This paper proposes a framework for generating diverse counterfactual explanations for machine learning classifiers. The framework uses determinantal point processes (DPPs) to ensure diversity among counterfactual examples while also considering feasibility and proximity to the original input. The framework is evaluated on four real-world datasets and shows that it generates counterfactuals that are both diverse and well-approximate local decision boundaries, outperforming prior approaches. The framework also provides quantitative evaluation metrics for counterfactual examples, allowing for fine-tuning of the method for specific scenarios. The framework is implemented at https://github.com/microsoft/DiCE. The paper also discusses the importance of causal feasibility in counterfactual examples, noting that features are not independent and that changes to one feature may affect others. The paper concludes that counterfactual explanations can be useful for both end-users and model builders, as they can help identify biases in machine learning models.This paper proposes a framework for generating diverse counterfactual explanations for machine learning classifiers. The framework uses determinantal point processes (DPPs) to ensure diversity among counterfactual examples while also considering feasibility and proximity to the original input. The framework is evaluated on four real-world datasets and shows that it generates counterfactuals that are both diverse and well-approximate local decision boundaries, outperforming prior approaches. The framework also provides quantitative evaluation metrics for counterfactual examples, allowing for fine-tuning of the method for specific scenarios. The framework is implemented at https://github.com/microsoft/DiCE. The paper also discusses the importance of causal feasibility in counterfactual examples, noting that features are not independent and that changes to one feature may affect others. The paper concludes that counterfactual explanations can be useful for both end-users and model builders, as they can help identify biases in machine learning models.