Available FAT software

• Transparency (interpretability and explainability)
  • Algorithms
  • Packages
• Fairness

A list of available fairness, accountability and transparency software packages:

Transparency (interpretability and explainability)

Algorithms

• Anchor;
• LIME;
• Individual Conditional Expectation Plot – ICE;
• Forestspy.

Packages

• Skater:
  • feature importance,
  • partial dependance plots,
  • LIME,
  • relevance propagation & integrated gradient,
  • bayesian rule lists;
• eli5:
  • LIME,
  • permutation importance;
• shap (this one is really nice: it has great visualisations, examples and sample notebooks):
  • LIME,
  • Shapley sampling values,
  • DeepLIFT,
  • QII,
  • Layer-wise relevance propagation,
  • Shapley regression values,
  • Tree interpreter.

Fairness

• fairlearn;
• fairml;
• fairtest:
  • developed by Columbia University,
  • uses Python and calls R from Python – it’s fairly difficult to run because of this dependency;
• BlackBoxAuditing:
  • there’s a FAT* 2018 tutorial that uses this package here,
  • Certifying and removing disparate impact,
  • Auditing Black-box Models for Indirect Influence,
  • implements Gradient Feature Auditing;
• fairness-comparison:
  • three naive Bayes approaches for discrimination-free classification,
  • A comparative study of fairness-enhancing interventions in machine learning.