Machine Learning
Supervised and unsupervised learning with scikit-learn: regression, trees, ensembles, SVMs, clustering and the math underneath.
What Is Machine Learning?
Define machine learning, contrast it with rules-based programming, and survey where ML wins in production.
Types of Learning
Supervised, unsupervised and reinforcement learning — and the sub-tasks that fall under each.
The Learning Problem
Formalize what a model learns: features, targets, hypotheses and loss — the vocabulary of every ML paper.
The Python ML Stack
Get fluent with scikit-learn: estimators, fit/predict, pipelines, and the dataset zoo used everywhere.
Linear Regression
The workhorse of prediction: least squares, multiple predictors, and interpreting coefficients.
Gradient Descent
The optimization loop under every model: compute the gradient, step downhill, repeat.
Logistic Regression
Probability for classification: the sigmoid, log loss, and decision boundaries that stay interpretable.
Decision Trees
Greedy splits, impurity, and why trees are the most interpretable nonlinear models.
Ensemble Methods: Bagging & Random Forests
Combine many weak models into one strong predictor with bagging and random forests.
Gradient Boosting
Boosted trees — the tabular-data champion — fit sequentially to the mistakes of previous models.
Support Vector Machines
Max-margin classification, the kernel trick, and when SVMs beat other models.
K-Nearest Neighbors
The simplest nonparametric model: predict by voting among the closest training points.
Naive Bayes
Bayes' rule applied to text and beyond — fast, simple, and shockingly effective on the right problems.
Feature Scaling & Selection
Prepare features so models learn well: scaling, encoding, selection, and dropping noise.
Regularization
Penalize complexity: L1 and L2 regularization, and the bias-variance trade-off made concrete.
Cross-Validation
k-fold cross-validation gives a more honest error estimate than a single split — and catches tuning leaks.
Hyperparameter Tuning
Search parameter space systematically with grid, random and Bayesian strategies — without leaking into the test set.
Classification Metrics Deep Dive
Precision, recall, F1, ROC curves and calibration — choose the metric that matches the business cost.
K-Means & Clustering
Cluster unlabeled data with k-means, pick k responsibly, and interpret clusters as business segments.
Dimensionality Reduction with PCA
Project high-dimensional data onto its principal components — for visualization, denoising and speed.
Machine Learning Roadmap
Synthesize the course, plan the next projects, and chart the path into deep learning and production ML.