Jupyter notebooks for coursework: local install, Google Colab or course-hosted labs

What a notebook file actually is

A Jupyter notebook is a single .ipynb file. The nbformat specification describes it as a JSON document with four top-level keys: metadata, nbformat, nbformat_minor and cells. Cells come in three types: code cells, which hold source code for the notebook's kernel; Markdown cells for text; and raw cells that are passed through unchanged when the notebook is converted.

The detail beginners miss is that outputs live inside the file. Each code cell keeps an outputs list, covering printed text, rich display data such as charts, returned values and error tracebacks. That is why a notebook you open on another machine still shows results you ran yesterday, why a notebook with many plots can grow large, and why sharing a notebook can share more than your code.

Every option below opens the same .ipynb format, so you can move work between them. What differs is where the kernel runs, where the file is saved, and what disappears when a session ends.

• Code cell: runs in the kernel; its output is saved in the file.
• Markdown cell: your notes and explanations; good for writing up a course exercise.
• Kernel: the running process that executes code. Restarting it clears variables, not the saved file.

Option 1: a local install

Jupyter's install page gives two pip commands. For JupyterLab, the fuller interface, run pip install jupyterlab and start it with jupyter lab. For the classic Notebook interface, run pip install notebook and start it with jupyter notebook. On macOS or Linux, Homebrew also offers brew install jupyterlab. The install page does not name a minimum Python version, so you need a working Python and pip first.

When the server starts, Jupyter's documentation says it runs on your own computer (localhost) at port 8888 by default and moves to the next free port if 8888 is busy. The dashboard lists the notebooks, files and folders in the directory where you started the server, so start it from the top folder of your course work rather than your home directory. Press Control-C in the terminal to stop the server and shut down all kernels.

If a course expects specific package versions, the install page points to environment managers such as conda, mamba and pipenv, and recommends the conda-forge channel when installing with conda or mamba. A separate environment per course stops one course's package versions from breaking another's.

• Best for: courses that run for months, work you want offline, and projects you will publish later.
• Cost: none beyond your own computer.
• Watch for: package setup taking time in week 1, and heavy machine learning work that is slow without a suitable GPU.

Option 2: Google Colab

Colab's FAQ says Colab is free of charge, but that resources are not guaranteed or unlimited and usage limits fluctuate. Free notebooks can run for at most 12 hours, runtimes time out when idle, and the GPU and TPU types available vary over time. Google does not publish exact limits because they change.

Notebooks are stored in Google Drive or can be loaded from GitHub, and you can download a notebook from Drive or the File menu; the FAQ says Colab uses the open-source Jupyter notebook format. The virtual machine itself is deleted when idle and has a maximum lifetime, so libraries you installed and files you created on it are not kept. The FAQ's advice is to put install commands in a notebook cell so the notebook can rebuild its own environment.

Colab Pro, Pro+ and Pay As You Go buy more compute availability through compute units; Pro+ supports continuous execution for up to 24 hours with enough units, and paid users drop back to free-tier rules when their balance runs out. The FAQ also lists prohibited uses, including using multiple accounts to get around limits.

• Best for: short exercises, borrowed or low-powered computers, and courses that link to Colab notebooks.
• Save habit: copy a shared course notebook to your own Drive before editing, and save data outputs to Drive rather than the virtual machine.
• Sharing caution: a shared notebook includes code, output and comments unless you turn on the setting to omit code cell output.

Option 3: course-hosted labs

Some courses, including programming courses on Coursera that use Coursera Labs, open notebooks in a lab environment attached to the assignment. Because the course sets up that environment, you skip installing packages yourself and work in the setup the assignment was written for.

A DeepLearning.AI community guide to Coursera Labs describes the routine for keeping your work. In the Jupyter Notebook environment, use File, Save and Checkpoint, then File, Download as, Notebook (.ipynb). To get data and helper files too, open Lab Files and use Download all files. To get the latest version of an assignment, rename your notebook first so your copy is kept, then use Help, Get latest version, Update Lab; the guide says the page then refreshes with the latest version of the notebook. It also notes that courses use either a Jupyter Notebook or a JupyterLab environment, with different menus, so check which one yours uses.

This is community guidance, not a Coursera help page, and the menus can change between course versions. Download before you rename or delete anything. If you plan to show the work publicly later, remember that Coursera's Honor Code prohibits making solutions to homework, quizzes, exams or projects available to anyone else unless an assessment explicitly permits it.

• Best for: graded programming assignments and anything submitted to an autograder.
• Save habit: download the notebook and the workspace files at the end of each week.
• Watch for: files you edit in a lab are not on your computer until you download them.

Which option fits which course situation

Most learners end up using two options: the course lab for graded work, and a local install or Colab for practice and personal projects. Use the table to pick a default.

A save-and-export checklist for any option

Worked example: a ten-week course with two notebooks a week produces about 20 notebooks. If each lives only in a lab or on a Colab virtual machine, one lost session can cost a week of work. A five-minute weekly routine avoids that.

• Keep one folder per course with subfolders by week; start a local server from that course folder.
• At the end of each session, save, then download the .ipynb (lab or Colab) or commit it (local).
• Put package installs in the first cell, so a fresh Colab or lab session can rebuild the environment.
• Clear outputs before sharing a notebook if they include data you should not publish.
• Rename and download a lab notebook before getting the latest version, so your edited copy is kept under its own name.
• Keep graded solutions private; publish only work you extended with your own data or questions.

Sources & verification

Product details and prices can change. Check the linked provider before buying.

1. Project Jupyter: Installing Jupyter Accessed 30 Oct 2026
2. Jupyter documentation: Running the Notebook Accessed 30 Oct 2026
3. nbformat documentation: The Notebook file format Accessed 30 Oct 2026
4. Project Jupyter: Try Jupyter in your browser Accessed 30 Oct 2026
5. Google Colab: Frequently Asked Questions Accessed 30 Oct 2026
6. DeepLearning.AI Community: Downloading your notebook, downloading your workspace and refreshing your workspace Accessed 30 Oct 2026
7. Coursera: Terms of Use and Honor Code Accessed 30 Oct 2026