Generating Random Floating-Point Numbers by Dividing Integers: a Case Study

Frédéric Goualard (1)

1 LS2N - Laboratoire des Sciences du Numérique de Nantes

A method widely used to obtain IEEE 754 binary floating-point numbers with a standard uniform distribution involves drawing an integer uniformly at random and dividing it by another larger integer. We survey the various instances of the algorithm that are used in actual software and point out their properties and drawbacks, particularly from the standpoint of numerical software testing and data anonymization.

Mots clés: floating-point number; IEEE 754; random number

Domaines: Arithmétique des ordinateurs

Frédéric Goualard. Generating Random Floating-Point Numbers by Dividing Integers: a Case Study. Proceedings of ICCS 2020, Jun 2020, Amsterdam, Netherlands. ⟨hal-02427338⟩