The raise of machine learning hyperparameter constraints in Python code - Inria - Institut national de recherche en sciences et technologies du numérique Access content directly
Conference Papers Year : 2022

The raise of machine learning hyperparameter constraints in Python code

Ana Milanova
  • Function : Author
  • PersonId : 1199999
Guillaume Baudart
  • Function : Author
  • PersonId : 1119186
Julian Dolby
  • Function : Author
  • PersonId : 1200000

Abstract

Machine-learning operators often have correctness constraints that cut across multiple hyperparameters and/or data. Violating these constraints causes the operator to raise runtime exceptions, but those are usually documented only informally or not at all. This paper presents the first interprocedural weakest-precondition analysis for Python to extract hyperparameter constraints. The analysis is mostly static, but to make it tractable for typical Python idioms in machine-learning libraries, it selectively switches to the concrete domain for some cases. This paper demonstrates the analysis by extracting hyperparameter constraints for 181 operators from a total of 8 ML libraries, where it achieved high precision and recall and found real bugs. Our technique advances static analysis for Python and is a step towards safer and more robust machine learning.
Fichier principal
Vignette du fichier
issta22.pdf (312.85 Ko) Télécharger le fichier
Origin Files produced by the author(s)

Dates and versions

hal-03891774 , version 1 (09-12-2022)

Identifiers

Cite

Ingkarat Rak-Amnouykit, Ana Milanova, Guillaume Baudart, Martin Hirzel, Julian Dolby. The raise of machine learning hyperparameter constraints in Python code. ISSTA 2022 - 31st ACM SIGSOFT International Symposium on Software Testing and Analysis, Jul 2022, Virtual, South Korea. ⟨10.1145/3533767.3534400⟩. ⟨hal-03891774⟩
25 View
120 Download

Altmetric

Share

Gmail Mastodon Facebook X LinkedIn More