# Extracting from vector to multidimensional array using indices provided

1 view (last 30 days)
Andris Cerins on 14 Apr 2020
Commented: Andris Cerins on 18 Apr 2020
Can this extraction be performed better without loops?
% data vector
data = [1 1 1 1 1 4 4 4 4 4 7 7 7 7 7 2 2 2 2 2 5 5 5 5 5 8 8 8 8 8 3 3 3 3 3 6 6 6 6 6 6 9 9 9 9 9]
% Desired output
data3d(:,:,1)
1 1 1 1 1
2 2 2 2 2
3 3 3 3 3
data3d(:,:,2)
4 4 4 4 4
5 5 5 5 5
6 6 6 6 6
data3d(:,:,3)
7 7 7 7 7
8 8 8 8 8
9 9 9 9 9
% indices for starts and ends of trials, rows are channels
dstart = [1 16 31; 6 21 36; 11 26 42]
dend = [5 20 35; 10 25 41; 15 30 46]
% note the annoying extra 6 in data
% can truncate now, dend_tr = dstart+4, or just use dstart:dstart+4
% or if target matrix pads then truncate later?
[numchannels,numsweeps] = size(dstart) %in case it is useful
% Here is how I think it would work using loops
% For all channels in that file
for ich = 1:numchannels
data_sweeps = [];
% For all sweeps in that file
for isw = 1:numsweeps
sweepdata = [];
% Get that sweep, truncate it to 5 samples (because some have extras)
sweepdata = data(dstart(ich,isw):dstart(ich,isw)+4);
% Make sweeps x samples matrix
data_sweeps(isw,:) = sweepdata;
end
% Adds each channel as a page in a 3d matrix
data3d(:,:,ich) = data_sweeps
end

Stephen Cobeldick on 17 Apr 2020
Edited: Stephen Cobeldick on 17 Apr 2020
In three lines (could be just one), no loops, and yes that extra 6 is taken into account!
>> D = [1,1,1,1,1,4,4,4,4,4,7,7,7,7,7,2,2,2,2,2,5,5,5,5,5,8,8,8,8,8,3,3,3,3,3,6,6,6,6,6,6,9,9,9,9,9];
>> S = [1,16,31;6,21,36;11,26,42];
>> E = [5,20,35;10,25,41;15,30,46];
>> F = @(s)reshape(s:s+4,[],1);
>> X = cell2mat(arrayfun(F,sort(S(:)),'uniformoutput',false));
>> A = permute(reshape(D(X),[5,3,3]),[3,1,2])
A(:,:,1) =
1 1 1 1 1
2 2 2 2 2
3 3 3 3 3
A(:,:,2) =
4 4 4 4 4
5 5 5 5 5
6 6 6 6 6
A(:,:,3) =
7 7 7 7 7
8 8 8 8 8
9 9 9 9 9

#### 1 Comment

Andris Cerins on 18 Apr 2020
Thank you Stephen, much appreciated!
Tested as below on a part of my actual data where data is 1x7005392 and datastart is a 4x350
Your code is 5x faster than my loop
trunc = 5000 % where I choose the truncation length
[numchannels, numsweeps] = size(datastart);
F = @(s)reshape(s:s+(trunc-1),[],1);
X = cell2mat(arrayfun(F,sort(datastart(:)),'uniformoutput',false));
data3d = permute(reshape(data(X),[trunc,numchannels,numsweeps]),[3,1,2]);

Guru Mohanty on 17 Apr 2020
I understand you are trying to extract data from a vector to multidimensional array without for loops.
It can be possible using MATLAB Array indexing. Here is a sample code for it.
% data vector
tic;
data = [1 1 1 1 1 ...
4 4 4 4 4 ...
7 7 7 7 7 ...
2 2 2 2 2 ...
5 5 5 5 5 ...
8 8 8 8 8 ...
3 3 3 3 3 ...
6 6 6 6 6 ...
9 9 9 9 9];
data=reshape(data,5,9);
dstart = [1 4 7;
2 5 8;
3 6 9];
data3d(:,:,1) = [data(:,dstart(1,1))';data(:,dstart(1,2))';data(:,dstart(1,3))'];
data3d(:,:,2) = [data(:,dstart(2,1))';data(:,dstart(2,2))';data(:,dstart(2,3))'];
data3d(:,:,3) = [data(:,dstart(3,1))';data(:,dstart(3,2))';data(:,dstart(3,3))'];
toc;
Using MATLAB Indexing instead of for loop will improve your Code performance. You can also check Execution time using tic & toc command.

#### 1 Comment

Stephen Cobeldick on 17 Apr 2020
@Guru Mohanty: your data vector is missing a 6 (as the original question points out, there are more of them). So your data vector should actually be defined as:
data = [1 1 1 1 1 ...
4 4 4 4 4 ...
7 7 7 7 7 ...
2 2 2 2 2 ...
5 5 5 5 5 ...
8 8 8 8 8 ...
3 3 3 3 3 ...
6 6 6 6 6 6 ... <- "note the annoying extra 6 in data"
9 9 9 9 9];
and then reshape fails with an error.
Using MATLAB permute instead of copy-and-pasting the same code nine times will improve your Code performance.