DALEX
In the Dimensions of explainability section of Chapter 1, Foundational Concepts of Explainability Techniques, we discussed the four different dimensions of explainability – data, model, outcome, and end user. Most explainability frameworks such as LIME, SHAP, and TCAV provide model-centric explainability.
DALEX (moDel Agnostic Language for Exploration and eXplanation) is one of the very few widely used XAI frameworks that tries to address most of the dimensions of explainability. DALEX is model-agnostic and can provide some metadata about the underlying dataset to give some context to the explanation. This framework gives you insights into the model performance and model fairness, and it also provides global and local model explainability.
The developers of the DALEX framework wanted to comply with the following list of requirements, which they have defined in order to explain complex black-box algorithms:
- Prediction's justifications: According to...