Shaping Up SHAP: Enhancing Stability through Layer-Wise Neighbor Selection
Authors: Gwladys Kelodjou, Laurence Rozé, Véronique Masson, Luis Galárraga, Romaric Gaudel, Maurice Tchuente, Alexandre Termier
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct a series of experiments to verify the effectiveness of ST-SHAP at mitigating the instability of Kernel SHAP. The experiments are conducted on a computer with a 12th generation Intel-i7 CPU (14 cores with HT, 2.3-4.7 GHz Turbo Boost). We use the official Kernel SHAP implementation provided by the authors (Lundberg and Lee 2017). ST-SHAP2 is based on this implementation. We use the scikit-learn python library3 to fit the black-box models that we aim to explain. |
| Researcher Affiliation | Academia | 1Univ Rennes, Inria, CNRS, IRISA UMR 6074, F35000 Rennes, France 2Univ Rennes, INSA Rennes, CNRS, Inria, IRISA UMR 6074, F35000 Rennes, France 3Sorbonne University, IRD, University of Yaound e I, UMI 209 UMMISCO, P.O. Box 337 Yaound e, Cameroon |
| Pseudocode | No | No pseudocode or clearly labeled algorithm block was found in the paper. |
| Open Source Code | Yes | ST-SHAP2 is based on this implementation. [footnote] 2https://github.com/gwladyskelodjou/st-shap |
| Open Datasets | Yes | We use several datasets from the UCI Machine Learning Repository and other popular real-world datasets, namely: Boston (Harrison Jr and Rubinfeld 1978) and Movie, which are regression datasets, and Default of Credit Card Clients (Yeh 2016), Adult (Becker and Kohavi 1996)4, Dry Bean5, Spambase (Hopkins et al. 1999), HELOC6, and Wisconsin Diagnostic Breast Cancer (Wolberg et al. 1995), which are classification datasets. Table 3 summarizes the information about the experimental datasets. |
| Dataset Splits | No | No specific training/test/validation split percentages, absolute sample counts for each split, or detailed cross-validation setup were explicitly provided. It mentions using a 'test set' but not details on how the data was split for training/validation/testing beyond that. |
| Hardware Specification | Yes | The experiments are conducted on a computer with a 12th generation Intel-i7 CPU (14 cores with HT, 2.3-4.7 GHz Turbo Boost). |
| Software Dependencies | No | The paper mentions using 'the official Kernel SHAP implementation' and 'scikit-learn python library' but does not specify version numbers for these or other software components, making it unreproducible. |
| Experiment Setup | No | While the paper describes the experimental protocol and general settings like datasets and models, it does not provide specific hyperparameter values (e.g., learning rate, batch size, number of epochs, optimizer settings) or other detailed training configurations for the models used in the experiments. |