Is it possible to optimize converting a C# array of serial times (double) to a Matlab array of datetimes?

Here's an example of what I'm doing:

function outputArg = TimeArray(obj)
    timeArray = DotnetClass.GetTimeArray(obj.DotnetArray);
    
    count = timeArray.Length;
    outputArg = NaT(1, count);
    
    for i=1:count
        netTime = timeArray(i);
        matTime = datetime(netTime, "ConvertFrom", "datenum");
        outputArg(i) = matTime;
    end
end

I'm converting an array of DateTimes in C# to a serial date which comes across as a double. I then need to convert these to a matlab array of datetimes for plotting.

The above implementation is taking about 30 seconds for ~110k datenums. I've used cellfun and it takes about 4 times as long, so I've ruled that out.

Thanks for any help!

Answer 1

[Steven Lord]

datetime can convert an array of serial date numbers at once, and this avoids the overhead of calling datetime repeatedly.

>> D = datenum(datetime('now') + 10*randn(110000, 1));
>> timeit(@() datetime(D, 'ConvertFrom', 'datenum'))

This took a (small) fraction of a second on my machine.

[Matthew Trahan]

This would be perfect, but for me, timeArray is a System.Object[] (Object because the cell(timeArray) method requires it to be an object, but I have the same results with a System.Double[]).

I can convert timeArray to a cell array with

cellTimeArray = cell(timeArray);

but it ends up in a format like

{[7.3725e+05]}    {[7.3725e+05]}    {[7.3725e+05]}    {[7.3725e+05]}

so I guess it's not quite converted how I need.

[Matthew Trahan]

Scratch that, when I return it as a System.Double[], it works. Thank you so much!