Unify timestamps as date strings - matlab

Question

MATLAB R2015b
I have a table containing a date string and a time string in various formats in two columns for each row:
11.01.2016 | 00:00:00 | data
10/19/16 | 05:29:00 | data
12.02.16 | 06:40 | data
I want to convert this two columns to one column with a common format:
31.12.2017 14:00:00
My current solution uses a loop over each row and combines the columns as strings, checks for the various formats to use datetime with an appropriate format string and then uses datestr with the desired format string. Datetime was not able to automatically determine the format of the input string.
As you can imagine, this is horribly slow for large tables (approx. 50000 rows).
Is there any faster solution?
Thanks in advance.

Answer 1

I gave a try to vectorize the code. The trick is to
convert tables > cell > char-array, then
manipulate char strings, then
convert back from char-array > cell > table
Also, there is an important bit to pad all cells having shorter lenths with 'null' character in a vectorized way. Without this, it will not be possible to convert from cell > char-array. Here is the code.
clc
clear all
%% create Table T
d={'11.01.2016';
'10/19/16';
'12.02.16'};
t={'00:00:00';
'05:29:00';
'06:40'};
dat=[123;
456;
789];
T = table(d,t,dat);
%% deal with dates in Table T
% separate date column and convert to cell
dd = table2cell(T(:,1));
% equalize the lengths of all elements of cell
% by padding 'null' in end of shorter dates
nmax=max(cellfun(#numel,dd));
func = #(x) [x,zeros(1,nmax-numel(x))];
temp1 = cellfun(func,dd,'UniformOutput',false);
% convert to array for vectorized manipulation of char strings
ddd=cell2mat(temp1);
% replace the separators in 3rd and 6th location with '.' (period)
ddd(:,[3 6]) = repmat(['.' '.'], length(dd),1);
% find indexes of shorter dates
short_year_idx = find(uint16(ddd(:,nmax)) == 0);
% find the year value for those short_year cases
yy = ddd(short_year_idx,[7 8]);
% replace null chars with '20XX' string in desirted place
ddd(short_year_idx,7:nmax) = ...
[repmat('20',size(short_year_idx,1),1) yy];
% convert char array back to cell and replace in table
dddd = mat2cell(ddd,ones(1,size(d,1)),nmax);
T(:,1) = table(dddd);
%% deal with times in Table T
% separate time column and convert to cell
tt = table2cell(T(:,2));
% equalize the lengths of all elements of cell
% by padding 'null' in end of shorter times
nmax=max(cellfun(#numel,tt));
func = #(x) [x,zeros(1,nmax-numel(x))];
temp1 = cellfun(func,tt,'UniformOutput',false);
% convert to array for vectorized manipulation of char strings
ttt=cell2mat(temp1);
% find indexes of shorter times (assuming only ':00' in end is missing
short_time_idx = find(uint16(ttt(:,nmax)) == 0);% dirty hack, as null=0 in ascii
% replace null chars with ':00' string
ttt(short_time_idx,[6 7 8]) = repmat(':00',size(short_time_idx,1),1);
% convert char array back to cell and replace in table
tttt = mat2cell(ttt,ones(1,size(t,1)),nmax);
T(:,2) = table(tttt);

Answer 2

If you call the two columns cell arrays c1 and c2, then something like this should work:
c = detestr(datenum(strcat(c1,{' '},c2)), 'dd.mm.yyyy HH:MM:SS')
Then you would need to drop the old columns and put this one c in their place. On the inside, datenum must be doing something similar to what you're doing, however, so I'm not sure if this will be faster. I suspect that it is because (we can hope) the standard functions are optimized.
If your table isn't representing those as cell arrays, then you may need to do a pre-processing step to form the cell arrays for strcat.