Consequence-Aware Sequential Counterfactual Generation

Counterfactuals have become a popular technique nowadays for interacting with black-box machine learning models and understanding how to change particular instance obtain desired outcome from the model. However, most existing approaches assume instant materialization of these changes, ignoring that they may require effort specific order application. Recently, methods been proposed also consider in which actions are applied, leading so-called sequential counterfactual generation problem. In this work, we propose model-agnostic method generation. We formulate task as multi-objective optimization problem present genetic algorithm approach find optimal sequences counterfactuals. Our cost model considers not only direct effect an action, but its consequences. Experimental results show compared state-of-the-art, our generates less costly solutions, is more efficient provides user diverse set solutions choose from.