BENCHMARK of ORDINAL REGRESSION: datasets and results
Chu Wei 2004.08.11
Reference: W. Chu and Z. Ghahramani (2005), Gaussian processes for ordinal regression. [pdf][ps] JMLR 6(Jul):1019--1041
Source Code: http://www.gatsby.ucl.ac.uk/~chuwei/README.gpor
We tried three approaches on the following datasets:
1. Gaussian processes for ordinal regression with Laplace approximation and evidence maximization (MAP)
2. Gaussian processes for ordinal regression with expectation propagation and variational methods (EP)
3. Support vector machines for ordinal regression using k-fold cross validation (SVM)
All the results of SET I: http://www.gatsby.ucl.ac.uk/~chuwei/result/allinone.zip
1. DIABETES (30/13, 20 folds) Click to Download
Results of MAP with Gaussian Kernel on each fold 5 bins 10 bins
Results of EP with Gaussian Kernel on each fold 5 bins 10 bins
Results of SVM with Gaussian Kernel on each fold 5 bins 10 bins
*Note: each line contains the results (Zero-OneError AbsoluteError MeanAverageError) of one fold.
2. PYRIMIDINES (50/24, 20 folds) Click to Download
Results of MAP with Gaussian Kernel on each fold 5 bins 10 bins
Results of EP with Gaussian Kernel on each fold 5 bins 10 bins
Results of SVM with Gaussian Kernel on each fold 5 bins 10 bins
*Note: each line contains the results (Zero-OneError AbsoluteError MeanAverageError) of one fold.
3. TRIAZINES (100/86, 20 folds) Click to Download
Results of MAP with Gaussian Kernel on each fold 5 bins 10 bins
Results of EP with Gaussian Kernel on each fold 5 bins 10 bins
Results of SVM with Gaussian Kernel on each fold 5 bins 10 bins
*Note: each line contains the results (Zero-OneError AbsoluteError MeanAverageError) of one fold.
4. WISCONSIN (130/64, 20 folds) Click to Download
Results of MAP with Gaussian Kernel on each fold 5 bins 10 bins
Results of EP with Gaussian Kernel on each fold 5 bins 10 bins
Results of SVM with Gaussian Kernel on each fold 5 bins 10 bins
*Note: each line contains the results (Zero-OneError AbsoluteError MeanAverageError) of one fold.
5. MACHINE CPU (150/59, 20 folds) Click to Download
Results of MAP with Gaussian Kernel on each fold 5 bins 10 bins
Results of EP with Gaussian Kernel on each fold 5 bins 10 bins
Results of SVM with Gaussian Kernel on each fold 5 bins 10 bins
*Note: each line contains the results (Zero-OneError AbsoluteError MeanAverageError) of one fold.
6. AUTO MPG (200/192, 20 folds) Click to Download
Results of MAP with Gaussian Kernel on each fold 5 bins 10 bins
Results of EP with Gaussian Kernel on each fold 5 bins 10 bins
Results of SVM with Gaussian Kernel on each fold 5 bins 10 bins
*Note: each line contains the results (Zero-OneError AbsoluteError MeanAverageError) of one fold.
7. BOSTON HOUSING (300/206, 20 folds) Click to Download
Results of MAP with Gaussian Kernel on each fold 5 bins 10 bins
Results of EP with Gaussian Kernel on each fold 5 bins 10 bins
Results of SVM with Gaussian Kernel on each fold 5 bins 10 bins
*Note: each line contains the results (Zero-OneError AbsoluteError MeanAverageError) of one fold.
8. STOCK DOMAIN (600/350, 20 folds) Click to Download
Results of MAP with Gaussian Kernel on each fold 5 bins 10 bins
Results of EP with Gaussian Kernel on each fold 5 bins 10 bins
Results of SVM with Gaussian Kernel on each fold 5 bins 10 bins
*Note: each line contains the results (Zero-OneError AbsoluteError MeanAverageError) of one fold.
9. ABALONE (1000/3177, 20 folds) Click to Download
Results of MAP with Gaussian Kernel on each fold 5 bins 10 bins
Results of EP with Gaussian Kernel on each fold 5 bins 10 bins
Results of SVM with Gaussian Kernel on each fold 5 bins 10 bins
*Note: each line contains the results (Zero-OneError AbsoluteError MeanAverageError) of one fold.
All the results of SET II: http://www.gatsby.ucl.ac.uk/~chuwei/result/allinone2.zip
10. BankDomain (1) Click to Download
11. BankDomain (2) Click to Download
12. Computer Activity (1) Click to Download
13. Computer Activity (2) Click to Download
14. California Housing Click to Download
15. CensusDomain (1) Click to Download
16. CensusDomain (2) Click to Download
Run the Matlab script to generate the data folds we used by 10 equal-frequency binning.
In Matlab: splitequfre(data file name, 100, training size, 10), where 'training size' is listed in Table 1 of Chu and Ghahramani (2005) for each dataset.