hirohashi / wtopk

Learning Weighted Top-k Support Vector Machine, ACML 2019

Learning Weighted Top-k Support Vector Machine

This repository provides the core implementation of our paper entitled "Learning Weighted Top-k Support Vector Machine" presented in ACML 2019.

Dependencies

This implementation requires the following softwares.

• Python3
• Numpy (version >= 1.15 is required for "take_along_axis" function)
• Scipy

Usage

Basically, the weighted top-k SVM training for the dummy data (k.demo450_01.mat) with the regularization parameter "C=10.0", and "exponentially decreased weights of k=3" can be executed by the following command.

$ python train_wtopk.py --dataset k.demo450_01.mat --c_svm 10.0 --rho_dist topk_exp --rho_param 3

For technical details, please check our paper.