// the find
Yorko/mlcourse.ai
Open Machine Learning Course
A 10-week structured ML course built around Jupyter notebooks, covering the classic supervised/unsupervised ML stack from pandas EDA through gradient boosting. Comes with video lectures, demo assignments, and three Kaggle Inclass competitions. Aimed at practitioners who want the math alongside the code, not just API calls.
The Vowpal Wabbit topic is genuinely rare in a course at this level — online learning on gigabyte-scale data with hashing tricks is something most ML courses skip entirely. The gradient boosting topic goes deep enough to explain what XGBoost, LightGBM, and CatBoost are actually doing differently from each other, not just how to call .fit(). Assignments are tied to real Kaggle competitions, so you get a leaderboard as external validation rather than grading against a hidden test set the instructor controls. The author's Kaggle Master background shows — the feature engineering section covers tricks that actually move competition scores.
The course ends at gradient boosting in 2020 and has not added a neural networks or transformers track; anything involving embeddings, LLMs, or modern deep learning is simply absent. The 'open' framing is undercut by the Patreon paywall on the non-demo assignments — the free assignments are visibly easier than what the course is building toward, so you hit a wall right when the material gets interesting. The time series section is built around Facebook Prophet, which Meta effectively abandoned and whose forecasting assumptions are widely criticized; a new learner following this section will pick up bad habits about what time series modeling should look like. Vowpal Wabbit coverage, while thorough, points students toward a tool with a shrinking ecosystem when PyTorch and its online learning patterns are now the practical choice.