Mercurial > hg > camir-aes2014

annotate toolboxes/FullBNT-1.0.7/bnt/examples/static/HME/fhme.m @ 0:e9a9cd732c1e tip

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
first hg version after svn
author wolffd
date Tue, 10 Feb 2015 15:05:51 +0000
parents
children
rev   line source
wolffd@0 1 function risultati = fhme(net, nodes_info, data, n)
wolffd@0 2 %HMEFWD Forward propagation through an HME model
wolffd@0 3 %
wolffd@0 4 % Each row of the (n x class_num) matrix 'risultati' containes the estimated class posterior prob.
wolffd@0 5 %
wolffd@0 6 % ----------------------------------------------------------------------------------------------------
wolffd@0 7 % -> pierpaolo_b@hotmail.com or -> pampo@interfree.it
wolffd@0 8 % ----------------------------------------------------------------------------------------------------
wolffd@0 9 %
wolffd@0 10 ns=net.node_sizes;
wolffd@0 11 if nargin==3
wolffd@0 12 ndata=n;
wolffd@0 13 else
wolffd@0 14 ndata=size(data, 1);
wolffd@0 15 end
wolffd@0 16 altezza=size(ns,2);
wolffd@0 17 coeff=cell(altezza-1,1);
wolffd@0 18 for m=1:ndata
wolffd@0 19 %- i=2 --------------------------------------------------------------------------------------
wolffd@0 20 s=struct(net.CPD{2});
wolffd@0 21 if nodes_info(1,2)==0,
wolffd@0 22 mu=[]; W=[]; predict=[];
wolffd@0 23 mu=s.mean(:,:);
wolffd@0 24 W=s.weights(:,:,:);
wolffd@0 25 predict=mu(:,:)+W(:,:,:)*data(m,:)';
wolffd@0 26 coeff{1,1}=predict';
wolffd@0 27 elseif nodes_info(1,2)==1,
wolffd@0 28 coeff{1,1}=fglm(s.glim{1}, data(m,:));
wolffd@0 29 else,
wolffd@0 30 coeff{1,1}=fmlp(s.mlp{1}, data(m,:));
wolffd@0 31 end
wolffd@0 32 %----------------------------------------------------------------------------------------------
wolffd@0 33 if altezza>3,
wolffd@0 34 for i=3:altezza-1,
wolffd@0 35 s=[]; f=[]; dpsz=[];
wolffd@0 36 f=family(net.dag,i); f=f(2:end-1); dpsz=prod(ns(f));
wolffd@0 37 s=struct(net.CPD{i});
wolffd@0 38 for j=1:dpsz,
wolffd@0 39 if nodes_info(1,i)==1,
wolffd@0 40 coeff{i-1,1}(j,:)=coeff{i-2,1}(1,j)*fglm(s.glim{j}, data(m,:));
wolffd@0 41 else
wolffd@0 42 coeff{i-1,1}(j,:)=coeff{i-2,1}(1,j)*fmlp(s.mlp{j}, data(m,:));
wolffd@0 43 end
wolffd@0 44 end
wolffd@0 45 app=cat(2, coeff{i-1,1}(:)); coeff{i-1,1}=app'; clear app;
wolffd@0 46 end
wolffd@0 47 end
wolffd@0 48 %- i=altezza ----------------------------------------------------------------------------------
wolffd@0 49 if altezza>2,
wolffd@0 50 i=altezza;
wolffd@0 51 s=[]; f=[]; dpsz=[];
wolffd@0 52 f=family(net.dag,i); f=f(2:end-1); dpsz=prod(ns(f));
wolffd@0 53 s=struct(net.CPD{i});
wolffd@0 54 if nodes_info(1,i)==0,
wolffd@0 55 mu=[]; W=[];
wolffd@0 56 mu=s.mean(:,:);
wolffd@0 57 W=s.weights(:,:,:);
wolffd@0 58 end
wolffd@0 59 for j=1:dpsz,
wolffd@0 60 if nodes_info(1,i)==0,
wolffd@0 61 predict=[];
wolffd@0 62 predict=mu(:,j)+W(:,:,j)*data(m,:)';
wolffd@0 63 coeff{i-1,1}(j,:)=coeff{i-2,1}(1,j)*predict';
wolffd@0 64 elseif nodes_info(1,i)==1,
wolffd@0 65 coeff{i-1,1}(j,:)=coeff{i-2,1}(1,j)*fglm(s.glim{j}, data(m,:));
wolffd@0 66 else
wolffd@0 67 coeff{i-1,1}(j,:)=coeff{i-2,1}(1,j)*fmlp(s.mlp{j}, data(m,:));
wolffd@0 68 end
wolffd@0 69 end
wolffd@0 70 end
wolffd@0 71 %----------------------------------------------------------------------------------------------
wolffd@0 72 risultati(m,:)=sum(coeff{altezza-1,1},1);
wolffd@0 73 clear coeff; coeff=cell(altezza-1,1);
wolffd@0 74 end
wolffd@0 75 return
wolffd@0 76
wolffd@0 77 %-------------------------------------------------------------------
wolffd@0 78
wolffd@0 79 function [y, a] = fglm(net, x)
wolffd@0 80 %GLMFWD Forward propagation through 1-layer net->GLM statistical model
wolffd@0 81
wolffd@0 82 ndata = size(x, 1);
wolffd@0 83
wolffd@0 84 a = x*net.w1 + ones(ndata, 1)*net.b1;
wolffd@0 85
wolffd@0 86 nout = size(a,2);
wolffd@0 87 % Ensure that sum(exp(a), 2) does not overflow
wolffd@0 88 maxcut = log(realmax) - log(nout);
wolffd@0 89 % Ensure that exp(a) > 0
wolffd@0 90 mincut = log(realmin);
wolffd@0 91 a = min(a, maxcut);
wolffd@0 92 a = max(a, mincut);
wolffd@0 93 temp = exp(a);
wolffd@0 94 y = temp./(sum(temp, 2)*ones(1,nout));
wolffd@0 95
wolffd@0 96 %-------------------------------------------------------------------
wolffd@0 97
wolffd@0 98 function [y, z, a] = fmlp(net, x)
wolffd@0 99 %MLPFWD Forward propagation through 2-layer network.
wolffd@0 100
wolffd@0 101 ndata = size(x, 1);
wolffd@0 102
wolffd@0 103 z = tanh(x*net.w1 + ones(ndata, 1)*net.b1);
wolffd@0 104 a = z*net.w2 + ones(ndata, 1)*net.b2;
wolffd@0 105 temp = exp(a);
wolffd@0 106 nout = size(a,2);
wolffd@0 107 y = temp./(sum(temp,2)*ones(1,nout));
wolffd@0 108
wolffd@0 109 %-------------------------------------------------------------------