BENCHMARK of REGRESSION: datasets and results
Chu Wei 2003.05.23
We tried three approaches on the following datasets:
1. Bayesian support vector regression using a unified loss function with evidence maximization (BSVR)
2. Standard Gaussian processes for regression with evidence maximization (GPR)
3. Classical Support vector regression using 5-fold cross validation (SVR)
Reference: W. Chu, S. S. Keerthi and C. J. Ong (2003), Bayesian support vector regression using a unified loss function. [pdf][ps] IEEE Transactions of Neural Networks 15(1):29-44
Source Code: http://www.gatsby.ucl.ac.uk/~chuwei/README.bisvm
1. BOSTON HOUSING (481/25, 100 folds) Click to Download
Results of BSVR with Gaussian Kernel on each fold
Results of BSVR with ARD Gaussian Kernel on each fold
Results of GPR with Gaussian Kernel on each fold
Results of GPR with ARD Gaussian Kernel on each fold
Results of SVR with Gaussian Kernel on each fold
*Note: each line contains the results (AAE ASE TIME) of one fold.
2. ABALONE (3000/1177, 10 folds) Click to Download
Results of BSVR with Gaussian Kernel on each fold
Results of BSVR with ARD Gaussian Kernel on each fold
Results of GPR with Gaussian Kernel on each fold
Results of GPR with ARD Gaussian Kernel on each fold
Results of SVR with Gaussian Kernel on each fold
*Note: each line contains the results (AAE ASE TIME) of one fold.
3. Computer Activity (2000/6192, 10 folds) Click to Download
Results of BSVR with Gaussian Kernel on each fold
Results of BSVR with ARD Gaussian Kernel on each fold
Results of GPR with Gaussian Kernel on each fold
Results of GPR with ARD Gaussian Kernel on each fold
Results of SVR with Gaussian Kernel on each fold
*Note: each line contains the results (AAE ASE TIME) of one fold.
5. ROBOT ARM originally generated by David J. C. MacKay is available at http://wol.ra.phy.cam.ac.uk/mackay/bigback/dat/