camir-aes2014: toolboxes/FullBNT-1.0.7/bnt/examples/static/Misc/mixexp

first hg version after svn

rev	line source
wolffd@0	1 function plot_mixexp(theta, eta, data)
wolffd@0	2 % PLOT_MIXEXP Plot the results for a piecewise linear regression model
wolffd@0	3 % plot_mixexp(theta, eta, data)
wolffd@0	4 %
wolffd@0	5 % data(l,:) = [x y] for example l
wolffd@0	6 % theta(i,:) = regression vector for expert i
wolffd@0	7 % eta(i,:) = softmax (gating) params for expert i
wolffd@0	8
wolffd@0	9 numexp = size(theta, 1);
wolffd@0	10
wolffd@0	11 mn = min(data);
wolffd@0	12 mx = max(data);
wolffd@0	13 xa = mn(1):0.01:mx(1);
wolffd@0	14 x = [ones(length(xa),1) xa'];
wolffd@0	15 % pr(i,l) = posterior probability of expert i on example l
wolffd@0	16 pr = exp(eta * x');
wolffd@0	17 pr = pr ./ (ones(numexp,1) * sum(pr));
wolffd@0	18 % y(i,l) = prediction of expert i for example l
wolffd@0	19 y = theta * x';
wolffd@0	20 % yg(l) = weighted prediction for example l
wolffd@0	21 yg = sum(y .* pr)';
wolffd@0	22
wolffd@0	23 subplot(3,2,1);
wolffd@0	24 plot(xa, y(1,:));
wolffd@0	25 title('expert 1');
wolffd@0	26
wolffd@0	27 subplot(3,2,2);
wolffd@0	28 plot(xa, y(2,:));
wolffd@0	29 title('expert 2');
wolffd@0	30
wolffd@0	31 subplot(3,2,3);
wolffd@0	32 plot(xa, pr(1,:));
wolffd@0	33 title('gating 1');
wolffd@0	34
wolffd@0	35 subplot(3,2,4);
wolffd@0	36 plot(xa, pr(2,:));
wolffd@0	37 title('gating 2');
wolffd@0	38
wolffd@0	39 subplot(3,2,5);
wolffd@0	40 plot(xa, yg);
wolffd@0	41 axis([-1 1 -1 2])
wolffd@0	42 title('prediction');
wolffd@0	43
wolffd@0	44 subplot(3,2,6);
wolffd@0	45 title('data');
wolffd@0	46 hold on
wolffd@0	47 plot(data(:,1), data(:,2), '+');
wolffd@0	48 hold off
wolffd@0	49

Mercurial > hg > camir-aes2014