Generating array of consecutive timestamps in double

I need to generate an array of consecutive timestamps, in double and yyyymmddHHMMSS format, for specified start and end times. For example:

timestamps(1) = 2.012082104000000e+013;
timestamps(2) = 2.012082104000100e+013;
...
...
timestamps(120) = 2.012082104020000e+013;

I wrote a working piece of code, but the str2double/datestr combination in the parfor loop is too slow (I used 120 seconds as an example, but in actual use, I will be generating about 10^7 consecutive seconds of timestamps). Is there a faster way to do this?

if matlabpool('size') == 0
    matlabpool local
end

startDate = num2str(2.012082104000000e+013);
endDate = num2str(2.012082104020000e+013);

startDateyyyy = str2double(startDate(:,1:4));
startDatemm = str2double(startDate(:,5:6));
startDatedd = str2double(startDate(:,7:8));
startDateHH = str2double(startDate(:,9:10));
startDateMM = str2double(startDate(:,11:12));
startDateSS = str2double(startDate(:,13:14));
startDateVec = [startDateyyyy startDatemm startDatedd startDateHH ...
    startDateMM startDateSS];
startDate = datenum(startDateVec);

endDateyyyy = str2double(endDate(:,1:4));
endDatemm = str2double(endDate(:,5:6));
endDatedd = str2double(endDate(:,7:8));
endDateHH = str2double(endDate(:,9:10));
endDateMM = str2double(endDate(:,11:12));
endDateSS = str2double(endDate(:,13:14));
endDateVec = [endDateyyyy endDatemm endDatedd endDateHH ...
    endDateMM endDateSS];
endDate = datenum(endDateVec);

timestamps = startDate:(1/24*1/60*1/60):endDate;

parfor i = 1:length(timestamps)
    timestamps(i) = str2double(...
        datestr(timestamps(i),'yyyymmddHHMMSS'));
end

Thanks in advance!

[Oleg Komarov]

Why not serial dates as generated by datenum()?

[Ephedyn]

^Thanks for your prompt response! It has to do with two reasons, mainly for standardization and what looks more pleasing to the eye. The function that I'm writing imports time series data fetched from a brokerage API, and this is the standard format for their timestamps and the rest of the programs I've written to manipulate the data.

Thereafter, the function runs a loop of tests to catch missing data points from the data feed and cleverly fill them in so there are no discontinuities - and other things, so the timestamps I've generated must match the format of the timestamps from the data feed to carry out these tests. I've written it all to work in this format before going back to profile my code, and figured it would be easier to optimize this part alone than to change everything. (: Any suggestion?

[Oleg Komarov]

As you can see from my answer, the appearance of dates can be controlled with datestr() and I still recommend to store dates as MATLAB serial numbers produced with datenum(). I am posting an alternative solution to let you see how bothersome is to store dates in a different way.

Answer 1

[Oleg Komarov]

I recommend to use MATLAB serial dates as produced by datenum(). Any manipulation with dates becomes much easier. The appearance can be controlled with datestr().

% Generate serial dates increasing by one second from 2012-08-21 04:00:00
dates = datenum(2012,08,21,4,0,0:120)

% Visualize them in format 31 yyyy-mm-dd HH:MM:SS
datestr(dates,31)

Alternatively:

st   = 20120821040000;
en   = 20120821040200;

sten = [st; en];
y    = floor(sten/1e10);
m    = floor(rem(sten, 1e10)/1e8);
d    = floor(rem(sten, 1e8)/1e6);
HH   = floor(rem(sten, 1e6)/1e4);
MM   = floor(rem(sten, 1e4)/1e2);
SS   = rem(sten, 1e2);

% Again MATLAB serial dates
stenser = datenum(y,m,d,HH,MM,SS);
dates2  = stenser(1):1/(60^2*24):stenser(2);

% Controlling appearance
datestr(dates2,31);

% Now you have to reverse into yyyymmddHHMMSS
dtvc = datevec(dates2);
out  = dtvc(:,1)*1e10 +  dtvc(:,2)*1e8 + dtvc(:,3)*1e6 +...
       dtvc(:,4)*1e4  + dtvc(:,5)*1e2  + dtvc(:,6);

[Ephedyn]

Exactly what I need for now! I will take your advice and try to clean up everything to MATLAB serial dates as produced datenum. Thanks!

Answer 2

[Amarjit Dhillon]

Let's say we want to generate timestamps in the format of yyyymmddHHMMSS for 10 days

1-Jan-2017 at 12:00:00 to 10-Jan-2017 till 12:00:00, This can be easily generated by below-written code

t1 = datetime(2017,1,1,12,0,0);
t2 = datetime(2017,1,10,12,0,0);
data = t1:seconds(1):t2;
datestr(data,'yyyymmddHHMMSS')

[Peter Perkins]

Amarjit, your suggestion to use datetime is a good one (the OP was from 1012, before datetime existed in MATLAB). But actually, there's usually no reason to convert it to text. Just change the format:

>> data.Format = 'yyyymmddHHMMSS'
data = 
  1×3 datetime array
   20170001120100   20170001120100   20170001120100 ...