rsomhebbv02.m

a neural network,Recursive SOM and Marge SOM ,can be use for time series and data fit.

function Net = RSOMHebbV02(Net , I , RSOMState , WinnerInd)
%   RSOMHebbV02     Hebbian learning for RSOM network, the same as V01 but
%                   without learning rates decay.
%   
%   Without parameters decay
%   ---------------------------------------------------------
%   Amir Reza Saffari Azar Alamdari
%   http://www.ymer.org/main.htm , [e-mail]
%   ---------------------------------------------------------

%-----Initializing
INNum       = Net.Dim.INNum;
RSOMNum     = Net.Dim.RSOMNum;
RSOMSize    = Net.Dim.RSOMSize;

WINRSOM     = Net.W.WINRSOM;
WRSOM       = Net.W.WRSOM;

RSOMType    = Net.RSOM.RSOMType;
RSOMPosD    = Net.RSOM.RSOMPosD;
RSOMNeiFun  = Net.RSOM.RSOMNeiFun;
RSOMNeiFunP = Net.RSOM.RSOMNeiFunP;

RSOMUnSLStepIN  = Net.Learn.RSOMUnSLStepIN;
RSOMUnSLStepC   = Net.Learn.RSOMUnSLStepC;

%-----Learning
WinnerDist  = RSOMPosD(WinnerInd , :);

Epoch       = Net.State.Epoch;
MaxEpoch    = Net.State.MaxEpoch;

switch RSOMNeiFun
    
    case 'Guass'
        
        WinnerNei   = exp(-WinnerDist.^2/RSOMNeiFunP^2);
        
    otherwise
        
        error('Unknown RSOMNeiFun !!!')
        
end

DWINRSOM    = RSOMUnSLStepIN*repmat(WinnerNei , INNum , 1).*(repmat(I , 1 , RSOMNum) - WINRSOM);
DWRSOM      = RSOMUnSLStepC*repmat(WinnerNei , size(RSOMState , 1) , 1).*(repmat(RSOMState , 1 , RSOMNum) - WRSOM);

WINRSOM     = WINRSOM + DWINRSOM;
WRSOM       = WRSOM + DWRSOM;

%-----Storing
Net.W.WINRSOM   = WINRSOM;
Net.W.WRSOM     = WRSOM;

return