interactive-sklearn-series

Master the hyperplane. We’ll explore how SVMs use kernels (Linear, RBF, Poly) to project data into higher dimensions, allowing for precise separation of complex, non-linear classes.

Classify data with speed. We’ll compare Logistic Regression for probability estimation against the SGDClassifier, which efficiently handles massive datasets using gradient descent.

How good is your prediction? We’ll evaluate performance using standard metrics like RMSE and MAE, and learn when to use R-squared versus Median Absolute Error for different business goals.

Not all relationships are straight lines. Learn to capture curvature using Polynomial Features and Support Vector Regression (SVR) to model complex, non-linear data patterns.

Outliers can wreck linear models. We’ll use robust estimators like RANSAC and Theil-Sen that ignore anomalous data points to find the true underlying trend in noisy datasets.

Start with the basics. We’ll cover Ordinary Least Squares (OLS) and explore regularization techniques like Ridge (L2) and Lasso (L1) to prevent overfitting and perform feature selection.

Avoid redundant computation. We’ll enable caching in pipelines to store the results of expensive transformers, significantly speeding up iterative experiments and hyperparameter tuning.

Complex pipelines can be opaque. We’ll configure sklearn to display interactive visual diagrams of your pipeline structure, making it easier to understand and debug the flow of data.

Combine features generated by different transformers. We’ll use FeatureUnion to concatenate results from multiple independent transformer pipelines into a single, rich feature set.

Different data types need different treatments. Learn to apply specific transformations to specific columns (e.g., scaling numbers vs. encoding text) simultaneously within a single workflow.