Skip to content

Fidex algorithms

The Fidex ensemble consists of four algorithms, all derived from the core eponymous algorithm. The Fidex algorithm itself focuses on extracting local rules to explain the predictions of a pre-trained model for given data samples. Each of the other algorithms builds upon this foundation to offer additional functionalities. For more details on the Fidex algorithm, you can refer to this paper.

Architecture

The ensemble is built as shown below:

graph TD;
    Fidex[Fidex] 
    FidexGloRules[FidexGloRules] 
    FidexGloStats[FidexGloStats] 
    FidexGlo[FidexGlo]

    Fidex --> FidexGloRules
    FidexGloRules --> FidexGloStats
    Fidex --> FidexGlo
    FidexGloRules --> FidexGlo

Each algorithm has its purpose:

  • Fidex: Extracts a local explanation rule for one or more given samples.
  • FidexGloRules: Extracts a set of global explanation rules from a given training dataset.
  • FidexGloStats: Computes statistics from a given global ruleset.
  • FidexGlo: Extracts an explanation for each given sample by generating one or more explaining rules, using the ruleset created by FidexGloRules, or Fidex if necessary.