GLOBE-CE: A Translation Based Approach for Global Counterfactual Explanations

Authors: Dan Ley, Saumitra Mishra, Daniele Magazzeni

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental evaluation with publicly available datasets and user studies demonstrate that GLOBE-CE performs significantly better than the current state-of-the-art across multiple metrics (e.g., speed, reliability).
Researcher Affiliation Collaboration 1Harvard University 2J.P. Morgan AI Research.
Pseudocode Yes Algorithm 1 GLOBE-CE Framework Input: B, X, G, n, k, cost
Open Source Code No The paper does not provide any explicit statement about releasing source code or a link to a code repository for the described methodology.
Open Datasets Yes We employ four publicly available datasets to assess our methods: COMPAS (Larson et al., 2016), German Credit (Dua & Graff, 2019), Default Credit (Yeh & Lien, 2009) and HELOC (FICO, 2018).
Dataset Splits No The paper states 'We elect to train models on 80% of the data' and provides a 'No. Test' column in Table 5 (e.g., COMPAS: 4937 Train, 1235 Test), implying an 80/20 train/test split. However, it does not explicitly mention a separate validation set split or its percentage.
Hardware Specification No The paper mentions 'average CPU time taken in computing GCEs' as a metric for efficiency but does not specify any particular CPU model, GPU, or other hardware components used for running the experiments.
Software Dependencies No The paper mentions general libraries like 'Py Torch library', 'xgboost library', and 'sklearn library' but does not specify their version numbers, which are crucial for reproducibility.
Experiment Setup Yes We train 3 model types: Deep Neural Network (DNN), XGBoost (XGB), and Logistic Regression (LR). Best parameters for each dataset and model are chosen. ... Table 6. Summary of the DNNs used in our experiments. (Width, Depth, Dropout) ... Table 7. Summary of the XGB models used in our experiments. (Depth, Estimators, γ, α, λ) ... Table 8. Summary of the LR models used in our experiments. (Max Iterations, Class Weights)