What problem does it solve? Choosing and applying dimensionality reduction incorrectly wastes accuracy, leaks validation data, and produces misleading visualizations. This Skill provides decision rules, correct scikit-learn code patterns, and anti-patterns so reduction is applied only when justified and implemented without common mistakes. ## Core Features & Use Cases - Method Selection Guidance: Decide when reduction is warranted (p ≫ n, collinearity, latency, visualization) versus when gradient boosting or feature selection is the better path. - Correct Implementations: Ready-to-use patterns for PCA with StandardScaler, TruncatedSVD for sparse TF-IDF data, LDA for supervised reduction, and t-SNE/UMAP for visualization, all fitted inside cross-validation pipelines. - Anti-Pattern Checklist: Thirteen checklist items covering leakage, sparse-matrix densification, t-SNE misinterpretation, and rank tuning against downstream metrics. - Use Case: You have a 50,000-column TF-IDF matrix and need 300 components for a classifier. The Skill directs you to TruncatedSVD instead of PCA, avoiding memory exhaustion from centring a sparse matrix. ## Quick Start Ask the AI to help reduce the dimensionality of your dataset and it will select the right method, write the scikit-learn pipeline, and check it against the anti-pattern list.
