Infix-Extensible Record Types for Tabular Data
summary
A row-polymorphic record calculus that adds scoped labels and first-class labels and rows plus record concatenation, with a Hindley-Milner–style inference algorithm that is sound but loses most-general-unifier guarantees in rare cases. Types let projection, deletion and concatenation express column shadowing and optional results; the system successfully types most Brown Benchmark functions (fails on pivots, positional indexing and multi-column rename) and models left-join as copying left rows and attaching an optional right-value column.
AI-written summary · about
chapters (5)
- 0:00Record types for tabular data
- 2:48Record calculus sketch
- 8:26Formal definition and results
- 12:36Brown Benchmark evaluation
- 16:48Conclusion and future work
related talks
25mRecursion Schemes Without the Barbed Wire (Part 5)
21mRecursion Schemes Without the Barbed Wire (Part 4)
32mRecursion Schemes Without the Barbed Wire (Part 3)
42mTables Considered Helpful
24mRecursion Schemes Without the Barbed Wire (Part 6)
41mRecursion Schemes Without the Barbed Wire (Part 2)
15mDive Deep into Metadata with Tplyr
13mInReAcTable: LLM-powered Interactive Visual Data Story Construction from Tabular Data