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
, orFidex
if necessary.