Efficient Top-K Feature Selection Using Coordinate Descent Method
Authors: Lei Xu, Rong Wang, Feiping Nie, Xuelong Li
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments exhibit the efficiency of CD-LSR, as well as the discrimination ability of l2,0-norm to identify informative features. |
| Researcher Affiliation | Academia | 1 School of Computer Science, Northwestern Polytechnical University, Xi an 710072, P.R. China 2 School of Artificial Intelligence, OPtics and Electro Nics (i OPEN), Northwestern Polytechnical University, Xi an 710072, P.R. China |
| Pseudocode | Yes | Algorithm 1: CD-LSR |
| Open Source Code | Yes | The source MATLAB code are available at: https://github.com/solerxl/Code For AAAI 2023. |
| Open Datasets | Yes | Dataset Type # Samp. # Dim. # Class SRBCT Bioinformatics 83 2308 4 USPS Handwritten Digit 9298 256 10 UMIST Handwritten Digit 575 644 20 JAFFE Human Face 213 676 10 COIL20 Object Image 1440 1024 20 Isolet Sound 1560 617 2 Table 1: Dataset descriptions. Dataset As shown in Table 1, we use six datasets to validate the performance of our method, including one bioinformatics dataset (i.e., SRBCT (Khan et al. 2001)), two handwritten digit datasets (i.e., USPS (Hull 1994) and UMIST (Hou et al. 2014)), one human face datasets (i.e., JAFFE (Lyons, Budynek, and Akamatsu 1999)), one object image dataset (i.e., COIL20 (Nene et al. 1996)), and one sound dataset (i.e., Isolet (Fanty and Cole 1990)). |
| Dataset Splits | Yes | For each dataset, we employ the five-fold cross-validation method for parameter tuning and performance evaluation. |
| Hardware Specification | Yes | All algorithms are test on a Windows machine with 3.4GHz, i7-6700CPU and 32 GB RAM memory. |
| Software Dependencies | No | The paper mentions 'MATLAB code' but does not specify the version of MATLAB or any other software libraries with their version numbers required for reproducibility. |
| Experiment Setup | Yes | For each dataset, we employ the five-fold cross-validation method for parameter tuning and performance evaluation. Since l2,0-norm is non-convex, the solution of the proposed method depends on the initialization. Similar to (Pang et al. 2019), we run each method 40 times and report the maximal accuracy as the final performance. In each trial, the solutions of all methods are initialized with the same random matrix. We set k = 5 and run all methods 40 times, with the average computational time recorded as their final result. We regard the objective function value converges as long as the rate of change of the objective function value is less than 0.01%. We also set the maximum number of iterations to 1000 so as to prevent an excessively long optimization process. |