Random machine-le 6EFA arning questions, solved two ways:
- from-scratch — core algorithms implemented by hand in NumPy/PyTorch, no shortcuts
- applied — end-to-end work on real datasets: EDA → features → modeling → honest evaluation
Every notebook runs top-to-bottom from a clean environment — verified automatically by CI.
uv sync
uv run jupyter labPython ≥ 3.12 is installed automatically by uv.
| # | Notebook | Track | Topics | Status |
|---|---|---|---|---|
| 01 | Cosine similarity from scratch | from-scratch | similarity metrics, vectorized NumPy | ✅ CI-verified |
| 02 | tfidf | from-scratch | ⬜ pending | |
| 03 | one hot encoding | from-scratch | ⬜ pending |
The easy way:
uv run python scripts/new_notebook.pyIt asks three questions — the problem name, whether it uses external data, and the
track (from-scratch / applied) — then does the rest for you:
- picks
templates/problem-template.ipynb, ortemplates/data-problem-template.ipynbwhen external data is involved - names the file with the next free number:
NN-your-name.ipynb(data problems automatically get-using-dataappended so CI skips them) - fills in the
<Problem Title>heading from the name you gave - appends a row to the index table above
Then it's on you to:
- fill in every template section (and the
<dataset-slug>in data notebooks) - run all cells top-to-bottom from a clean kernel
- replace the index row's
<topics>placeholder
Manual fallback: copy a template into notebooks/<track>/ yourself, prefixed with
the next number. Data-problem filenames must end in -using-data.
| Filename | CI behavior | Index Status |
|---|---|---|
ends -using-data.ipynb |
skipped (downloads aren't reproducible) | runs locally |
| anything else | re-executed top-to-bottom on every push | ⬜ pending → ✅ CI-verified |
Data-track notebooks aren't CI-verified, so their Status should say they run locally.