adanet_test.m~

cfg.maxNodes = 50*ones(1,5);
cfg.maxWeightNorms = 1*ones(size(cfg.maxNodes)); %capital lambda
% cfg.maxWeightNorms = 1:-0.05:0.55;
cfg.pnorm = 3;
cfg.maxWeightMag = 10*ones(size(cfg.maxNodes));
% cfg.maxWeightMag = linspace(1e2,1e5,length(cfg.maxNodes));
cfg.maxBiasMag = 10;
% cfg.complexityRegWeight = .001*ones(size(cfg.maxNodes));
cfg.complexityRegWeight = logspace(-4,-3,length(cfg.maxNodes));
cfg.normRegWeight = 0.03*ones(size(cfg.maxNodes)); %beta
cfg.augment = true;
cfg.augmentLayers = true;
cfg.activationFunction = 'tanh';
cfg.numEpochs = 300;

% cfg.featureMap = @quadMap;
cfg.featureMap = [];


cfg.lossFunction = 'binary';
cfg.surrogateLoss = 'logistic';

%% create synthetic data set
% [x,y] = cubicsplit;
addpath datasets/
data = twospirals(44100/2, 720, .1,1);
y = 2*data(:,end)-1;
x = data(:,1:2);
x = bsxfun(@times, x, 1./std(x));
x = bsxfun(@plus, x, -mean(x));
 %figure,scatter(x(:,1),x(:,2),2,c)
 
% two spirals
xs = twospirals(44100,540,0,1);
ys = 2*(xs(:,end))-1;
xs = xs(:,1:end-1);
%gscatter(xs(:,1),xs(:,2),ys)

% xs = [xs, xs.^2, xs(:,1).*xs(:,2)];
% xs = bsxfun(@times,xs,1./std(xs));
% xs = bsxfun(@plus, xs,-mean(xs));

% % circle
% xs = clusterincluster(44100,0.25,1);
% ys = 2*(xs(:,end))-1;
% xs = xs(:,1:end-1);
[adaParams,history] = adanet(xs, ys, cfg);

%% Visualize when nodes are added
vl = 1; %which hidden layer to visualize
plot(adaParams.lossStore); hold on;
plot(history.newN(:,1),history.newN(:,vl+1).*adaParams.lossStore(history.newN(:,1)).','g*')
ylim([0.01,0.5])
hold off;

adaParams.feedforward = false
adanet_plot(adaParams,[-2,-2;2,2],200,x,y)

%% Visualize weight modifications over time
% paramsT.activationFunction = history.activationFunction;
% paramsT.augment = adaParams.augment;
% paramsT.augmentLayers = adaParams.augmentLayers;
% paramsT.feedforward = false;
% paramsT.W_bias = adaParams.W_bias;
% paramsT.lossFunction = adaParams.lossFunction;
% paramsT.featureMap = adaParams.featureMap;
% Tselect = [1:cfg.numEpochs];
% minX = 1.1*min(min(xs));
% maxX = abs(minX);
% for ti = Tselect
%     paramsT.W = history.Wt{ti+1};
%     paramsT.u = history.ut{ti+1};
%     paramsT.numLayers = history.numL(ti);
%     paramsT.numNodes = history.numN(ti,:);
%     adanet_plot(paramsT,[minX,minX;maxX,maxX],200,xs,ys)
%     titleStr = sprintf('End of Epoch: %d, numNodes: = (',Tselect);
%     for ni = 1:length(paramsT.numNodes)-1,
%         titleStr = sprintf([titleStr,'%d,'],paramsT.numNodes(ni));
%     end
%     titleStr = sprintf([titleStr,'%d)'],paramsT.numNodes(end));
%     title(titleStr)
% %     pause;
% end