Library: | nn |
See also: | nnrnet ann nnrpredict2 |
Quantlet: | nnrpredict | |
Description: | estimates the response for a given net and a dataset |
Usage: | yh = nnrpredict(x, net) | |
Input: | ||
x | n x p matrix, input variables | |
net | list, composed object from nnrnet | |
Output: | ||
yh | list, composed from yh.result and yh.hess | |
yh.result | n x q matrix, result variables | |
yh.hess | matrix, the hessian matrix |
library("nn") x = read("kredit") t = read("tkredit") randomize(55) y = x[,1] x = x[,2:21] x =(x-min(x))./(max(x)-min(x)) net = nnrnet(x, y, matrix(rows(x)), 10) yh = nnrpredict(x, net) yh.result
runs a neural network with 10 hidden units for the kredit data of Fahrmeier and Hammerle and computes the predicted values. Contents of ts [1,] "A 20 - 10 - 1 network:" [2,] "# weights : 221" [3,] "linear output : no" [4,] "error function: least squares" [5,] "log prob model: no" [6,] "skip links : no" [7,] "max. weight : 0.70" [8,] "decay : 0" [9,] "max. Iterat : 100" Contents of result [ 1,] 0.71751 [ 2,] 1 [ 3,] 1 [ 4,] 1 [ 5,] 0.99967 ... [ 996,] 5.4697e-06 [ 997,] 0 [ 998,] 1 [ 999,] 1 [1000,] 0