Fast Algorithms for Stackelberg Prediction Game with Least Squares Loss
Authors: Jiali Wang, He Chen, Rujun Jiang, Xudong Li, Zihao Li
ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we conduct numerical experiments on both synthetic and real world datasets to verify the superior performance of our proposed algorithms in terms of both the computational time and the learning accuracy. |
| Researcher Affiliation | Academia | 1School of Data Science, Fudan University, China 2School of Mathematical Sciences, Fudan University, China. |
| Pseudocode | Yes | Algorithm 1 SOCP method for solving (2) |
| Open Source Code | Yes | Our code is available at https://github.com/JialiWang12/SPGLS. |
| Open Datasets | Yes | We first test our methods on the red wine dataset (Cortez et al., 2009), which contains 1599 instances each with 11 features. ... We next compare our algorithms on the blogfeedback dataset from the UCI data repository (Dua & Graff, 2019). |
| Dataset Splits | Yes | to evaluate the learning accuracy of the algorithms, we perform 10-fold cross-validation and compare their average MSE for 40 different values of the parameter γ [1 10 3, 0.75] in (2). |
| Hardware Specification | Yes | All simulations are implemented using MATLAB R2019a on a PC running Windows 10 Intel(R) Xeon(R) E5-2650 v4 CPU (2.2GHz) and 64GB RAM. |
| Software Dependencies | Yes | We apply the powerful commercial solver MOSEK (MOSEK, 2021) to solve all the SDPs and SOCPs in the bisection method and ours. All simulations are implemented using MATLAB R2019a on a PC running Windows 10 Intel(R) Xeon(R) E5-2650 v4 CPU (2.2GHz) and 64GB RAM. |
| Experiment Setup | Yes | The function make regression in scikit-learn (Pedregosa et al., 2011) is used to build artificial datasets of controlled size and complexity. In particular, we specify the noise as 0.1, which is the standard deviation of the Gaussian noise applied to the output y, and all other arguments are set as default. ... In all tests, the parameter γ is set as 0.01. ... In the test, the algorithm would not be run in larger dimension case (denoted by ), if its wallclock time at current dimension exceeds 1800 seconds. |