MLP Live session

This live session from the Machine Learning Practice channel covers regression techniques, focusing on K-nearest neighbors (KNN) and decision trees. The instructor begins with a recap of prior lessons, emphasizing the distinction between classification and regression, the use of dummy regressors as baselines, and the importance of scaling and cross-validation. They demonstrate building a baseline model, then introduce polynomial features to handle non-linear relationships, and discuss model evaluation via learning curves. The core of the session deals with KNN: its non-parametric, distance-based nature, the necessity of feature scaling, and hyperparameters like n_neighbors and weight functions. Practical examples using scikit-learn pipelines and named_steps are shown, along with methods to select the optimal k using elbow plots. Strengths and weaknesses of KNN are outlined, including its computational cost and sensitivity to irrelevant features. The session then briefly introduces decision tree regressors, explaining how they partition the feature space into non-overlapping regions. Throughout, student questions clarify pipeline usage, model selection, and the trade-offs between parametric and non-parametric methods. The delivery is instructional and interactive, suitable for learners with prior exposure to basic ML concepts.