nt_mmat

PURPOSE

SYNOPSIS

DESCRIPTION

y=nt_mmat(x,m) -  matrix multiplication (with convolution)

  y: result
 
  x: input data (2D or more)
  m: matrix to apply (2D: right multiply, 3D: same with convolution)

  If m is 3D, the last index (k) is lag.

CROSS-REFERENCE INFORMATION

• nt_fold y=fold(x,epochsize) - fold 2D to 3D
• nt_mmat y=nt_mmat(x,m) - matrix multiplication (with convolution)
• nt_unfold y=nt_fold(x) - unfold 3D to 2D

• nt_bias_cluster [c0,c1,A,todss,pwr0,pwr1]=nt_bias_cluster(x,dsr,flags) - cluster covariance
• nt_cca_aad [D,E,R]=nt_cca_match_aad(xA,xB,y,ssize) - calculate metrics for match-mismatch task
• nt_cca_crossvalidate [AA,BB,RR,iBest]=nt_cca_crossvalidate(xx,yy,shifts,ncomp,A0,B0) - CCA with cross-validation
• nt_cca_crossvalidate2 [A,B,RR]=nt_cca_crossvalidate2(xx,yy,shifts) - CCA with cross-validation
• nt_cca_crossvalidate_3 [AA,BB,RR,iBest]=nt_cca_crossvalidate(xx,yy,shifts,ncomp,A0,B0) - CCA with cross-validation
• nt_cca_mm [D,E,R]=nt_cca_mm(x,y,ssize,ldaflag,nccs) - calculate metrics for match-mismatch task
• nt_cca_mm3 [D,E,R]=nt_cca_mm(x,y,ssize,ldaflag,nccs) - calculate metrics for match-mismatch task
• nt_cluster_jd [IDX,todss,SCORE,COVS]=nt_cluster_jd(x,dsr,smooth,flags,init,verbose) - cluster with joint diagonalization
• nt_dprime [d,e]=nt_dprime(x,y,jd_flag) - calculate d' (discriminability) of two distributions
• nt_dss1 [todss,pwr0,pwr1]=nt_dss1(x,w,keep1,keep2) - evoked-biased DSS denoising
• nt_eyeblink [y,z,mask]=nt_eyeblink(x,eyechans,nremove) - project out eyeblinks
• nt_lsp [Y,scores,removed]=nt_LSP(X,npass,thresh,tol,guard) - local subspace pruning
• nt_mmat y=nt_mmat(x,m) - matrix multiplication (with convolution)
• nt_narrowband_scan A=nt_narrowband_scan(x,freqs,sr,Q,plotflag) - scan for narrowband components using DSS
• nt_pca0 [topcs,pwr,y]=nt_pca0(x,shifts,nkeep,threshold,w) - time-shift pca
• nt_qca [squares,quads,D]=nt_qca(x,npcs,nsmooth,nquads) - maximize induced power using quadratic component analysis
• nt_qca0 [tosquares,quads,D]=nt_qca0(x,npcs,nsmooth,nquads) - maximize induced power using quadratic component analysis
• nt_qca02 [tosquare,quad,tosquare2,quad2,D]=nt_qca02(x,npcs,nsmooth) - maximize induced power using quadratic component analysis
• nt_qca2 [square,quad,square2,quad2,D]=nt_qca(x,npcs,nsmooth) - maximize induced power using quadratic component analysis
• nt_qpca [squares,quads]=nt_qpca(x,npcs,nsmooth,nquads) - quadratic PCA
• nt_qpca0 [tosquares,quads,D]=nt_qpca0(x,npcs,nsmooth,nquads) - quadratic PCA
• nt_split_jd [idx,score_vector,todss]=nt_split_dss(x,thresh,depth) - segmentation based on joint diagonalization
• nt_subspace_prune5 [Y]=nt_subspace_prune(X,npass,thresh) - local cleaning matrices
• nt_subspace_prune6 [Y]=nt_subspace_prune6(X,npass,thresh) - local cleaning matrices
• nt_zapline [y,yy]=nt_zapline(x,fline,nremove,p,plotflag) - remove power line artifact

SUBFUNCTIONS

• function x=nt_mmat0(x,m)

SOURCE CODE

0001 function y=nt_mmat(x,m)
0002 %y=nt_mmat(x,m) -  matrix multiplication (with convolution)
0003 %
0004 %  y: result
0005 %
0006 %  x: input data (2D or more)
0007 %  m: matrix to apply (2D: right multiply, 3D: same with convolution)
0008 %
0009 %  If m is 3D, the last index (k) is lag.
0010 
0011 if nargin<2; error('!'); end
0012 
0013 if iscell(x)
0014     for iCell=1:length(x)
0015         y{iCell}=nt_mmat(x{iCell},m);
0016     end
0017     return;
0018 end
0019 
0020 if ndims(x)>3
0021     % concatenate the last dimensions, process, then de-concatenate
0022     sz=size(x);
0023     x=reshape(x,[sz(1),sz(2),prod(sz(3:end))]);
0024     x=nt_mmat(x,m);
0025     x=reshape(x,[size(x,1),sz(2),sz(3:end)]);
0026 
0027 else
0028 
0029     if ndims(m)==2;
0030         
0031         % no convolution
0032         y=nt_mmat0(x,m); 
0033     
0034     else
0035         
0036 % does anyone use this ?????
0037 
0038         [nRows,nCols,nLags]=size(m);
0039         [nSamples,nChans,nTrials]=size(x);
0040         if nChans~=nRows; 
0041             error('ncols(x) ~= nrows(m)');
0042         end
0043         % convolution: for each k, multiply x by m(:,:,k) and add with
0044         % shift of (k-1)
0045         y=zeros(nSamples+nLags-1,nCols,nTrials);
0046         for iLag=1:nLags
0047             y(iLag:iLag+nSamples-1,:,:) = y(iLag:iLag+nSamples-1,:,:) + nt_mmat0(x,m(:,:,iLag));
0048         end
0049         
0050     end
0051 end    
0052 
0053 function x=nt_mmat0(x,m)
0054 x=nt_fold(nt_unfold(x)*m,size(x,1));