Learning Bayesian Network Classifiers to Minimize the Class Variable Parameters
Authors: Shouta Sugahara, Koya Kato, Maomi Ueno
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | This section describes experiments conducted to demonstrate the benefits of the proposed method. First, we compare the classification accuracies of the following ten methods using the benchmark datasets in Table 1. |
| Researcher Affiliation | Academia | Shouta Sugahara, Koya Kato, Maomi Ueno The University of Electro-Communications sugahara@ai.lab.uec.ac.jp, kato@ai.lab.uec.ac.jp, ueno@ai.lab.uec.ac.jp |
| Pseudocode | No | The paper describes the depth-first branch and bound algorithm and illustrates it with an example in Figure 1b, but it does not provide formal pseudocode or an algorithm block. |
| Open Source Code | No | The source code is available at http://www.ai.lab.uec.ac.jp/software/. To ensure double-blind review, the code will be made publicly available after the manuscript is reviewed and accepted for publication. |
| Open Datasets | Yes | This experiment uses 24 real datasets from the UCI repository (Lichman 2013). |
| Dataset Splits | Yes | We ascertain the ESS N {1, 10, 100, 1, 000} of BDeu scores in GBN-BDeu, ANB-BDeu, fs ANB-BDeu, Proposed(BFS), and Proposed(DFB&B) using ten-fold cross validation to obtain the highest classification accuracy. |
| Hardware Specification | Yes | As described throughout this paper, our experiments are conducted on a computer with a 3.2 GHz 16-core processor and 128 GB of memory. |
| Software Dependencies | No | The paper states: "The proposed methods are implemented in C++. The other methods are implemented in Java." It does not provide specific version numbers for these languages or any libraries used, which is required for reproducibility. |
| Experiment Setup | Yes | We ascertain the ESS N {1, 10, 100, 1, 000} of BDeu scores in GBN-BDeu, ANB-BDeu, fs ANB-BDeu, Proposed(BFS), and Proposed(DFB&B) using ten-fold cross validation to obtain the highest classification accuracy. Continuous variables are discretized using a standard discretization algorithm proposed by (Fayyad and Irani 1993). In addition, data with missing values are removed from the datasets. We use EAP estimators with N ijk = 1/(riqi) as conditional probability parameters of the respective classifiers (Ueno 2010, 2011). |