Low-level strongly typed dataframes for machine learning and statistical computing in resource-constrained devices
Jayaraj Poroor · 2019
Strongly typed dataframe-like abstraction suitable for resource-constrained devices is presented. The proposed dataframe type system assigns types to memory addresses rather than symbolic names, supports memory reuse, and has zero runtime overheads. A subsumption relation is defined on dataframe types that allows flexible, strongly-typed access to rows, columns, and fragments. A specialized form of separation logic is proposed to formally reason about dataframe typing, allowing compositional reasoning. Hoare-style formal rules are defined to reason about the type-safety of programs operating on dataframes.