Learning to Binarize Continuous Features for Neuro-Rule Networks
Authors: Wei Zhang, Yongxiang Liu, Zhuo Wang, Jianyong Wang
IJCAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct comprehensive experiments on public datasets and demonstrate the effectiveness of Auto Int in boosting the performance of NRNs. 4 Experiments 4.1 Experimental Setup |
| Researcher Affiliation | Academia | School of Computer Science and Technology, East China Normal University; Shanghai Institute for AI Education; Department of Computer Science and Technology, Tsinghua University |
| Pseudocode | No | The paper describes computational procedures (e.g., Equation 1, 3, 4, 6, 7) and provides a high-level overview of Auto Int's modules, but it does not include a formally labeled 'Pseudocode' or 'Algorithm' block. |
| Open Source Code | Yes | The source code of Auto Int is available at https://github. com/yxliu99/Auto Int. |
| Open Datasets | Yes | We use six public datasets from the UCI dataset repository1. All of them satisfy to contain some continuous features for testing the effectiveness of binarization methods. Table 1 summarizes the statistics of the 6 datasets. 1https://archive.ics.uci.edu/ml/datasets.php |
| Dataset Splits | Yes | To have a reliable performance evaluation, we adopt 5-fold cross-validation and report the average performance, the same as [Wang et al., 2021]. When there are hyperparameters needing to be tuned, we use 80% of the training set for optimization and the left for validation. |
| Hardware Specification | No | The paper does not provide any specific hardware details such as GPU models, CPU types, or memory specifications used for the experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers (e.g., 'PyTorch 1.9', 'Python 3.8'). It only states 'The source code of Auto Int is available at https://github. com/yxliu99/Auto Int.', which might imply dependencies are within the repo, but they are not explicitly listed in the text. |
| Experiment Setup | Yes | For the proposed Auto Int, both the temperature τ1 and τ2 are selected in {500, 1000, 2000}, K is searched in {5, 10, 15, 20, 30, 50}, and λ is tuned in {0.005, 0.01, 0.05, 0.1, 0.2, 0.5}. The intervals are initialized by Fre Int, if not otherwise stated. Besides, all the continuous feature values are pre-processed through standard normalization. For the neuro-rule network RRL and XGBoost (XGBoost directly uses the continuous features as input), we also follow [Wang et al., 2021] to set their hyper-parameters. |