skrub like a pro: clean, prepare, and transform your data faster

1. Introduction of the topic
   • Example: preparing and cleaning data with traditional tools
   • Common pitfalls: missing data, data leakage
   • First glance at how skrub mitigates these problems
2. Data exploration and sanitization
   • The TableReport and how it speeds up exploratory data analysis
   • The Cleaner and how it addresses common data preparation issues
   • Practical exercise: use both the TableReport and the Cleaner to explore and prepare a dataset
3. Transforming columns
   • Problem statement: why transforming columns is important
   • The ApplyToCols meta-transformer and how it helps with encoding columns according to simple rules
   • The skrub selectors, and how they expand column transformation to arbitrary rules and dataframe libraries
   • Practical exercise: use ApplyToCols and the selectors to select and transform features in a dataframe
4. Building a strong baseline
   • The TableVectorizer: a standalone transformer that converts arbitrary tables into fully-numeric feature matrices for ML models
   • The tabular pipeline: a full machine learning pipeline that combines data preparation, feature engineering, and model training in a single object
   • Practical exercise: build a TableVectorizer by combining the tools explained up to this point
5. Going deeper: skrub encoders
   • How skrub converts text and categorical data to numerical features
   • Robust numerical scaling with the SquashingScaler
   • Encoding datetimes with the DatetimeEncoder
6. Avoiding leakage and multi-table operations: skrub Data Ops
7. A brief introduction of the skrub Data Ops, and how they help in multi-table scenarios.

SKILLS YOU’LL GAIN:

• Explore and diagnose tabular data with TableReport.
• Clean and engineer features using skrub transformers.
• Build end-to-end scikit-learn pipelines with skrub and compare performance to baseline setups.

Updated on August 2026