel) before it will be recognized.

There are two main package managers for working with Python: pip and conda. (More advanced projects may use virtual environment, such as venv or poetry.

Pip

Pip is the official Python package installer. Once Python is installed, you can generally use pip on the command line to manage packages.

Installing a package

pip install PACKAGENAME

Update a package

pip install --upgrade PACKAGENAME

Change to a particular version

pip install --pip install --upgrade PACKAGENAME==3.2.16

Pip documentation

See the official pip documentation and/or the Python Package Index (PyPi)

Conda

The Conda package manager is usually part of an Anaconda solution, such as Anaconda Cloud, Anaconda, or Miniconda. It can be used from the command line, but there are also graphical ways to manage packages with Conda. See the particular documentation for the version of Anaconda you are using.

Common text analysis packages used in these notebooks

• gensim - Create word embeddings, topic models, and other natural language processing outputs using modern statistical machine learning.
• matplotlib - Create static, animated, and interactive visualizations in Python.
• nltk (Natural Language Tool Kit) - Work with over 50 corpora and lexical resources, such as Wordnet. Deploy a suite of text proc