do-SHAP explanations made practical with Structural Causal Models and a novel algorithm to accelerate do-SHAP

We propose a novel feature attribution method that disentangles attributions based on a feature's value and its position within a sequence.