Revert "fucked up everthing, reverting back"

This reverts commit 3aa9a44.

Author: jovo

classifiers/bayes_plugin/plugin_classify.m

@@ -13,17 +13,13 @@
 %   yhat:       list of estimated class identity for each graph
 
 siz     = size(As);
-if length(siz)==3
-    s       = siz(end);                         % # of samples (robust to As being an arbitrary array)
-else
-    s = 1;
-end
+s       = siz(end);                         % # of samples (robust to As being an arbitrary array)
 poiss   = any(As(:))>1;          % whether data is Poisson or Bernoulli
 n_subs  = length(subspace);
 yhat    = nan(s,n_subs);
 
 for i=1:s
-    datum = As(:,:,i);               % this line only makes sense when data are graphs
+    datum = As(:,:,i);               % this line only makes sense when data are graphs    
     for j=1:n_subs
         ind = subspace(j).indices;         % for code legibility
 

meta/classify_par.m

@@ -14,27 +14,25 @@
 %   xval:       xval structure after changing it
 
 %% ensure that xval fields are set
-% if ~isfield(xval,'s0_trn'),
-%     disp(['s0 = ' num2str(constants.s0)])
-%     xval.s0_trn = input('\s0_trn [positive integer]: ');
-% end
-% if ~isfield(xval,'s1_trn'),
-%     disp(['s1 = ' num2str(constants.s1)])
-%     xval.s1_trn = input('\s1_trn [positive integer]: ');
-% end
-
-% if num_iters is not specified, do leave-one-out cross-validation
+if ~isfield(xval,'s0_trn'),
+    disp(['s0 = ' num2str(constants.s0)])
+    xval.s0_trn = input('\s0_trn [positive integer]: ');
+end
+if ~isfield(xval,'s1_trn'),
+    disp(['s1 = ' num2str(constants.s1)])
+    xval.s1_trn = input('\s1_trn [positive integer]: ');
+end
 if ~isfield(xval,'num_iters'),
-    xval.num_iters = constants.s; %choose(constants.s0,xval.s0_trn)*choose(constants.s1,xval.s1_trn);
+    xval.num_iters = choose(constants.s0,xval.s0_trn)*choose(constants.s1,xval.s1_trn);
 end
 
 
 
 %% run classifiers
 Lstats_tmp(1:xval.num_iters)=struct('avg',nan,'sem',nan);
-for i=1:xval.num_iters
+parfor i=1:xval.num_iters
     if mod(i,100)==0, disp(['repeat # ' num2str(i)]), end
-    [Atrn Gtrn Atst Gtst inds] = crossvalprep3(As,constants,xval,i); % seperate data into training and testing sets
+    [Atrn Gtrn Atst Gtst inds] = crossvalprep2(As,constants,xval,i); % seperate data into training and testing sets
 
     % naive_bayes classifiers
     if ~isempty(subspace)

plots/plot_means_sig.m

@@ -0,0 +1,219 @@
+clear, clc
+
+
+alg.save=1;
+alg.figdir = '/Users/jovo/Research/figs/MRI/BLSA/';
+alg.fname = 'BLSA50';
+
+
+%% plot model and significance
+clc
+figure(1), clf,
+fs=10;
+
+load('/Users/jovo/Research/data/MRI/BLSA/results/edge_significance.mat')
+
+
+alg.save=1;
+alg.figdir = '/Users/jovo/Research/figs/MRI/BLSA/';
+alg.fname = 'BLSA50';
+
+
+params = get_params_mle(As,constants);
+
+emax=max(max([params.E0(:) params.E1(:) abs(params.E0(:)-params.E1(:))]));
+
+subplot(131)
+image(64*params.E0/emax)
+colormap('gray')
+title('female','fontsize',fs)
+xlabel('vertices','fontsize',fs)
+ylabel('vertices','fontsize',fs)
+set(gca,'fontsize',fs,'DataAspectRatio',[1 1 1])
+
+subplot(132)
+image(64*params.E1/emax)
+title('male','fontsize',fs)
+set(gca,'fontsize',fs)
+set(gca,'fontsize',fs,'DataAspectRatio',[1 1 1])
+
+subplot(133)
+imagesc(1-d_pval)
+title('significance of difference','fontsize',fs)
+set(gca,'fontsize',fs)
+set(gca,'fontsize',fs,'DataAspectRatio',[1 1 1])
+
+
+
+if alg.save
+    wh=[6 3];   %width and height
+    set(gcf,'PaperSize',wh,'PaperPosition',[0 0 wh],'Color','w');
+    figname=[alg.figdir alg.fname '_params'];
+    print('-dpdf',figname)
+    print('-deps',figname)
+    saveas(gcf,figname)
+end
+
+%% plot Lhat vs signal subgraph size
+
+clear, clc
+load('/Users/jovo/Research/data/MRI/BLSA/results/BLSA50_Lhat_200_subspaces.mat')
+alg.save=1;
+plot_Lhat_vs_signal_subgraph_size
+
+%% make table of m most significant edges
+
+clear, clc,
+load('/Users/jovo/Research/data/MRI/BLSA/results/edge_significance.mat')
+
+[foo IX] = sort(d_pval(:));
+
+[I J] = ind2sub([70 70],IX);
+
+CorticalLabels = importdata('/Users/jovo/Research/data/MRI/BLSA/misc/IACL-corticallabeldescriptions-1110.txt',' ');
+
+
+clc
+
+clear orderedregions
+m=find(foo<0.999999999999, 1, 'last' );
+for i=1:m
+    
+    if I(i)<36, ii=I(i)+2; else ii=I(i)+3; end
+    if J(i)<36, jj=J(i)+2; else jj=J(i)+3; end
+    
+    if ii<11, ii_start=7; elseif ii<39, ii_start=8; else ii_start=9; end
+    if jj<11, jj_start=7; elseif jj<39, jj_start=8; else jj_start=9; end
+    
+    maxlen=19;
+    label1=CorticalLabels{ii}(ii_start:end-1);
+    if length(label1)>maxlen, label1=label1(1:maxlen); end
+    label2=CorticalLabels{jj}(jj_start:end-1);
+    if length(label2)>maxlen, label2=label2(1:maxlen); end
+    
+    orderedregions{i,1}=foo(i);
+    orderedregions{i,2}=label1;
+    orderedregions{i,3}=label2;
+    
+end
+
+%% remove good edges
+
+clear, clc,
+load('/Users/jovo/Research/data/MRI/BLSA/results/edge_significance.mat')
+
+alg.save=0;
+alg.num_repeats=100;
+alg.s0_trn=15;
+alg.s1_trn=15;
+[pvals ind]=sort(d_pval(:),'ascend');
+
+k=0;
+for m=1:500
+    k=k+1;
+    subspace.manual=ind(m:end);
+    [Lhat Lsem] = classify_holdout(As,constants,alg,subspace);
+    
+    Lhats.manual(k)=mean(Lhat.manual);
+    Lsems.manual(k)=std(Lsem.manual)/sqrt(alg.num_repeats);
+end
+
+
+% alg.num_repeats=1000;
+% subspace.manual=ind(50:end)
+% [LhatX LsemX] = classify_holdout(As,constants,alg,subspace);
+% mean(LhatX.manual)
+
+
+alg.save=0;
+alg.figdir = '/Users/jovo/Research/figs/MRI/BLSA/';
+% alg.postdir = '/Users/jovo/Research/data/MRI/BLSA/results/';
+alg.fname = 'BLSA50';
+if alg.save
+    wh=[6 3];   %width and height
+    set(gcf,'PaperSize',wh,'PaperPosition',[0 0 wh],'Color','w');
+    figname=[alg.figdir alg.fname '_removed_subspace'];
+    print('-dpdf',figname)
+    print('-deps',figname)
+    saveas(gcf,figname)
+end
+
+
+%% Lhat vs. m
+
+clear, clc
+load('/Users/jovo/Research/data/MRI/BLSA/results/edge_significance.mat')
+
+
+alg.save=1;
+alg.num_repeats=100;
+alg.s0_trn=18;
+alg.s1_trn=18;
+alg.postdir='/Users/jovo/Research/data/MRI/BLSA/results/';
+alg.fname='Lhat_v_m';
+
+naive_bayes=find(sum(As,3)>0);
+
+
+[foo fis_ind]=sort(d_pval(:),'ascend');
+
+params = get_params_mle(As,constants);
+
+[foo inc_ind]=sort(params.d_pos(:),'ascend');
+[foo cor_ind]=sort(params.d_opt(:),'ascend');
+
+
+subspace.fis=[];
+subspace.inc=[];
+subspace.cor=[];
+
+subspace_names  = fieldnames(subspace);
+n_subspaces=length(subspace_names);
+
+for m=1:length(naive_bayes)
+    
+    disp(m)
+    subspace.fis=fis_ind(1:m);
+    subspace.inc=inc_ind(1:m);
+    subspace.cor=cor_ind(1:m);
+    [Lhat Lsem] = classify_holdout(As,constants,alg,subspace);
+    
+    for k=1:n_subspaces
+        Lhats.(subspace_names{k})(m)=mean(Lhat.(subspace_names{k}));
+        Lsems.(subspace_names{k})(m)=std(Lsem.(subspace_names{k}))/sqrt(alg.num_repeats);
+    end
+    
+end
+
+
+%%
+deg=sum(d_pval)+sum(d_pval');
+[foo deg_ind] = sort(deg,'ascend');
+
+clear subspace Lhat Lsem
+for m=5:length(deg_ind)
+    
+    disp(m)
+    
+    ind_max_mat=zeros(70);
+    ind_max_mat(deg_ind(1:m),deg_ind(1:m))=1;
+    ind_max_mat=triu(ind_max_mat,+1);
+    subspace.coh=find(ind_max_mat);
+    
+    [Lhat Lsem] = classify_holdout(As,constants,alg,subspace);
+    
+    Lhats.coh(m)=mean(Lhat.coh);
+    Lsems.coh(m)=std(Lsem.coh)/sqrt(alg.num_repeats);
+    
+end
+
+
+
+
+
+
+
+
+
+
+