function [x,resnorm,residual,exitflag,output]=LM_mat(InpPat,TarPat, Wini,n_iter,NP)
%
%[crit,WW]=LM_mat(InpPat,TarPat, Wini,n_iter,NP)
%
%This function implements a Levenberg-Marquardt type of learning rule, using the function lsqnonçin available in the optimization toolbox
%
%Input Arguments:	InpPat - Matrix of input data
%						TarPat - Matrix of output desired data
%						Wini - vector of initial weights
%						n_iter - how many iterations
%						NP - toplogy of the MLP - Two hidden layers network, with just 1 output
%							The weights are arranged as:
%								ni*NP(1)
%								NP(1)`*NP(2)
%								NP(2)*1
%								Bias for the 1st hidden layer
%								Bias for the 2nd hidden layer
%								Bias for the output neuron
%
%Output arguments:x - final weights

options = optimset('LargeScale','off','Jacobian','on','LevenbergMarquardt','on','TolFun',10^-8,'Display','iter','MaxIter',n_iter);
[x,resnorm,residual,exitflag,output] = lsqnonlin('funlm',Wini,[],[],options,InpPat,TarPat,NP);