rri_multiblock_perm.m

绝对经典,老外制作的功能强大的matlab实现PLS_TOOBOX

%RRI_MULTIBLOCK_PERM Apply Behavioral or Task PLS test and permutation test
%	on RRI scan.
%
%   See also PLS_PERM_TEST, PLS_DEVIATION_PERM_TEST, BEHAVPLS_PERM, TASKPLS_PERM
%

%   I (ismean) - 1 if using grand mean deviation;
%   I (ishelmert) - 1 if using helmert matrix;
%   I (iscontrast) - 1 if using contrast data;
%   I (isbehav) - 1 if using behavior data;
%   I (newdata_lst) - A group list of datamat files;
%   I (num_cond_lst) - A group list of condition numbers;
%   I (num_subj_lst) - A group list of subject numbers;
%   I (behavdata_lst) - A group list of behav data with selected columns;
%   I (helmertdata_lst) - A group list of helmert matrix with selected columns;
%   I (contrastdata_lst) - A group list of contrast data with selected columns;
%			contrastdata_lst has been orthchecked in
%			erp_get_common function
%   I (num_perm) - Number of Permutation;
%   I (posthoc) - posthoc data;
%
%   O (brainlv) - Left singular value vector. It is LV for the brain dimension.
%   O (s) - Singular values vector.
%   O (behavlv) - Right singular vector. It is LV for the behavior OR contrast
%		dimension for each scan.
%   O (brainscores) - The brain score.
%   O (behavscores) - If use GRAND_MEAN method, behavscores = behavlv, then
%		expand each condition for all the subjects. If not, it is
%		calculated for each set of condition, then stack together.
%   O (lvcorrs) - Correlates brain scores with behavior data for multiple scans,
%		return [] if for Task PLS or if use GRAND_MEAN method.
%   O (origpost) - posthoc result.
%   O (perm_result) - A Structure array containing the permutation result data.
%
%   Created on 03-OCT-2002 by Jimmy Shen for PLS test
%   Modified on 24-OCT-2002 by Jimmy Shen to add permutation test
%   Modified Jan 10,03 to add contrast & helmert
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

function [brainlv,s,designlv,behavlv,brainscores,designscores,behavscores, ...
		lvcorrs, origpost, perm_result, datamatcorrs_lst, b_scores, ...
		behav_row_idx, behavdata_lst] = ...
                rri_multiblock_perm(ismean, ishelmert, iscontrast, isbehav, ...
                newdata_lst,num_cond_lst,num_subj_lst, ...
                behavdata_lst, helmertdata_lst, contrastdata_lst, ...
                num_perm, posthoc, bscan);

    % Init
    %
    brainlv = [];
    s = [];
    designlv = [];
    behavlv = [];
    brainscores = [];
    designscores = [];
    behavscores = [];
    lvcorrs = [];
    origpost = [];
    b_scores = [];
    datamatcorrs_lst = {};

    stacked_behavdata = [];
    stacked_helmertdata = [];
    stacked_contrastdata = [];
    stacked_TBdatamatcorrs = [];
    stacked_datamatcorrs = [];
    stacked_datamat = [];
    stacked_mask = [];

    perm_result = [];

    num_groups = length(newdata_lst);

    progress_hdl = rri_progress_ui('initialize');

    msg = 'Working on PLS ...';
    rri_progress_ui(progress_hdl, '', msg);

        k = num_cond_lst(1);
        kk = length(bscan);

    row_idx = [];

    % loop accross the groups, and
    % calculate datamatcorrs for each group
    %
    for i = 1:num_groups

        n = num_subj_lst(i);

        datamat = newdata_lst{i};

        rri_progress_ui(progress_hdl,'',2/10+5/10*(i-1)/(num_groups)+1/(10*num_groups));

       % compute task mean
       %
       Tdatamatcorrs = rri_task_mean(datamat,n)-ones(k,1)*mean(datamat);

        % compute correlation or covariance
        %
            Bdatamatcorrs = rri_corr_maps_notall(behavdata_lst{i}, datamat, n, bscan);
            datamatcorrs_lst = [datamatcorrs_lst, {Bdatamatcorrs}];

        rri_progress_ui(progress_hdl,'',2/10+5/10*(i-1)/(num_groups)+3/(10*num_groups));

       %  stack task and behavior - keep un-normalize data that will be
       %  used to recover the normalized one
       %
       TBdatamatcorrs = [Tdatamatcorrs; Bdatamatcorrs];

       %  stack task and behavior - normalize to unit length to reduce
       %  scaling difference
       %
       datamatcorrs = [normalize(Tdatamatcorrs,2); normalize(Bdatamatcorrs,2)];

       stacked_TBdatamatcorrs = [stacked_TBdatamatcorrs; TBdatamatcorrs];

            stacked_behavdata = [stacked_behavdata; behavdata_lst{i}];

        stacked_datamat = [stacked_datamat; datamat];
        stacked_datamatcorrs = [stacked_datamatcorrs; datamatcorrs];

         tmp_row_idx = reshape(1:size(stacked_behavdata,2)*k, [size(stacked_behavdata,2) k]);
         tmp_row_idx = tmp_row_idx(:,bscan);
         row_idx = [row_idx tmp_row_idx(:)+size(stacked_behavdata,2)*k*(i-1)];

        rri_progress_ui(progress_hdl,'',2/10+5/10*(i-1)/(num_groups)+5/(10*num_groups));

    end		% for

    if ~isempty(posthoc)
       row_idx = row_idx(:);
       posthoc = posthoc(row_idx,:);
    end

    % actually, all the groups must have the same condition number
    %
    num_cond = num_cond_lst(1);

    % Singular Value Decomposition
    %
    [r c] = size(stacked_datamatcorrs);
    if r <= c
        % transpose datamatcorrs to ensure SVD operation will be
        % on smallest of RxC dimension
        %
        [brainlv,s,v] = svd(stacked_datamatcorrs',0);
    else
        [v,s,brainlv] = svd(stacked_datamatcorrs,0);
    end

    s = diag(s);

    original_v = v * diag(s);

    rri_progress_ui(progress_hdl,'',9/10);

   %  Since the 2 matrices that went into the SVD were unit normal, we should
   %  go backwards from the total Singular value Sum of Squares (SSQ)

      %  Calculate total SSQ
      %
      total_s = sum(stacked_TBdatamatcorrs(:).^2);

      %  Calculate distribution of normalized SSQ across LVs
      %
      per = s.^2 / sum(s.^2);

      %  Re-calculate singular value based on the distribution of SSQ
      %  across normalized LVs
      %
      org_s = sqrt(per * total_s);

      %  Re-scale v (block LV) with singular value
      %
      org_v = v * diag(org_s);


      %  Separate v into 2 parts: designlv and behavlv
      %
      designlv = [];
      behavlv = [];

      for g = 1:num_groups
         t = size(stacked_behavdata, 2);

         designlv = [designlv; v((g-1)*k+(g-1)*kk*t+1 : (g-1)*k+(g-1)*kk*t+k,:)];
         behavlv = [behavlv; v((g-1)*k+(g-1)*kk*t+k+1 : (g-1)*k+(g-1)*kk*t+k+kk*t,:)];
      end

      num_col = size(designlv, 2);

      % expand the num_subj for each row (cond)
      % did the samething as testvec
      %
      designscores = [];
      row_idx = [];
      last = 0;

      for g = 1:num_groups
         n = num_subj_lst(g);

         tmp = reshape(designlv((g-1)*k+1:(g-1)*k+k,:),[1, num_col*k]);
         tmp = repmat(tmp, [n, 1]);		% expand to num_subj
         tmp = reshape(tmp, [n*k, num_col]);

         designscores = [designscores; tmp];		% stack by groups

         %  take this advantage (having g & n) to get row_idx
         %
         tmp = 1:n*k;
         tmp = reshape(tmp, [n k]);
         tmp = tmp(:, bscan);
         behavdata_lst{g} = behavdata_lst{g}(tmp(:),:);

         row_idx = [row_idx ; tmp(:) + last];
         last = last + n*k;
      end

    behav_row_idx = row_idx;

    % calculate behav scores
    %
        b_scores = stacked_datamat * brainlv;
        behav = stacked_behavdata;

        [brainscores, behavscores, lvcorrs] = ...
		rri_get_behavscores(stacked_datamat(row_idx,:), ...
		stacked_behavdata(row_idx,:), ...
		brainlv, behavlv, kk, num_subj_lst);

%      lvcorrs = original_v;

    if ~isempty(posthoc)
        origpost = rri_xcor(posthoc,behavlv);
        porigpost = zeros(size(origpost));
    else
        origpost = [];
    end


    rri_progress_ui(progress_hdl,'',1);

    %  Begin permutation loop
    %
    sp = zeros(size(s));
    dp = zeros(size(v));

    Treorder = rri_perm_order(num_subj_lst, k, num_perm);
    rand('state',sum(100*clock));

        for p = 1:num_perm
            Breorder(:,p) = [randperm(size(stacked_datamat,1))'];
        end

    for p = 1:num_perm

        msg = ['Working on Permutation:  ',num2str(p),' out of ',num2str(num_perm)];
        rri_progress_ui(progress_hdl, '', msg);
        rri_progress_ui(progress_hdl,'',p/num_perm);

            data_p = stacked_datamat(Treorder(:,p),:);
            behav_p = stacked_behavdata(Breorder(:,p),:);

        stacked_TBdata = [];
        stacked_data = [];

        for g=1:num_groups

            k = num_cond_lst(g);
            n = num_subj_lst(g);
            span = sum(num_subj_lst(1:g-1)) * num_cond;

                if num_groups == 1
                   Tdata = rri_task_mean(data_p,n)-ones(k,1)*mean(data_p);
                else
                   Tdata = rri_task_mean(data_p(1+span:n*k+span,:),n)-ones(k,1)*mean(data_p(1+span:n*k+span,:));
                end

            if 1

		% Check for upcoming NaN and re-sample if necessary.
		% this only happened on behavior analysis, because the
		% 'xcor' inside of 'rri_corr_maps' contains a 'stdev', which
		% is a divident. If it is 0, it will cause divided by 0
		% problem.
		% since this happend very rarely, so the speed will not
		% be affected that much.
		%
                min1 = min(std(behav_p(1+span:n*k+span,:)));
                while (min1 == 0)
                    Breorder(:,p) = [randperm(size(stacked_datamat,1))'];
                    behav_p = stacked_behavdata(Breorder(:,p),:);
                    min1 = min(std(behav_p(1+span:n*k+span,:)));
                    count = count + 1;
                    if count > 100
                       msg = 'Please check your behavior data, and make ';
                       msg = [msg 'sure none of the columns are all the '];
                       msg = [msg 'same for each group'];
                       uiwait(msgbox(msg, 'Program can not proceed', 'modal'));
                       brainlv = [];
                       return;
                    end
                end

		% Notice here that stacked_datamat is used, instead of
		% boot_p. This is only for behavpls_perm.
		%
                if num_groups == 1
                   Bdata = rri_corr_maps_notall(behav_p, stacked_datamat, n, bscan);
                else
                   Bdata = rri_corr_maps_notall(behav_p(1+span:n*k+span,:), ...     
				stacked_datamat(1+span:n*k+span,:), n, bscan);
                end
            end

            TBdata = [Tdata; Bdata];
            data = [normalize(Tdata,2); normalize(Bdata,2)];

            stacked_TBdata = [stacked_TBdata; TBdata];
            stacked_data = [stacked_data; data];

        end		% for num_groups

        % Singular Value Decomposition
        %
        [r c] = size(stacked_data);
        if r <= c
            % transpose datamatcorrs to ensure SVD operation will be
            % on smallest of RxC dimension
            %
            [pbrainlv, sperm, pv] = svd(stacked_data',0);
        else
            [pv, sperm, pbrainlv] = svd(stacked_data,0);
        end

        rotatemat = rri_bootprocrust(v,pv);
        pv = pv * sperm * rotatemat;
        sperm = sqrt(sum(pv.^2));

        ptotal_s = sum(stacked_TBdata(:).^2);
        per = diag(sperm).^2 / sum(diag(sperm).^2);
        sperm = sqrt(per * ptotal_s);
        pv = normalize(pv) * diag(sperm);

        sp = sp + (sperm>=org_s);
        dp = dp + (abs(pv) >= abs(org_v));

        Bpv = [];

        for g = 1:num_groups
            t = size(stacked_behavdata, 2);
            Bpv = [Bpv; pv((g-1)*k+(g-1)*kk*t+k+1 : (g-1)*k+(g-1)*kk*t+k+kk*t,:)];
        end

        if ~isempty(posthoc)
            tmp = rri_xcor(posthoc, Bpv);
            porigpost = porigpost + (abs(tmp) >= abs(origpost));
        end

    end		% for num_perm

    if num_perm ~= 0

        perm_result.sprob = sp ./ (num_perm + 1);
        % perm_result.dprob = dp ./ num_perm;
        perm_result.vprob = dp ./ (num_perm + 1);
        perm_result.num_perm = num_perm;
        perm_result.Tpermsamp = Treorder;
        perm_result.Bpermsamp = Breorder;
        perm_result.sp = sp;
        % perm_result.dp = dp;

        if ~isempty(posthoc)
            perm_result.posthoc_prob = porigpost / num_perm;
        end

    end

    return;					% rri_multiblock_perm


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
%   load_contrast_file: turn the contrast file into a contrast matrix,
%		which is used for behavdata or design data.
%
%   I (contrast_file) - filespec of the contrast file
%   I (num_subj) - number of subjects in each condition
%   O (contrasts) - contrast matrix
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

function contrasts = load_contrast_file(contrast_file, num_subj)

    load(contrast_file);

    num_contrasts = length(pls_contrasts);
    num_conditions = length(pls_contrasts(1).value);

    contrasts = zeros(num_conditions,num_contrasts);
    for i=1:num_contrasts
        contrasts(:,i) = pls_contrasts(i).value';
    end

    % expand each row of condition for all the subjects
    %
    tmp = contrasts(:)';
    tmp = repmat(tmp,num_subj,1);
    contrasts = reshape(tmp, num_subj*num_conditions, num_contrasts);

    return;						% load_contrast_file