Skrub: Less data wrangling, more machine learning
summary
Claims that Skrub slashes the 80% data‑prep burden by providing a Python library that converts messy tables into model‑ready features, shipping a table vectorizer as a scikit‑learn Transformer, a Gap encoder for substring categorical structure, interactive table report exploration, and an optional pre‑trained graph model (cart) that tops benchmarks in a large comparison. Logistic regression plus the vectorizer is a cheap strong baseline; deep models work best but are notably heavier to run.
AI-written summary · about
chapters (11)
- 0:00Tabular data for AI
- 2:12Data prep challenges
- 4:18Introducing Skrub’s vectorizer
- 6:07Gap encoder for strings
- 8:36Pre-training neural networks
- 10:25Cost and efficiency in DL
- 12:12Benchmarking preprocessing
- 13:12Balancing power and access
- 13:58Data exploration with TableReport
- 15:24Bridging pandas and machine learning
- 17:10Open source and Skrub’s future
related talks
26mSkrub: machine learning for dataframes
20mSkrub DataOps, Probabl
29mDirty Data Science Machine Learning On Non Curated Data
21mKeynote: Easier Machine Learning Thoughts From Scikit-Learn
32mPreparing Messy Data for Supervised
35mScikit-Learn can do THAT?!
32mUnlock the full predictive power of your multi-table data
33mUsing Scikit-Learn's interface for turning Spaghetti Data Science into Maintainable Software