Skoltech ML homeworks go open source

This feels less like a class handout and more like a reusable blueprint for teaching ML the way engineers actually learn, by building the thing and then breaking it against tests.

• –The starter-template-plus-test-suite approach is a strong antidote to blank-page paralysis while still forcing students to implement core algorithms themselves.
• –Auto-grading and grading scripts scale much better than hand review, and they give students immediate feedback on both correctness and spec compliance.
• –Generating separate test data for each homework is a smart anti-overfitting measure that makes the assignments harder to game.
• –The MIT license makes this genuinely reusable curriculum infrastructure for other instructors, not just a one-off course archive.
• –The quiz add-on is a nice reminder that even great tests do not replace conceptual understanding.