MML-Book: one reader's exercise solutions for Mathematics for Machine Learning

What the README is honest about

The title of this project is its own qualifier: Mathematics for Machine Learning, my Self Study. That framing is carried through consistently and it is worth taking at face value, because the alternative reading, a maintained solutions repository for the textbook, is not what the contents support.

The README divides the work into three parts. The first is Python code, described as code the author played with to better understand various concepts, with one specific note: the mix of Gaussians was tricky in particular. The second is chapter exercise solutions, described as solutions for each chapter, with a practical warning that notebooks might take a bit of time to render all the LaTeX. The third is listed simply as Misc, followed by a colon and a closing bracket, which conveys the same message about that directory that the title conveys about the whole project.

So the author is not claiming completeness or authority. The wording describes a record of one person's progress. That is not a criticism, since self-study repositories are useful precisely because they show the path a reader actually walked, but it does change what you should expect before you start reading.

Six chapters of solutions, and which ones are missing

The exercise section links chapters 2 through 7, and each chapter gets two links: a Jupyter notebook rendered through nbviewer, and a PDF committed to the repository. The pattern is uniform across all six, which suggests a deliberate and completed effort rather than a partial one that was abandoned halfway.

What is missing is the other half of the book. The MML textbook runs well beyond seven chapters, with later sections on neural networks and reinforcement learning, and none of those appear in the README. Chapter 1 is absent too, though chapter 1 in that book is matrix algebra with no exercises, so its absence is unremarkable. The omission of the later chapters is not.

The gap matters for anyone using this as a check on their own work, because the later chapters are where the book's material diverges most from current practice. Early chapters on linear algebra, calculus and probability decomposition still map cleanly onto what people teach today. A notebook of derivations for a reinforcement learning chapter would raise harder questions about whether the reader's approach still holds, and no such check is provided here.

The link targets are also worth noting. They point at the `master` branch through nbviewer rather than at a commit SHA, so what you see rendered today is whatever the branch currently contains.

The textbook PDF sits in the repository root

The repository tree contains `mml-book.pdf` at the root, next to the README and the exercise directory. This is a copyrighted textbook, and the commit places a full copy inside a public Git repository.

There is no license file. The license field for the project is null, the tree has no LICENSE entry, and the README carries no licensing statement. That is a real constraint on what you can do with the notebooks even setting aside copyright, because code with no license is code you have no explicit permission to copy, modify or redistribute. If you want to reuse a derivation from one of these notebooks in your own project, there is no grant here to point at, and the author's intent is not recorded anywhere.

The same absence has a second effect: anyone auditing this repository for licensing questions gets no answer. For a personal study repository that is a defensible omission, since the author may not have intended redistribution at all. It only becomes a problem when the notebooks get quoted, forked, or lifted into coursework, which is precisely the use most people arrive at a solutions repository with.

Notebooks, a stray .DS_Store, and no environment

GitHub classifies the repository's primary language as Jupyter Notebook, which is an accurate signal about where the substance lives. The tree itself is small and legible: a README, the `end of chapter exercises` directory, a `misc` directory, a `python code to play with` directory, the PDF, and one file that does not belong, `.DS_Store`.

That `.DS_Store` is a macOS Finder metadata artifact, and its presence tells you the repository was worked on from the desktop rather than maintained as a checked-out project. It is harmless. It is also the kind of detail that tells you the repository was not set up with a contributing guide, a pre-commit hook, or any of the scaffolding that accompanies intentional ongoing maintenance.

More consequential is the absence of an environment file. There is no requirements.txt, no environment.yml, no pyproject.toml, and no lockfile anywhere in the tree. For notebooks this is a genuine usability question rather than a style complaint: notebooks written several years ago often depend on versions of NumPy, matplotlib or scikit-learn whose behaviour has since changed, particularly around array printing and plotting defaults. Anyone reproducing these notebooks is left to guess a working set from the imports visible in each file.

The repository also has no releases. There is no tag history, no version, and nothing to pin, so there is no way to refer to a known-good state of the notebooks other than a commit hash.

Activity without a maintenance policy

The repository has 1265 stars, 201 forks and 21 open issues, and it was pushed on 2026-09-30. Those numbers describe a repository people find useful and return to, which is about the highest compliment you can pay a personal study project.

Recent pushes and lack of releases are not in conflict. A person returning to a study repository tends to add a notebook or fix a rendering problem, not to prepare a release with a changelog. The default branch is `master`, the older convention, which is consistent with the repository having been created before `main` became standard and simply never migrated.

The open issue count is the most interesting number here, because 21 issues against a repository with no contributing guide means those threads are the de facto support channel. If you are stuck on an exercise, opening or reading an issue is more likely to get you an answer than anything in the README, which contains no usage guidance beyond the install-free observation that this is code to read.

For a repository whose topics include self-study, that is the right shape. The star count reflects the value of seeing someone else's worked derivations; the issue count reflects that nobody promised to maintain them.

Editorial conclusion

Treat this repository as one person's notebook rather than a maintained solution set, because that is exactly what the README says it is. It has real value for a specific job: if you are working through Mathematics for Machine Learning and want to compare your derivation against someone else's for chapters 2 through 7, having both notebook and PDF renderings side by side saves time. It has no value as a dependency, since there is nothing to install, no release to pin, and no license, which means no clear permission to reuse the code in your own work. The chapters that matter most for modern practice, the neural network chapters and the reinforcement learning chapter, are exactly the ones with no solutions here.

Frequently asked questions

Official sources

1. ilmoi/MML-Book on GitHub
2. Issues
3. README