lard.m

竞争学习的matlab工具箱

% MATLAB implementation of local AR models built through the clustered data
% vectors.
%
% Based on the prediction method described in the following paper:
% 
% Lehtokangas et al. (1996). "A network of autoregressive processing units
% for time series modelling", Applied Mathematics and Computation,
% 75:151-165.
%
% Authors: Guilherme A. Barreto
% Date: September 21st 2006

clear; clc; close all;

%--------------- Organize the training data ------------
load train_series_data.dat;    % Load the time series to be clustered

%% Building the input vectors from an univariate time series
p=5;       % Dimension of the input vector (length of the time window)
lap=p-1;      % Amount of overlapping between consecutive input vectors
Dw=buffer(train_series_data,p,lap);   % Build the data vectors
if lap>0,
     Dw=Dw(:,p:end)';  % Eliminate the first 'p-1' vectors with zeros)
else Dw=Dw';
end
Dw=fliplr(Dw);
Dw=Dw+0.01*randn(size(Dw));  % Add some gaussian noise to the data
%-------------------------------------------------------

Mx = 10;           % Number of neurons in the X-dimension
My = 1;            % Number of neurons in the Y-dimension
msize = [Mx My];   % Size of 2-D SOM map

sMap  = som_randinit(Dw, 'msize', msize,'lattice','rect','shape','sheet');   % Random weight initialization

% Train SOM
sMap = som_seqtrain(sMap,Dw,'radius',[8 0],'sample_order','random','neigh','gaussian','trainlen',50);

% Divide the training vectors into clusters based on learned prototypes
[V I] = som_divide(sMap,Dw);

% Build local AR models from data
WW=[];
for i=1:Mx*My,
     y=V{i}(:,1);      % Prediction vector
     X=[ones(size(y)) V{i}(:,2:end)];  % Regression matrix
     w=X\y;            % Estimated coefficient vector w=inv(X'*X)*X'*y
     WW=[WW; w'];      % Store the coefficient vectors
end

%--------------- Organize the testing data ------------
load test_series_data.dat;    % Load testing time series
Dw=buffer(test_series_data,p,lap); % Build input data vectors

if lap>0,
     Dw=Dw(:,p:end)';  % Eliminate the first 'p-1' vectors containing zeros
else Dw=Dw';
end
Dw=fliplr(Dw);
Dw=Dw+0.01*randn(size(Dw));  % Add some gaussian noise to the data

Ytrue=Dw(2:end,1);       % Desired output values
%------------------------------------------------------

[LEN_DATA DIM_INPUT]=size(Dw);  % Data matrix size (1 input vector per row)

for t=1:LEN_DATA-1,
    win = som_bmus(sMap,Dw(t,:),1);    % Find the winning neuron (bmu)
    Yhat(t) = dot(WW(win,:),Dw(t,:));  % Predicted value of neuron 'win'
    error(t)=Ytrue(t)-Yhat(t);          % Corresponding prediction error
end

% Normalized mean squared error
NMSE=var(error)/var(Ytrue)

% Plot target and predicted time series
figure; plot(Ytrue,'r-'); hold on; plot(Yhat,'b-'); hold off