stage 04 · Core
Classical Machine Learning
Frame defensible prediction problems, implement central model families, and turn evaluation evidence into bounded decisions.
5 sections 42 units about 145 h 0/42
Sections
In dependency order
Capstone
Classical ML Investigation Pack (8 to 12 h)
Complete one compact investigation on a controlled structured dataset. Three artifacts are enough to show formulation, mechanism fluency and decision quality without collecting unrelated notebooks.
Milestone deliverables
- Contract and leakage pack - prediction contract, immutable manifest, split builder, schema and availability tests, two naive baselines, and one preserved leakage failure with its repair
- Classical mechanisms suite - tested NumPy implementations of regularized linear or logistic regression, naive Bayes, a deterministic tree core plus one ensemble, kNN or a small kernel model, PCA and k-means
- Decision dossier - locked model budget, raw predictions, paired uncertainty, calibration and threshold policy where relevant, slices, error analysis, one shift or failure injection, a negative result and a bounded next decision
Rubric
| Criterion | Pass | Strong |
|---|---|---|
| Contract and leakage pack | The section 1 mini-project meets Pass on every criterion. | It meets Strong on leakage and lineage, and one command runs every audit. |
| Classical mechanisms suite | Tested NumPy implementations of a regularized linear or logistic model, naive Bayes, a deterministic tree core with one ensemble, kNN or a small kernel model, PCA and k-means, each checked against an independent oracle. | One test command runs the whole suite, and every mechanism has property tests over seeds and shapes. |
| Decision dossier | The section 5 mini-project meets Pass on every criterion, on the same controlled dataset as the contract pack. | It includes one injected failure from the contract pack and shows the audit catching it before the comparison runs. |
| Rebuild from a clean clone | A README lists the commands that rebuild each artifact. | One script rebuilds every reported number on a laptop CPU, with its run time stated. |
| Honest limits | Each artifact states what its compact dataset and models cannot show. | Each limit names the later stage or track that addresses it. |
Capstone hours · 8 to 12
Begin the first section