The arboretum procedure
Yüklə 3,07 Mb. Pdf görüntüsü
|
- Bu səhifədəki naviqasiya:
- Levenberg-Marquardt Termination Criteria
- Levenberg-Marquardt Iteration Log
- Parameter Estimates (MLP Weights)
- List Report of Selected Variables in the Score OUTFIT= Data Set
12 H3_C1 -5.975478 -0.00229 H3 -> C1 13 H1_C2 4.632142 -0.00268 H1 -> C2 14 H2_C2 4.809949 -0.00996 H2 -> C2 15 H3_C2 -4.803055 -0.00695 H3 -> C2 16 BIAS_C1 -11.870050 0.00869 BIAS -> C1 17 BIAS_C2 -7.664741 0.00352 BIAS -> C2 Value of Objective Function = 0.1531232148
This section lists the termination criteria for the Levenberg-Marquardt optimization. MLP with 3 Hidden Units Levenberg-Marquardt Optimization *** Termination Criteria *** Minimum Iterations . . . . . . . . . . . . . . . -1 Maximum Iterations . . . . . . . . . . . . . . . 100 Maximum Function Calls. . . . . . . . . . . . . . 2147483647 Maximum CPU Time . . . . . . . . . . . . . . . . 604800 ABSGCONV Gradient Criterion . . . . . . . . . . . 0.0000100 GCONV Gradient Criterion . . . . . . . . . . . . 1E-8 GCONV2 Gradient Criterion . . . . . . . . . . . . 0 ABSFCONV Function Criterion . . . . . . . . . . . 0 FCONV Function Criterion . . . . . . . . . . . . 0.0001000 FCONV2 Function Criterion . . . . . . . . . . . . 0 FSIZE Parameter . . . . . . . . . . . . . . . . . 0 ABSXCONV Parameter Change Criterion . . . . . . . 0 XCONV Parameter Change Criterion . . . . . . . . 0 XSIZE Parameter . . . . . . . . . . . . . . . . . 0 ABSCONV Function Criterion . . . . . . . . . . . -1.341E154 *** Other Control Parameters *** Trust Region Initial Radius Factor . . . . . . . 1.00000 Singularity Tolerance (SINGULAR) . . . . . . . . 1E-8
At the start of optimization, the procedure lists the number of active constraints, the current value of the objective function, the maximum gradient element, and the radius. The iteration history includes the following: the iteration number (iter) q the number of iteration restarts (rest) q the number of active constraints (act) q the value of the objective function (optcrit) q the difference between the adjacent objective functions (difcrit) q the maximum of the absolute (projected) gradient components (maxgrad) q the value of lamba q the value of rho q The optimization results section lists information specific for the optimization technique. MLP with 3 Hidden Units Levenberg-Marquardt Optimization Scaling Update of More (1978) Number of Parameter Estimates 17 Optimization Start: Active Constraints= 0 Criterion= 0.153
Maximum Gradient Element= 0.046 Radius= 1.000 Iter rest nfun act optcrit difcrit maxgrad lambda rho 1 0 5 0 0.1007 0.0524 0.0708 0.0138 0.582 2 0 9 0 0.0883 0.0124 0.0589 0.417 0.330 3 0 15 0 0.0546 0.0336 0.0493 0.00993 0.674 4 0 19 0 0.0255 0.0291 0.0726 0.00154 0.667 5 0 21 0 0.0233 0.00227 0.136 0 0.0960 6 0 23 0 0.003321 0.0200 0.0154 0 0.897 7 0 24 0 0.000742 0.00258 0.00889 5.65E-7 0.816 8 0 25 0 0.000283 0.000459 0.00683 0 0.636 9 0 26 0 0.0000632 0.00022 0.00147 0 0.789 10 0 27 0 6.4334E-6 0.000057 0.00018 0 0.908 11 0 28 0 1.1769E-6 5.257E-6 0.00006 0 0.823 12 0 29 0 2.6787E-7 9.09E-7 0.00002 0 0.776 13 0 30 0 1.2592E-7 1.42E-7 0.00002 0 0.532 14 0 32 0 1.6879E-8 1.09E-7 2.89E-6 308E-15 0.869 Optimization Results: Iterations= 14 Function Calls= 33 Jacobian Calls= 15 Active Constraints= 0 Criterion= 1.6879056E-8 Maximum Gradient Element= 2.8916E-6 Lambda= 3.08E-13 Rho= 0.8685 Radius= 26.94 NOTE: ABSGCONV convergence criterion satisfied. Parameter Estimates (MLP Weights) Optimization Results Parameter Estimates ------------------------------------------------------------------------------ Parameter Estimate Gradient Label ------------------------------------------------------------------------------ 1 X_H1 -0.377915 1.2972E-6 X -> H1 2 Y_H1 -0.665105 2.8916E-6 Y -> H1 3 X_H2 1.720212 8.9422E-8 X -> H2 4 Y_H2 -0.043669 1.63254E-7 Y -> H2 5 X_H3 0.487788 -3.1918E-7 X -> H3 6 Y_H3 -0.801489 -5.7188E-7 Y -> H3 7 BIAS_H1 0.875418 1.87012E-6 BIAS -> H1 8 BIAS_H2 1.340638 -1.1246E-7 BIAS -> H2 9 BIAS_H3 -0.649604 -7.5073E-8 BIAS -> H3 10 H1_C1 388.611382 -1.7823E-9 H1 -> C1 11 H2_C1 289.584211 -1.6285E-9 H2 -> C1 12 H3_C1 -309.435371 1.04029E-9 H3 -> C1 13 H1_C2 177.046552 -1.9849E-9 H1 -> C2 14 H2_C2 118.043485 1.37679E-8 H2 -> C2 15 H3_C2 -149.150661 -1.1747E-8 H3 -> C2 16 BIAS_C1 -580.555136 -2.0973E-9 BIAS -> C1 17 BIAS_C2 -213.290409 1.39802E-8 BIAS -> C2 Value of Objective Function = 1.6879056E-8 List Report of Selected Variables in the Score OUTFIT= Data Set The example PROC PRINT report of the OUTFIT = data set lists selected fit statistics for the training data. By default, the OUTFIT data set contains two observations. These observations are distinguished by the value of the _NAME_ variable: _NAME_ = 'Target Variable Name' q _NAME_ = 'OVERALL' q Because a WHERE statement was used to select only values of _NAME_ = 'OVERALL', this report contains a single observation. Yüklə 3,07 Mb. Dostları ilə paylaş:
|