scikit-learn, the parts you'll actually use

Pipelines, cross-validation, and the five models that solve 80% of tabular problems.

scikit-learn is enormous, but most working code uses ~10% of the API. This is that 10%.

What’s in here

• Pipeline and ColumnTransformer: keep preprocessing and model in one object
• the five models that solve most tabular problems: linear/logistic, ridge/lasso, tree, random forest, gradient boosting
• cross_val_score and why a single train/test split lies to you
• GridSearchCV vs RandomizedSearchCV, and when each one wins
• saving and loading models without breaking versioning

Prerequisites

• Tutorial 1 (training basics) recommended
• Comfortable with pandas DataFrames