Active Feature Selection for the Mutual Information Criterion
Authors: Shachar Schnapp, Sivan Sabato9497-9504
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We explain and experimentally study the choices that we make in the algorithm, and show that they lead to a successful algorithm, compared to other more naive approaches. and 6 Experiments We first report experiments for the single-feature estimation problem, comparing the approaches suggested in Section 4. These clearly show that the I-CP approach is preferable. Then, we report experiments on several benchmark data sets for the AFS algorithm, comparing it to three natural baselines. We further report ablation studies, demonstrating the necessity of each of the mechanisms of the algorithm. |
| Researcher Affiliation | Academia | Shachar Schnapp and Sivan Sabato Department of Computer Science, Ben-Gurion University of the Negev, Beer-Sheva, Israel schnapp@post.bgu.ac.il, sabatos@cs.bgu.ac.il |
| Pseudocode | Yes | Algorithm 1 AFS: Active Feature Selection for the Mutual Information Criterion |
| Open Source Code | Yes | The code is available at the following url: https://github.com/Shachar Schnapp/Active Feature Selection and Python 3.6 code for all experiments is available at: https://github. com/Shachar Schnapp/Active Feature Selection. |
| Open Datasets | Yes | For the real-data experiments, we tested the 14 features of the Adult data set (Dua and Graff 2019) with their true pv values. and In addition, we tested on the MUSK data set (Dua and Graff 2019), and on the MNIST data set (Le Cun and Cortes 2010) for three pairs of digits. |
| Dataset Splits | No | The paper does not explicitly provide training/validation/test dataset splits, specific percentages, or detail a cross-validation setup. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running its experiments. It only states: "All experiments can be run on a standard personal computer.". |
| Software Dependencies | No | The paper mentions "Python 3.6 code" but does not provide specific ancillary software details like library or solver names with version numbers. |
| Experiment Setup | No | The paper mentions parameters like `k`, `B`, `δ`, `Λ`, and `ψ`, and states that `δ = 0.05` and `Λ = 30` were selected after testing. However, it does not provide comprehensive specific experimental setup details such as hyperparameter values (e.g., learning rate, batch size, optimizer settings) for training the models. |