pseudoinverse.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 outputWeights = pseudoinverse(stateCollectMat, teachCollectMat)

% PSEUDOINVERSE computes the outputWeights using the standard pseudoinverse
% operation
%
% inputs:
% stateCollectMat = matrix of size (nTrainingPoints-nForgetPoints) x
% nInputUnits + nInternalUnits 
% stateCollectMat(i,j) = internal activation of unit j after the 
% (i + nForgetPoints)th training point has been presented to the network
% teacherCollectMat is a nSamplePoints * nOuputUnits matrix that keeps
% the expected output of the ESN
% teacherCollectMat is the transformed(scaled, shifted etc) output see
% compute_teacher for more documentation
%
% output:
% outputWeights = vector of size (nInputUnits + nInternalUnits) x 1
% containing the learnt weights
%


% Created April 30, 2006, D. Popovici
% Copyright: Fraunhofer IAIS 2006 / Patent pending

  
outputWeights = (pinv(stateCollectMat)*teachCollectMat)' ;