Researchers from LinkedIn open-source the FastTreeSHAP package, a Python module introduced in the paper, ‘Fast TreeSHAP: Accelerating SHAP Value Computation for Trees’. SHAP package allows for the efficient interpretation of tree-based machine learning models by estimating sample-level feature significance values. The package proposed two new algorithms: FastTreeSHAP v1 and FastTreeSHAP v2, both of which improve TreeSHAP’s computational efficiency by taking a different approach.
SHAP, LIME, and Integrated Gradient are examples of state-of-the-art sample-level model interpretation techniques. SHAP (Shapley Additive exPlanation) uses concepts from game theory and local explanations to produce SHAP values, which quantify the contribution of each feature. SHAP evaluates the average effect of adding an element to the model by considering all potential subsets of the other components.
FastTreeSHAP, a package developed by the Data Science Applied Research team at LinkedIn, has been widely used for explaining tree-based models due to its desirable theoretical properties and polynomial computational complexity. The current version of FastTreeSHAP package supports one-time usage scenarios. However, researchers are working towards extending it to multi-time usage scenarios (having a stable model in the backend and receiving new scoring data to be explained on a regular basis) with parallel computing.]
Sign up for your weekly dose of what's up in emerging technology.
For more details, click here