scikit-learn tips and tricks

Explore scikit-learn's toy datasets, including iris, diabetes, digits, wine, and breast cancer, and learn how to load them as data frames or arrays for classification, regression, and multi-output tasks.

Learn how to load real world data with scikit-learn using fetch and frame options, and explore california housing, labeled faces, unstructured text, and openml datasets for practical regression and classification.

Generate synthetic data with scikit-learn by using functions such as make_classification, make_blobs, make_moons, and make_s_curve, and explore parameters for features, samples, noise, and transformations across classification, clustering, and regression.

Explain why and how to use cross-validation in scikit-learn, compare cross_val_score, cross_validate, and cross_val_predict, and examine folds like k-fold, leave-one-out, leave-P-out, random permutation (shuffle), stratified, and time-series.

Examine cross validation generators, including k fold, repeated k fold, stratified k fold, leave one out, leave p out, shuffle split, group and time series splits, with practical visualizations.

Master scikit-learn metrics and scoring, including recall, precision, accuracy, and confusion matrices, and learn how cross-validation scoring and macro versus weighted averages inform model evaluation.

Learn to define custom scorers, apply grid search and cross-validation, and evaluate models with metrics like r2, mae, mse, precision-recall, roc curves, and learning curves.

Explore how to tune hyperparameters with grid search and cross-validation in scikit-learn, including pipeline, column transformers, and randomized search to balance performance and computation.

Explore out-of-the-box text feature engineering with scikit-learn, converting unstructured text to numeric features via count vectorizer and tf-idf, and tune n-grams and minimum frequency to optimize weights.

Explore how scaling and nonlinear transformations in scikit-learn preprocess data, using standard, minmax, robust, and max scalers, plus log, quantile, and power transformers, within pipelines.

Explore encoding techniques for categorical data, including ordinal and one hot encoding, with handling for unknowns, missing values, and infrequent categories, plus practical examples.

Master feature engineering for categorical and numerical data with k bins discretization, one-hot encoding, cut labeling, polynomial features, and robust imputation plus pipelines.

Explore how scikit-learn pipelines streamline preprocessing, model fitting, and grid search while preventing data leakage, with practical guidance on using make_pipeline, PCA, and SVC.

Explore how to use feature union to run parallel pipelines, adding new features with function transformers and column multipliers, and compare stateless versus stateful transformers.