mlcourse.ai: a free 10-week course whose best assignments are the paywall
Open Machine Learning Course
At a glance
- What is it?
- mlcourse.ai is OpenDataScience's self-paced machine learning course, built as a Jupyter Book source repository and mirrored into five language folders. The free weeks cover ten weeks from Pandas to gradient boosting; the ten assignments described as best non-demo versions sit behind a 17 dollar monthly Patreon tier, and the only GitHub release is from January 2022.
- Who is it for?
- Work through the free weeks if you want a structured path from exploratory analysis to gradient boosting with notebooks you can run locally, and treat the paid pack as optional rather than essential. Before you start, check that Python 3.13 is available to you, because pyproject.toml sets that floor, and read the licensing split carefully, since the free course content and the bonus assignments are covered differently.
- Can I use it commercially?
- Check first. The repository uses a licence we do not classify automatically, so read its LICENSE file before any commercial use.
- Is it still maintained?
- Yes. The repository last received commits 73 days ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 13, 2026, and from our analysis. They are not legal advice.
Star history
- Yorko/mlcourse.ai10,803 ★ · +93
| Repository | 2026-09-13 | 2026-11-13 |
|---|---|---|
| Yorko/mlcourse.ai | 10.7k | 10.8k |
93 Stars in the last 61 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
Self-paced means 10 weeks of read, watch, submit, with nobody checking
The course is explicitly in self-paced mode, and the guidance is a week-by-week path rather than a schedule: for each of the 10 weeks, from Pandas to Gradient Boosting, you are told which articles to read, which lectures to watch, and which assignments to accomplish. That structure is the product. The lectures carry the math formulae, and the practice arrives as assignments plus Kaggle Inclass competitions, which is the balance the author describes aiming for. What self-paced removes is equally concrete. There is no cohort start date, so nobody is holding you to a pace. If you are the kind of learner who stalls without external deadlines, the honest move is to put your own dates in a calendar before you start, because the course does not impose any.
The bonus pack costs 17 dollars a month and is copyrighted separately
The README is unusually direct about the commercial arrangement. The Bonus Assignments pack holds the best non-demo versions of the course assignments, it is sold at $17 per month through the Bonus Assignments tier on Patreon or a similar tier on Boosty for Russian readers, and the reasoning given is that you pay for the one to five months it takes you to work through the course content, with a single contribution enough to open access. Two payment details are worth noting before you subscribe: the first payment is charged when you join and the next on the first day of the following month, which is why buying early in the month is advised. The pack contains 10 assignments. In some of them you are challenged to beat a baseline in a Kaggle competition under thorough guidance, with baseline write-ups for Alice and Medium, and in others you implement an algorithm from scratch, namely an efficient stochastic gradient descent classifier and gradient boosting. Now the licensing split, which is the part people misread: unlike the rest of the course content, the bonus assignments are copyrighted, public sharing of the pack is prohibited, and the informal understanding is that sharing with two or three friends is tolerated. So the free course and the paid pack are not the same content under two prices.
This repository is a Jupyter Book source tree, not an installable library
What lives at the root explains what kind of project this is. There is mlcourse_ai_jupyter_book/, the book configuration that turns notebooks into the site at mlcourse.ai, and there are five notebook directories: jupyter_english/, jupyter_chinese/, jupyter_russian/, jupyter_french/ and jupyter_arabic/, plus data/ for datasets and img/ for figures. There is a Makefile whose four targets define the whole workflow:
install:
uv sync
build:
uv run jb build mlcourse_ai_jupyter_book
check_links:
uv run jupyter-book build mlcourse_ai_jupyter_book/ --builder linkcheck
deploy:
uv run ghp-import -n -p -f mlcourse_ai_jupyter_book/_build/html -c mlcourse.aiinstall syncs the environment with uv, build renders the book, check_links runs the linkcheck builder over the same source, and deploy pushes the generated HTML in mlcourse_ai_jupyter_book/_build/html to the live site. Note what that means for a reader: you cannot pip install mlcourse-ai and get a course. The pyproject file describes an environment for building a book, not a library with an API, and the way to consume the lessons is the built site, the Kaggle Dataset mirror holding the same notebooks as Kaggle Notebooks, or the translated notebook directories in this repository. If you want to run the assignments offline, you clone, install, and build, which is a different commitment from opening a website.
The Python floor is 3.13, and one dependency is pinned to an exact version
pyproject.toml declares the project as mlcourse-ai version 1.0.0 and sets requires-python to >=3.13, which is the first thing to check before planning a local setup. The dependency list is long because the notebooks do real work: pandas, numpy, scipy, scikit-learn, statsmodels, matplotlib, seaborn and plotly for analysis and drawing, xgboost and prophet for the boosting and forecasting topics, pytesseract and pillow for image and text work, reverse-geocoder, ua-parser and user-agents for the parsing exercises, wordcloud, eli5 for feature importance explanations, pydot and pydotplus for graph rendering, holidays, ipywidgets and jupyter for the interactive parts. One pin stands out. jupyter-book is pinned with an exact version, jupyter-book==1.0.4.post1, while everything else uses a floor, so the book builder is deliberately frozen and the rest of the stack floats upward. The dev group adds black, pylint, pre-commit, pyyaml, pyzmq, fonttools, gdown, ghp-import, pylatexenc and sphinx-jupyterbook-latex. A repository-level detail worth flagging for anyone running the notebooks on a fresh machine: pytesseract appears in the runtime dependencies.
Course content is mirrored across five languages and four platforms
One article, several homes, is the publication model here. The English articles are published on medium.com under the Open Machine Learning Course publication, Russian versions go to habr.com, Chinese notebooks are linked through nbviewer against raw notebook files in this repository, and English Kaggle Notebooks are linked per topic, sometimes in parts, as with linear models spread across five notebooks and ensembles across three. A Kaggle Dataset mirror carries the same notebooks as Kaggle Notebooks, and the README names mlcourse.ai as the main site. The consequence for a learner is that the content is fragmented by language and platform, and you cannot assume that a Chinese notebook and an English notebook are at the same revision, because they live in separate folders with separate link targets. A second consequence is structural: a course whose lessons are largely external links ages badly, and the Makefile carries a check_links target that runs the linkcheck builder. The practical instruction for anyone using this as a syllabus is to work from the repository notebooks rather than from the third-party posts, and to expect the published articles to move. Translation coverage is uneven by design as well: French and Arabic exist as notebook directories, and the Arabic version has its own README.
The only GitHub release is v1.0.0 from January 2022, so trust main instead
Release metadata is misleading if you read it as an activity signal. The repository has exactly one release, v1.0.0, labelled Self-paced mlcourse.ai and dated 2022-01-16, and pyproject.toml carries the same static version 1.0.0. The last commit on the default branch is dated 2026-09-03, and the project is not archived. So the course has continued to receive commits years after the version number was frozen, and there is no tagged release to install, pin, or diff against. For a learner this matters in one specific way: if you want to reproduce exactly what you studied six months ago, a commit hash is the only handle available, because a tag will not move you. For a reader trying to judge whether the content is current, look at the branch rather than the release list. The repository also carries the habits you would expect from a text-heavy project that has been maintained for years: a pre-commit configuration, a spell checker config in .yaspellerrc.json, CONTRIBUTING.md, and a GEMINI.md file at the root. The last of those is the most unusual entry, since it means the repository is set up to be read by an AI assistant as well as by contributors.
A link checker exists, which tells you how the course rots at the edges
Among the Makefile targets, check_links deserves attention because it checks the links the course depends on:
check_links:
uv run jupyter-book build mlcourse_ai_jupyter_book/ --builder linkcheckThat target builds the book with the linkcheck builder, which walks the assembled pages and reports broken references.
Editorial conclusion
Work through the free weeks if you want a structured path from exploratory analysis to gradient boosting with notebooks you can run locally, and treat the paid pack as optional rather than essential. Before you start, check that Python 3.13 is available to you, because pyproject.toml sets that floor, and read the licensing split carefully, since the free course content and the bonus assignments are covered differently.
Frequently asked questions
What is the AI and ML course at mlcourse.ai?
It is an open machine learning course by OpenDataScience led by Yury Kashnitsky, currently in self-paced mode, guiding you through 10 weeks from Pandas to gradient boosting with articles to read, lectures to watch and assignments to accomplish, plus Kaggle Inclass competitions.
Is there a 3-month AI ML course available?
mlcourse.ai is structured as 10 self-paced weeks rather than a fixed-length cohort course, so there is no start date to sign up for. It is described as open and free, with a paid Bonus Assignments pack offered separately for $17 per month.
Which website is good for practicing machine learning?
mlcourse.ai combines lectures with assignments and Kaggle Inclass competitions, and the repository mirrors the notebooks to a Kaggle Dataset and to five language folders including English, Chinese, Russian, French and Arabic. The site is built from the repository with Jupyter Book.
Which AI and ML course is best?
That depends on what you need, since the course here is theory-leaning with heavy practice: math formulae in the lectures, assignments in the course, and 10 bonus assignments sold separately that ask you to beat a baseline in a Kaggle competition or implement stochastic gradient descent and gradient boosting from scratch.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/yorko-mlcourse-ai)Star history chart for your README
[](https://hysenlabs.com/projects/yorko-mlcourse-ai)Was this useful?