nnet_denoise.m

function [lsizeOUT,ltypeOUT] = nnet_denoise(alldata,trainIND,testIND,runName,noiseIters,sig2noise,varargin)

seed = 1234;

randn('state', seed );
rand('twister', seed+1 );


%you will NEVER need more than a few hundred epochs unless you are doing
%something very wrong.  Here 'epoch' means parameter update, not 'pass over
%the training set'.
maxepoch = 150;


%CURVES
%%%%%%%%%%%%%%%%%
%this dataset (by Ruslan Salakhutdinov) is available here: http://www.cs.toronto.edu/~jmartens/digs3pts_1.mat

% tmp = load('digs3pts_1.mat');
% indata = tmp.bdata';
% %outdata = tmp.bdata;
% intest = tmp.bdatatest';
% %outtest = tmp.bdatatest;
% clear tmp
% 
% perm = randperm(size(indata,2));
% indata = indata( :, perm );

%it's an auto-encoder so output is input
% outdata = indata;
% outtest = intest;
%runName = 'HFb10_04122013';

% permInd=randperm(size(alldata,2));
% 
% indata= alldata(:,permInd(1:trainIND));
% outdata=alldata(:,permInd(1:trainIND));
% intest=alldata(:,permInd(testIND:end));
% outtest=alldata(:,permInd(testIND:end));

indata = alldata(:,trainIND);
intest = alldata(:,testIND);

indata = indata(:,randperm(size(indata,2)));
outdata = indata;
intest = intest(:,randperm(size(intest,2)));
outtest = intest;

sig = std(alldata,[],2);
noise2add = sig/sig2noise;
indata = repmat(indata,1,noiseIters) + randn(size(indata,1),size(indata,2)*noiseIters).*repmat(noise2add,1,size(indata,2)*noiseIters);
intest = repmat(intest,1,noiseIters) + randn(size(intest,1),size(intest,2)*noiseIters).*repmat(noise2add,1,size(intest,2)*noiseIters);
outdata = repmat(outdata,1,noiseIters);
outtest = repmat(outtest,1,noiseIters);

runDesc = ['seed = ' num2str(seed) 'd is reduced to PC scores with 95 of the variance' ];

%next try using autodamp = 0 for rho computation.  both for version 6 and
%versions with rho and cg-backtrack computed on the training set

if isempty(varargin)
    layersizes = [100 50 20 2 20 50 100];
else
    layersizes = varargin{1};
end
layertypes = {'tanh', 'tanh', 'tanh', 'linear', 'tanh','tanh', 'tanh', 'linear'};

% layersizes = [150 75 40 20 2 20 40 75 150];
% layertypes = {'logistic', 'logistic', 'logistic', 'logistic', 'linear', 'logistic', 'logistic', 'logistic', 'logistic', 'linear'};


resumeFile = [];

lsizeOUT = [size(alldata,1) layersizes size(alldata,1)];
ltypeOUT = layertypes;
ltypeOUT{find(layersizes == min(layersizes))} = 'linearSTORE';

paramsp = [];
Win = [];
bin = [];
%[Win, bin] = loadPretrainedNet_curves;

numchunks = noiseIters*4;
numchunks_test = noiseIters;

mattype = 'gn'; %Gauss-Newton.  The other choices probably won't work for whatever you're doing
%mattype = 'hess';
%mattype = 'empfish';

rms = 0;

hybridmode = 1;

%decay = 1.0;
decay = 0.95;

jacket = 0;
%this enables Jacket mode for the GPU
%jacket = 1;

errtype = 'L2'; %report the L2-norm error (in addition to the quantity actually being optimized, i.e. the log-likelihood)

%standard L_2 weight-decay:
weightcost = 2e-5;
%weightcost = 0

nnet_train_2( runName, runDesc, paramsp, Win, bin, resumeFile, maxepoch, indata, outdata, numchunks, intest, outtest, numchunks_test, layersizes, layertypes, mattype, rms, errtype, hybridmode, weightcost, decay, jacket);