_PROBLEM CoEPrA-2006_Regression_001 _GROUP_NAME Cesare Furlanello _GROUP_MEMBERS Stefano Merler Giuseppe Jurman Silvano Paoli Annalisa Barla Cesare Furlanello _ADDRESS ITC-irst, Via Sommarive 18, I-38050 Povo, Trento, Italy _MODELING_PROCEDURE The predictions result from the average of 100 Terminated-ramp regularization networks. The 100 models were obtained by randomly splitting 100 times the calibration data into training and validation sets, consisting of 80 and 9 points respectively. 2 Calibration model and statistics 2.1 Descriptors We used the original descriptors but for the 7 desciptors (namely 89,673,732,789,2661,718,3303) with null variance, which were eliminated. 2.2 Validation For each terminated-ramp regularization network the only free parameter to be optimized is the regularizer. For each training set, the optimal value of the regularizer were optimized by minimizing the MSE on the corresponding validation set. The following 401 values of the regularizer were considered: lambda = 1.1^l , l=-200,...,200. 2.3 Parameters We can not report all the model parameters. In fact, each terminated ramp function k_ij depends on 5781 parameters . The terminated ramp kernel depends on the combination of n(n - 1)/2 terminated ramp functions, where n is the number of training data. This means that, for each training set (n = 80) the kernel depends on 18264800 parameters. Of course, we could report the optimal values of the regularizer and the estimated regression coeficients c_i for each training set, but they are meaningless without the kernel. References [MJ06] S. Merler and G. Jurman. Terminated ramp - support vector machines: a data driven kernel. Neural Networks, 2006. in press. _PREDICTION Obj_00001 5.722 Obj_00002 5.415 Obj_00003 3.819 Obj_00004 4.814 Obj_00005 7.025 Obj_00006 6.075 Obj_00007 4.843 Obj_00008 4.319 Obj_00009 5.236 Obj_00010 4.812 Obj_00011 6.188 Obj_00012 4.815 Obj_00013 5.866 Obj_00014 4.425 Obj_00015 6.669 Obj_00016 5.882 Obj_00017 4.209 Obj_00018 5.592 Obj_00019 4.102 Obj_00020 3.438 Obj_00021 3.359 Obj_00022 5.283 Obj_00023 4.801 Obj_00024 4.865 Obj_00025 4.101 Obj_00026 6.536 Obj_00027 6.021 Obj_00028 5.456 Obj_00029 5.697 Obj_00030 4.583 Obj_00031 5.423 Obj_00032 5.641 Obj_00033 5.612 Obj_00034 6.052 Obj_00035 4.835 Obj_00036 6.288 Obj_00037 4.470 Obj_00038 5.571 Obj_00039 5.356 Obj_00040 4.909 Obj_00041 5.908 Obj_00042 5.080 Obj_00043 6.758 Obj_00044 5.019 Obj_00045 5.590 Obj_00046 5.501 Obj_00047 3.781 Obj_00048 5.928 Obj_00049 4.483 Obj_00050 4.881 Obj_00051 5.902 Obj_00052 5.817 Obj_00053 5.095 Obj_00054 4.164 Obj_00055 5.568 Obj_00056 5.772 Obj_00057 5.030 Obj_00058 4.573 Obj_00059 6.660 Obj_00060 4.113 Obj_00061 6.127 Obj_00062 4.093 Obj_00063 5.622 Obj_00064 5.444 Obj_00065 5.777 Obj_00066 6.080 Obj_00067 6.091 Obj_00068 5.348 Obj_00069 5.910 Obj_00070 4.511 Obj_00071 4.387 Obj_00072 5.354 Obj_00073 5.682 Obj_00074 5.354 Obj_00075 5.401 Obj_00076 4.519 Obj_00077 5.856 Obj_00078 5.873 Obj_00079 4.766 Obj_00080 5.999 Obj_00081 5.850 Obj_00082 5.613 Obj_00083 3.682 Obj_00084 5.938 Obj_00085 6.888 Obj_00086 5.328 Obj_00087 5.467 Obj_00088 5.293