Thread Subject: Making this section of code faster...

I have cells of different lengths,

Let's say,

A={[1:4];[1:5];[1:2]};

I am trying to make this into one matrix by completing the empty spaces with 0.

So A looks like this,

1 2 3 4 0
1 2 3 4 5
1 2 0 0 0

I am using the following code, (S_Seq is a cell of 5x1 with each cell being a (12x1 cell),

[Eij_CM tf] = padcat(S_Seq{s,1}{:});
Eij_CM(~tf) = 0;

padcat is in the FEX.

The thing is I am calling this line 1656000 times, taking a total of 319.317 sec.

I tried changing it to this code but takes longer,
lns = cellfun(@length,S_Seq{s,1});
mx = max(lns);
mat = 0;

for ii = 1:numel(S_Seq{s,1})
S_Seq{s,1}{ii,1} = [S_Seq{s,1}{ii,1} mat(1,ones(1,mx-lns(ii)))];
end

cell2mat(S_Seq{s,1})

If anyone has a clue how I can run this faster it would make my code a hell lot better! Thanks! Any questions or clarifications let me know!


FYI padcat function is,

function [M, TF] = padcat(varargin)
error(nargchk(1,Inf,nargin)) ;

SZ = cellfun(@size,varargin,'UniformOutput',false) ; % sizes
Ndim = cellfun(@ndims,varargin) ; %

if ~all(Ndim==2)
    error([mfilename ':WrongInputDimension'], ...
        'Input should be vectors.') ;
end

TF = [] ; % default second output so we do not have to check all the time

SZ = cat(1,SZ{:}) ;
maxSZ = max(SZ) ; % probable size of the longest vector

if ~any(maxSZ == 1), % hmm, not all elements are 1-by-N or N-by-1
    if any(maxSZ==0),
        M = [] ;
        return
    else
        error([mfilename ':WrongInputSize'], ...
            'Inputs should be all row vectors or all column vectors.') ;
    end
end

if nargin == 1,
    M = varargin{1} ;
else
    dim = (maxSZ(1)==1) + 1 ; % Find out the dimension to work on
    X = cat(dim, varargin{:}) ; % make one big list

    if maxSZ(dim) == 1,
        M = X ; % copy the list
    elseif all(SZ(:,dim)==SZ(1,dim)),
        M = reshape(X,SZ(1,dim),[]) ;% copy the list and reshape
    else
        M = zeros([maxSZ(dim)+1 nargin]) ;
        M(sub2ind(size(M), SZ(:,dim).'+1, 1:nargin)) = 1 ;
        M = cumsum(M(1:end-1,:),1) ; % remove last row

        if nargout>1,
            TF = ~M ;
            M(~TF) = NaN ; % put the fillers in
            M(TF) = X ; % put the values in
        else
            M(M==1) = NaN ; % put the fillers in
            M(M==0) = X ; % put the values in
        end
    end

    if dim == 2,
        M = M.' ;
        TF = TF.' ;
    end
end % nargin == 1

if nargout > 1 && isempty(TF),
    TF = true(size(M)) ;
end

As was shown in your original thread....


clear all
a = {[1:4];[1:5];[1:2]};
t = zeros(1,2);

tic
for ii = 1:10000
    lns = cellfun(@length,a);
    mt = zeros(length(lns),max(lns));
    for ii = 1:length(lns)
        mt(ii,1:lns(ii)) = a{ii};
    end
end
t(1) = toc;



tic
for ii = 1:10000
    [b,tf] = padcat(a{:});
    b(~tf) = 0 ;
end
t(2) = toc;


t/min(t)
ans =
       1 6.34968004321936






aiXXV5eeZVkVWjidtZd{]NbaV:tZtidW^ttZd#dVb\jcB/]{]cVdnt^`tVt

"Diego Zegarra" <diegozbb@gmail.com> wrote in message <gnfv9u$3as$1@fred.mathworks.com>...

<.. problem that can be handled by padcat

> padcat is in the FEX.
...
> FYI padcat function is,
< ... copy of padcat code ...

I am happy you have found some use of my PADCAT function, Diego. However, you better provide a link to the code, rather than copy it partially, leaving out the most important section: the help ....

Jos