# How to slice each string in a string array without using for loop

94 views (last 30 days)
YANAN ZHU on 18 Sep 2018
Edited: Cedric Wannaz on 19 Sep 2018
For a string array, for example,
celldata =
3×1 cell array
{'2018-12-12'}
{'2018-11-05'}
{'2018-09-02'}
Is there array operation (i.e. without using for loop) to extract the months from each cell and convert them to a 3*1 numeric matrix, which should be [12;11;09]. I don't want to use for loop because it was too slow.

Cedric Wannaz on 19 Sep 2018
Edited: Cedric Wannaz on 19 Sep 2018
If you favor performance over readability/maintainability, you can build an approach around the following:
buffer = vertcat( dates{:} ) ;
months = (buffer(:,6:7) - '00') * [10;1] ;
where dates is a cell array of date strings ( celldata in your example).
Here is the benchmark:
% Build "large" data set.
N = 1e4 ;
dates = repmat( {'2018-12-12'; '2018-11-05'; '2018-09-02'}, N, 1 ) ;
% Basic FOR loop, STR2DOUBLE.
tic ;
n = numel( dates ) ;
months_forStr2double = zeros( n, 1 ) ;
for k = 1 : n
months_forStr2double(k) = str2double( dates{k}(6:7) ) ;
end
fprintf( 'Basic FOR, STR2DOUBLE : %.3fs\n', toc ) ;
% Basic FOR loop, STR2NUM.
tic ;
n = numel( dates ) ;
months_forStr2num = zeros( n, 1 ) ;
for k = 1 : n
months_forStr2num(k) = str2num( dates{k}(6:7) ) ;
end
fprintf( 'Basic FOR, STR2NUM : %.3fs\n', toc ) ;
% Basic FOR loop, SSCANF.
tic ;
n = numel( dates ) ;
months_forScanf = zeros( n, 1 ) ;
for k = 1 : n
months_forScanf(k) = sscanf( dates{k}(6:7), '%d' ) ;
end
fprintf( 'Basic FOR, SSCANF : %.3fs\n', toc ) ;
% CELLFUN (hidden FOR), SSCANF.
tic ;
months_cellfun = cellfun( @(date) sscanf( date(6:7), '%d' ), dates ) ;
fprintf( 'CELLFUN, SSCANF : %.3fs\n', toc ) ;
% REGEXP
tic ;
months_regexp = str2double( regexp( dates,'(?<=-)\d+(?=-)','match','once' )) ;
fprintf( 'REGEXP : %.3fs\n', toc ) ;
% CELL2MAT, STR2NUM
tic ;
chardata = cell2mat( dates ) ;
months_cell2matStr2num = str2num( chardata(:,6:7) ) ;
fprintf( 'CELL2MAT, STR2NUM : %.3fs\n', toc ) ;
% DATETIME
tic ;
dt = datetime( dates) ;
months_datetime = month( dt ) ;
fprintf( 'DATETIME : %.3fs\n', toc ) ;
% SSCANF
tic ;
months_sscanf = sscanf([dates{:}],'%*4d-%2d-%*2d') ;
fprintf( 'SSCANF : %.3fs\n', toc ) ;
% EXTRACTBETWEEN
tic ;
months_extractBetween = extractBetween( dates, '-', '-' ) ;
months_extractBetween = cellfun( @str2double, months_extractBetween ) ;
fprintf( 'EXTRACTBETWEEN : %.3fs\n', toc ) ;
% Trick.
tic ;
buffer = vertcat( dates{:} ) ;
months_trick = (buffer(:,6:7) - '00') * [10;1] ;
fprintf( 'Trick: %.3fs\n', toc ) ;
% Check
disp( [isequal( months_forStr2num, months_forStr2double ), ...
isequal( months_forScanf, months_forStr2double ), ...
isequal( months_cellfun, months_forStr2double ), ...
isequal( months_regexp, months_forStr2double ), ...
isequal( months_cell2matStr2num, months_forStr2double ), ...
isequal( months_datetime, months_forStr2double ), ...
isequal( months_sscanf, months_forStr2double ), ...
isequal( months_extractBetween, months_forStr2double ), ...
isequal( months_trick, months_forStr2double )] ) ;
Output:
Basic FOR, STR2DOUBLE : 0.489s
Basic FOR, STR2NUM : 0.975s
Basic FOR, SSCANF : 0.356s
CELLFUN, SSCANF : 0.550s
REGEXP : 0.673s
CELL2MAT, STR2NUM : 0.015s
DATETIME : 0.201s
SSCANF : 0.023s
EXTRACTBETWEEN : 0.624s
Trick: 0.008s
1 1 1 1 1 1 1 1 1
##### 2 CommentsShowHide 1 older comment
Cedric Wannaz on 19 Sep 2018
Thanks Stephen, just added this to the benchmark!

### More Answers (4)

Paolo on 18 Sep 2018
Edited: Paolo on 18 Sep 2018
str2double(regexp(celldata,'(?<=-)\d+(?=-)','match','once'))
ans =
12
11
9

Christopher Wallace on 19 Sep 2018
chardata = cell2mat(a);
numdata = str2num(chardata(:,6:7));
##### 1 CommentShowHide None
Cedric Wannaz on 19 Sep 2018
No, I am CW ;-)

Star Strider on 19 Sep 2018
Using datetime and its functions:
celldata = [{'2018-12-12'}
{'2018-11-05'}
{'2018-09-02'}];
dt = datetime(celldata);
M = month(dt)
M =
12
11
9

Akira Agata on 19 Sep 2018
Another possible solution:
celldata = [{'2018-12-12'}
{'2018-11-05'}
{'2018-09-02'}];
M = extractBetween(celldata,'-','-');
M = cellfun(@str2double,M);