Tools for building intuition across the ML toolkit — cross-validation, regularization, clustering, and model selection — drawn from MIT, UC Berkeley, Carnegie Mellon, UNC, and UW curricula.
Explore classic ML methods on toy datasets — logistic regression, decision trees, support vector machines, clustering algorithms, principal component analysis — with cross-validation, regularization knobs, and side-by-side comparisons. Build grounded intuitions without shipping production code.
Coordinate a volunteer data team's modeling projects: each member proposes an approach (logistic regression, random forests, SVM, neural network), logs cross-validation results and regularization choices, and the team discusses tradeoffs in threaded comments tied to specific decisions.
Manage a portfolio of production models: log feature drift weekly, re-run model selection comparisons monthly, apply the feature engineering pipeline, and document every decision so a future data scientist can trace why each model looks the way it does.
Design a clinical risk model for sepsis prediction: define the prediction task formally, document feature engineering steps, run structured model selection comparing logistic regression vs. random forests vs. neural networks, and produce a model card the clinical advisory board can read.
Review ML models deployed in regulated decisions — lending, hiring, healthcare — for algorithmic fairness and statistical validity: evaluate cross-validation protocol, examine subgroup performance disparities, and scrutinize whether a complex neural network was justified over a simpler regularized regression.
