SVM-rank Experiments on LETOR 3.0
=================================
Author: Thorsten Joachims
Date: 21 March 2009
The experiments were conducted with SVM-rank V1.00 available at
http://svmlight.joachims.org/svm_rank.html
SVM-rank solves the same optimization problem as SVM-light with the '-z p' option, but is much faster (on average around a second on my travel laptop with a 1.2Ghz Pentium).
For all datasets the learner was called with
svm_rank_learn -c -e 0.001 -l 1 train.txt model
where takes the values (0.00001,0.00002,0.00005,0.0001,0.0002,0.0005,0.001,0.002,0.005,0.01,0.02,0.05,0.1,0.2,0.5). For each fold, the value with the best MAP performance on the validation set was selected and its test set performance is reported. The value that achieved best validation set performance is shown in the filename of the results files and the log files of each run. In particular, they are:
2003_hp_dataset/QueryLevelNorm/cv_l1_e0.001/fold1_l1_c0.1_e0.001.log
2003_hp_dataset/QueryLevelNorm/cv_l1_e0.001/fold2_l1_c0.002_e0.001.log
2003_hp_dataset/QueryLevelNorm/cv_l1_e0.001/fold3_l1_c0.002_e0.001.log
2003_hp_dataset/QueryLevelNorm/cv_l1_e0.001/fold4_l1_c0.02_e0.001.log
2003_hp_dataset/QueryLevelNorm/cv_l1_e0.001/fold5_l1_c0.05_e0.001.log
2003_np_dataset/QueryLevelNorm/cv_l1_e0.001/fold1_l1_c0.2_e0.001.log
2003_np_dataset/QueryLevelNorm/cv_l1_e0.001/fold2_l1_c0.5_e0.001.log
2003_np_dataset/QueryLevelNorm/cv_l1_e0.001/fold3_l1_c0.005_e0.001.log
2003_np_dataset/QueryLevelNorm/cv_l1_e0.001/fold4_l1_c0.01_e0.001.log
2003_np_dataset/QueryLevelNorm/cv_l1_e0.001/fold5_l1_c0.2_e0.001.log
2003_td_dataset/QueryLevelNorm/cv_l1_e0.001/fold1_l1_c0.2_e0.001.log
2003_td_dataset/QueryLevelNorm/cv_l1_e0.001/fold2_l1_c0.1_e0.001.log
2003_td_dataset/QueryLevelNorm/cv_l1_e0.001/fold3_l1_c0.01_e0.001.log
2003_td_dataset/QueryLevelNorm/cv_l1_e0.001/fold4_l1_c0.005_e0.001.log
2003_td_dataset/QueryLevelNorm/cv_l1_e0.001/fold5_l1_c0.5_e0.001.log
2004_hp_dataset/QueryLevelNorm/cv_l1_e0.001/fold1_l1_c0.2_e0.001.log
2004_hp_dataset/QueryLevelNorm/cv_l1_e0.001/fold2_l1_c0.1_e0.001.log
2004_hp_dataset/QueryLevelNorm/cv_l1_e0.001/fold3_l1_c0.5_e0.001.log
2004_hp_dataset/QueryLevelNorm/cv_l1_e0.001/fold4_l1_c0.5_e0.001.log
2004_hp_dataset/QueryLevelNorm/cv_l1_e0.001/fold5_l1_c0.01_e0.001.log
2004_np_dataset/QueryLevelNorm/cv_l1_e0.001/fold1_l1_c0.005_e0.001.log
2004_np_dataset/QueryLevelNorm/cv_l1_e0.001/fold2_l1_c0.001_e0.001.log
2004_np_dataset/QueryLevelNorm/cv_l1_e0.001/fold3_l1_c0.2_e0.001.log
2004_np_dataset/QueryLevelNorm/cv_l1_e0.001/fold4_l1_c0.0001_e0.001.log
2004_np_dataset/QueryLevelNorm/cv_l1_e0.001/fold5_l1_c0.0001_e0.001.log
2004_td_dataset/QueryLevelNorm/cv_l1_e0.001/fold1_l1_c0.5_e0.001.log
2004_td_dataset/QueryLevelNorm/cv_l1_e0.001/fold2_l1_c0.2_e0.001.log
2004_td_dataset/QueryLevelNorm/cv_l1_e0.001/fold3_l1_c0.5_e0.001.log
2004_td_dataset/QueryLevelNorm/cv_l1_e0.001/fold4_l1_c0.5_e0.001.log
2004_td_dataset/QueryLevelNorm/cv_l1_e0.001/fold5_l1_c0.1_e0.001.log
OHSUMED/QueryLevelNorm/cv_l1_e0.001/fold1_l1_c0.0002_e0.001.log
OHSUMED/QueryLevelNorm/cv_l1_e0.001/fold2_l1_c0.002_e0.001.log
OHSUMED/QueryLevelNorm/cv_l1_e0.001/fold3_l1_c0.01_e0.001.log
OHSUMED/QueryLevelNorm/cv_l1_e0.001/fold4_l1_c0.02_e0.001.log
OHSUMED/QueryLevelNorm/cv_l1_e0.001/fold5_l1_c0.01_e0.001.log