L-Shapley and C-Shapley: Efficient Model Interpretation for Structured Data
Authors: Jianbo Chen, Le Song, Martin J. Wainwright, Michael I. Jordan
ICLR 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate the performance of L-Shapley and C-Shapley on real-world data sets involving text and image classification. We compare L-Shapley and C-Shapley with several competitive algorithms for instancewise feature importance scoring on black-box models, including the regressionbased approximation known as Kernel SHAP (Lundberg & Lee, 2017), Sample Shapley (Štrumbelj & Kononenko, 2010), and the LIME method (Ribeiro et al., 2016). |
| Researcher Affiliation | Collaboration | UC Berkeley , Georgia Institute of Technology , Ant Financial , Voleon Group |
| Pseudocode | No | The paper describes algorithms with mathematical formulas but does not include structured pseudocode blocks or sections explicitly labeled 'Algorithm'. |
| Open Source Code | No | The paper does not provide a direct link to a code repository, nor does it explicitly state that the source code for the methodology is being released or available in supplementary materials. |
| Open Datasets | Yes | IMDB Review with Word-CNN The Internet Movie Review Dataset (IMDB) is a dataset of movie reviews for sentiment classification (Maas et al., 2011)... AG news with Char-CNN The AG news corpus... (Zhang et al., 2015)... Yahoo! Answers with LSTM The corpus of Yahoo! Answers Topic Classification Dataset... (Zhang et al., 2015)... MNIST The MNIST data set... (Le Cun et al., 1998)... CIFAR10 The CIFAR10 data set (Krizhevsky, 2009)... |
| Dataset Splits | Yes | IMDB Review (Maas et al., 2011) 2 25,000 25,000... AG s News (Zhang et al., 2015) 4 120,000 7,600... Yahoo! Answers (Zhang et al., 2015) 10 1,400,000 60,000... |
| Hardware Specification | Yes | taking less than one second in Tensor Flow on a Tesla K80 GPU for all the three models. |
| Software Dependencies | No | The paper mentions software like 'Tensor Flow', 'rmsprop', 'Adam optimizer', and 'Dropout' but does not provide specific version numbers for any of these dependencies. |
| Experiment Setup | Yes | We choose zero paddings as the reference point for all methods, and make 4 d model evaluations, where d is the number of words for each input. Given the average length of each input (see Table 1), this choice controls the number of model evaluations under 1, 000... For L-Shapley, we are able to consider the interaction of each word i with the two neighboring words in N1(i) given the budget. For C-Shapley, the budget allows the regression-based version to evaluate all n-grams with n 4. |