compute_error.m

回声状态神经网络(ESN)是一种性能优异的递归神经网络

function err = compute_error(estimatedOutput, correctOutput)
% Computes the NRMSE between estimated and correct ESN outputs.
% 
% input arguments:
% estimatedOutput: array of size N1 x outputDimension, containing network
% output data. Caution: it is assumed that these are un-rescaled and
% un-shifted, that is, the transformations from the original data format
% via esn.teacherScaling and esn.teacherShift are undone. This happens
% automatically when the estimatedOutput was obtained from calling
% test_esn.
%
% correctOutput: array of size N2 x outputDimension, containing the
% original teacher data. 
%
% output:
% err: a row vector of NRMSE's, each corresponding to one of the output
% dimensions.
%
% If length(correctOutput) > length(estimatedOutput), the first
% elements from correctOutput are deleted. This accounts for cases where
% some (nForgetPoints many) initial transient data points were cancelled
% from estimatedOutput, as occurs in calls to test_esn.
%
% Version 1.0, June 6, 2006, H. Jaeger
% Copyright: Fraunhofer IAIS 2006 / Patents pending
  
  
nEstimatePoints = length(estimatedOutput) ; 

nForgetPoints = length(correctOutput) - nEstimatePoints ; 

correctOutput = correctOutput(nForgetPoints+1:end,:) ; 

correctVariance = var(correctOutput) ; 
meanerror = sum((estimatedOutput - correctOutput).^2)/nEstimatePoints ; 

err = (sqrt(meanerror./correctVariance)) ;