trust.describable
Module Contents
Classes
A Trustworthy-AI interface that defines the capabilities of objects (e.g., Models, Graphs) |
- class trust.describable.Describable
A Trustworthy-AI interface that defines the capabilities of objects (e.g., Models, Graphs) that are Describable, i.e., they can self-describe their properties and behaviors. For example, a `Describable Model should be able to report the data that was used to train it, provide an importance-ranked list of its input features on a global basis, etc.
- property description(self)
- describe(self, constituency=Constituency.ANY, aspect=Aspect.ANY)
Returns a description of the model’s behavior and properties based on Who’s asking for what.
- Parameters:
constituent : Constituency: The Constituency asking for the explanation Who aspect : The Aspect of the question. What
- Returns:
out : Description of Model’s behavior and properties