ts = setabstime (ts, dates) gives the samples of the series ts the dates dates, one per sample, in order: a cell array of character vectors, a string array, a character matrix of one date per row, a character vector for a single sample, or a datetime array. The earliest date becomes TimeInfo.StartDate, stored as 'dd-mmm-yyyy HH:MM:SS' with milliseconds as '.FFF' when the seconds are not whole, and Time holds each date’s offset from it in the series’ own TimeInfo.Units, which are kept: a series in seconds given one date a day gets times [0 86400 172800 …]. Each date is read by its own format. Dates given as times of day alone, such as '13:00:00', give times since midnight and no start date, as in MATLAB.

ts = setabstime (ts, dates, format) reads every date strictly in the datestr format format. A date not written in format is refused. The start date is stored as above whatever the format, where MATLAB writes it in format, from which every later reading has to guess it again: 'dd/mm/yyyy' then comes back as a date of another month.

The dates need not be sorted: the samples, with their quality codes, are sorted with them, keeping the order of samples at equal dates, as the constructor sorts them.

MATLAB differs in four ways. It does not move the samples with unsorted dates, so each is given another sample’s date. It reads every date by the format of the first, so {'01-Jan-2024', '2024-01-02'} gives a date in year 7. It forces text into format when it does not fit, so 'yyyy-mm-dd' on '01-Jan-2024' gives a date in year 6. And it takes datenum values, datetime values and plain numbers as relative seconds, dropping the dates; here numbers are refused, since a datenum cannot be told from a plain number, and a datetime is read as its dates. Convert a datenum x with datetime (x, 'ConvertFrom', 'datenum').

See also: timeseries.getabstime, timeseries.setuniformtime

Source Code: timeseries