NeuromatchAcademy/course-content: what the NMA computational neuroscience notebooks actually contain

What NeuromatchAcademy/course-content is for

This is the syllabus repository for the Neuromatch Academy computational neuroscience course, a three-week summer programme whose 2025 run the README dates as July 7 to 25. The audience is narrow and specific: students with the prerequisites listed in the separate NeuromatchAcademy/precourse repository, teaching assistants who run the tutorials live, and instructors who want to reuse the material in their own course. The README states plainly that the content should primarily be accessed from the ebook at compneuro.neuromatch.io, which the README describes as under continuous development. That sentence matters more than it looks. The repository is the source, the ebook is the product, and the two are not guaranteed to be in step at any given moment. If you arrive expecting a Python library with an importable module, you are in the wrong place. The topics listed for the repository (dynamic systems, machine learning, neuroscience, stochastic processes) describe the subject matter of the notebooks, not the capabilities of a package.

How the notebooks, book sources and projects fit together

The top-level layout tells you the architecture without any documentation: tutorials/, book/, projects/, plus environment.yml and requirements.txt. Tutorials are Jupyter Notebooks, which is consistent with the repository's primary language being listed as Jupyter Notebook. The book/ directory holds the sources that build the ebook, so the same content exists in two forms and the repository is the origin of both. The projects/ directory is separate from the tutorials, which fits the structure of a summer school where students spend the final stretch on a group project rather than on guided exercises. Dependencies are declared twice, in environment.yml for a conda environment and in requirements.txt for pip. The requirements file pins almost nothing: requests, numpy, scipy, matplotlib, scikit-learn, torch, ipywidgets, tqdm, pathlib, pandas, xkcd, h5py, opencv-python, torchvision and natsort are all unpinned, and only decorator is fixed at ==5.0.9. That single pin is the clearest signal in the file that something in the notebook stack breaks when decorator moves. Everything else floats, which means a fresh install in 2026 will not reproduce the environment the notebooks were written against.

Installing the environment and opening a first tutorial

The README does not give install steps. It points to the ebook and to the prerequisites page in the precourse repository, and it lists environment.yml and requirements.txt among the repository's top-level entries. Those two files are the only dependency manifests on offer, so the practical route is to build an environment from environment.yml, or to install requirements.txt with pip. Neither file is quoted in the README, and the README does not name the conda environment it declares or a supported Python version. What the requirements file does show is the dependency set: requests, numpy, scipy, matplotlib, scikit-learn, torch, ipywidgets, tqdm, pathlib, pandas, xkcd, h5py, opencv-python, torchvision and natsort are listed without version constraints, while decorator is held at ==5.0.9. Once the environment resolves, the tutorials directory at the repository root is what you point a notebook server at. Expect to resolve version conflicts by hand on a modern Python, particularly around torch and torchvision, because nothing in the repository pins them to a known-good set.

The dependency pin that tells you where the pain is

A single pinned dependency in an otherwise open requirements.txt is worth reading as a bug report. decorator==5.0.9 is held while numpy, scipy, torch and the rest float. The README does not explain the pin, and no comment in the file does either, so the reason has to be inferred: some notebook in the stack depends on decorator behaviour that changed after that release. For an instructor rebuilding the course, this is the kind of detail that costs an afternoon. The rest of the file is a heavier problem. torch and torchvision unpinned means the notebooks may assume tensor APIs from a particular era, and a clean install today can pull a version where a call has moved or a default has changed. There is no lockfile, no constraints file and no CI configuration visible in the top-level entries that would pin the teaching environment to a known-good set. If you plan to run these notebooks for a class, freeze your own environment after the first successful run and keep that lockfile, because the repository will not do it for you.

Where this repository is the wrong tool

Treating course-content as a maintained software dependency is the main failure mode. The last push was on 2026-07-14 and the most recent release is v3.2.2 from the same day, so the material is current for the 2025 course cycle, but the versioning tracks the course, not a stable API. A release bump can reorganise a notebook, rename a section or change an exercise, and anything you built on top of a specific notebook path can move. The README also makes no promise about backward compatibility, and the statement that the ebook is under continuous development applies to the content itself. There is no documented rollback procedure, no changelog in the README, and no migration note for people who forked an earlier cycle. A second limit is scope: this teaches computational neuroscience, so a reader who wants a general machine learning curriculum will find the machine learning here bound to neural data and models. A third is that the repository is not a dataset. If you need the data the tutorials use, the README does not say where it lives.

How it compares with NeuromatchAcademy/precourse

The obvious alternative inside the same organisation is the precourse repository, which the README links for expected prerequisites. The difference in approach is one of entry point rather than subject. Precourse is the on-ramp: it exists to bring students up to the level the computational neuroscience notebooks assume, and the README treats it as the place to check whether you are ready. course-content assumes that readiness and starts teaching. If you are deciding where to spend your time, the README's own framing answers it: check the prerequisites page first, then use the ebook. Outside the organisation, the honest comparison is a university course taught from a textbook, where the reading list is stable and the exercises are not redistributed. Here the exercises are the repository, which is the advantage and the liability at once.

Licence, reuse and the cost of keeping a fork alive

The repository carries two licences, and the README explains how they divide: content is under Creative Commons Attribution 4.0 International, and software elements are additionally under the BSD 3-Clause licence, with derivative works permitted to use whichever is more appropriate to the context. LICENSE.md and LICENSE-CODE.md sit at the top level to match. For an instructor, CC BY 4.0 means you can adapt the notebooks for your own course provided you give attribution, and BSD 3-Clause covers the code portions. That is a permissive combination and it is the reason reuse is realistic. The maintenance cost is the part the README does not discuss. If you fork the material to teach your own version, you inherit the unpinned dependency problem above and you own the divergence from upstream. The repository does not document a rebase or upgrade path, so the practical question is whether you want to track the course cycle or freeze at one release and maintain it yourself. Neither choice is wrong, but the second one means the environment.yml and requirements.txt become your responsibility from day one.

Editorial conclusion

Adopt this repository if you teach computational neuroscience and want a ready-made syllabus of Jupyter Notebook tutorials with an environment.yml and requirements.txt you can rebuild, or if you are a self-learner willing to follow the ebook at compneuro.neuromatch.io. Do not adopt it as a Python package, as a stable API, or as a source of production code: it is a course. Verify first that the notebook versions you clone match the schedule in tutorials/Schedule/daily_schedules.md and that your PyTorch install satisfies the torch entry in requirements.txt, since the README does not document a supported version range.

Frequently asked questions

Official sources

1. License: CC-BY-4.0
2. NeuromatchAcademy/course-content on GitHub
3. Project website
4. README
5. Releases