How is "quartely" defined in the function retime?

Hi. I am using retime to get quarterly mean data from irregularly distributed measurements:

yTT = retime(xTT,'quarterly',@nanmean);

My first measurement is on the 1st of April. The new yTT timetable has now timesteps at April, July, September, and January. My question is now are these quarterly values means of the widely used Mar-Apr-May, Jun-Jul-Aug, Sep-Oct-Nov, and Dec-Jan-Feb? Or is the April quarterly value an average of Apr-May-Jun, and so on?

Answer 1

[Adam Danz]

According to Wiki,

In Gregorian calendar:

• First quarter, Q1: 1 January – 31 March (90 days or 91 days in leap years)
• Second quarter, Q2: 1 April – 30 June (91 days)
• Third quarter, Q3: 1 July – 30 September (92 days)
• Fourth quarter, Q4: 1 October – 31 December (92 days)

While in the Chinese calendar, the quarters are traditionally associated with the 4 seasons of the year:

• Spring: 1st to 3rd month
• Summer: 4th to 6th month
• Autumn: 7th to 9th month
• Winter: 10th to 12th month

https://en.wikipedia.org/wiki/Calendar_year

The months you see are the first month within the quarter.

[Fabrice Lambert]

I went to look at the code of the function "synchronize" and it appears that Matlab will not follow any Gregorian or Chinese calendar, but will just add 3 months to the first date stamp.

[Adam Danz]

Thanks for the comment, Fabrice Lambert. Where in the synchronize function did you see that (using r2020a, I assume)?

This example suggests otherwise. The timetable starts on Feb-1-2019 and contains a value for every 3 months but some of the 3-month periods are missing. When the retime fills in the table using the quarterly times, the output timetable only shows the starting month of each gregorian quarter.

startdate = datetime(2019,02,01); 
dates = startdate + calmonths(0:3:30)'; 
dates([2,5,10]) = []; 
TT = timetable(dates, (1:numel(dates))'); % Input
TTr = retime(TT,'quarterly','previous');  % Output retime'd

% TT =
%   8×1 timetable
%        dates       Var1
%     ___________    ____
%     01-Feb-2019     1  
%     01-Aug-2019     2  
%     01-Nov-2019     3  
%     01-May-2020     4  
%     01-Aug-2020     5  
%     01-Nov-2020     6  
%     01-Feb-2021     7  
%     01-Aug-2021     8  

% TTr =
%   12×1 timetable
%        dates       Var1
%     ___________    ____
%     01-Jan-2019    NaN 
%     01-Apr-2019      1 
%     01-Jul-2019      1 
%     01-Oct-2019      2 
%     01-Jan-2020      3 
%     01-Apr-2020      3 
%     01-Jul-2020      4 
%     01-Oct-2020      5 
%     01-Jan-2021      6 
%     01-Apr-2021      7 
%     01-Jul-2021      7 
%     01-Oct-2021      8 

[Fabrice Lambert]

Indeed. I may have misunderstood the code. Using R2020a, the relevant section in synchronize is between lines 495 and 518. The comment says "Given a choice of time step name and the min/max times of the data being synchronized, return nicely-aligned spanning time limits and the actual time step."

But then at line 514 they use timeshift to move the original tmin and tmax, so you may be right, the new time stamps will be between the two closest earlier quarters.

timeshift still doesn't define what a quarter actually is, but looking at your example it appears they use the gregorian ones.

Thanks!

[Adam Danz]

You're exactly right. That section of the code shifts the starting time of the timetable to the nearest quarter and then increments by 3-months from there.

I also agree that the documentation could be clearer. If you go to the retime() and synchronize() documentation pages online and scroll to the bottom, you can give those pages a star rating and then a window will appear allowing you to give feedback. I'll make this suggestion on both pages (you could, too, if you'd like).

[Fabrice Lambert]

will do