Group a list of Object arrays by timestamp in Java -

i have list < object[]> , 1 of columns in object[] localdatetime. other columns location (string) , item price (double).

basically, list looks this:

2017-01-01 02:05:00   newyork   26.89 2017-01-01 02:10:00   newyork   72.00 2017-01-01 02:15:00   newyork   73.10 2017-01-01 02:20:00   newyork   70.11 2017-01-01 02:25:00   newyork   79.90 2017-01-01 02:30:00   newyork   72.33 2017-01-01 02:35:00   newyork   75.69 2017-01-01 02:40:00   newyork   72.12 2017-01-01 02:45:00   newyork   73.09 2017-01-01 02:50:00   newyork   72.67 2017-01-01 02:55:00   newyork   72.56 2017-01-01 03:00:00   newyork   72.76  2017-01-01 02:05:00   boston    26.89 2017-01-01 02:10:00   boston    42.00 2017-01-01 02:15:00   boston    23.10 2017-01-01 02:20:00   boston    77.11 2017-01-01 02:25:00   boston    49.92 2017-01-01 02:30:00   boston    72.63 2017-01-01 02:35:00   boston    73.19 2017-01-01 02:40:00   boston    76.18 2017-01-01 02:45:00   boston    83.59 2017-01-01 02:50:00   boston    76.67 2017-01-01 02:55:00   boston    52.06 2017-01-01 03:00:00   boston    76.06 

what need time-weighted average of price on interval of 15-minutes, per city. datetime associated interval latest one. running algorithm on list above produce list looks this:

01-01-2017 02:15:00   newyork   57.33 (average of 2:05, 2:10 , 2:15) 01-01-2017 02:30:00   newyork   74.11 (average of 2:20, 2:25 , 2:30) 01-01-2017 02:45:00   newyork   73.63 (...) 01-01-2017 03:00:00   newyork   72.60 01-01-2017 02:15:00   boston    30.66 (average of 2:05, 2:10 , 2:15) 01-01-2017 02:30:00   boston    66.55 (average of 2:20, 2:25 , 2:30) 01-01-2017 02:45:00   boston    77.65 (...) 01-01-2017 03:00:00   boston    68.26 

i'm thinking first step in doing group records 15 minute interval , city. rest matter of iterating through groups , getting average, can figure out on own.

i have no idea how go grouping per localdatetime , less on 15 minute basis. 1 last thing mention there missing rows. intervals empty, in case can ignore interval altogether. appreciated.

update1: i'm assuming there better way group them sort them , iterating on each 1 , comparing timestamps. first answer in post: how group objects in list other lists attribute using streams & java 8?

update2: also, timestamps not every 5 minutes. @ random times , intervals have 3 or 5 rows in them.

update3: not duplicate, question grouping , not rounding down. understand how rounding down 15 minutes 1 way of doing it, afterwards, i'd have keep real timestamps perform time-weighted average. not way this.

design class hold data instead of object arrays. each object of class hold timestamp, location , item price. may hold result of rounding timestamp down whole 15 minutes; alternatively method rounding down. understand, can rounding (otherwise there’s inspiration found in this question: round time seconds).

with such class can use stream the answer linked to group , average. if prefer, may start stream list <object[]> , map each array object before further processing.

edit seem understand prefer without class rows. of course can done:

private static list<object[]> averagebyquarterofhour(final int indexoftime,          int othergroupingindex, int indextoaverage, list<object[]> mylist) {     return mylist.stream()             .collect(collectors.groupingby(arr                      -> arrays.aslist(rounduptowholequarterofhour((localdatetime) arr[indexoftime]),                              arr[othergroupingindex])))             .entryset()             .stream()             .map(e -> new object[] { e.getkey().get(0),                      e.getkey().get(1),                      e.getvalue().stream()                             .map(arr -> (number) arr[indextoaverage])                             .maptodouble(number::doublevalue)                             .average()                             .getasdouble() })             .collect(collectors.tolist()); }  static localdatetime rounduptowholequarterofhour(localdatetime timetoround) {     localdatetime truncated = timetoround.truncatedto(chronounit.minutes);     int minute = truncated.getminute();     if (truncated.isequal(timetoround) && minute % 15 == 0) { // on whole quarter         return timetoround;     }     int minutestoadd = 15 - (minute % 15);     return truncated.plusminutes(minutestoadd); } 

feeding list averagebyquarterofhour() gives:

[2017-01-01t03:00, newyork, 72.66333333333334] [2017-01-01t02:15, newyork, 57.330000000000005] [2017-01-01t02:30, newyork, 74.11333333333333] [2017-01-01t02:45, newyork, 73.63333333333334] [2017-01-01t02:45, boston, 77.65333333333334] [2017-01-01t03:00, boston, 68.26333333333334] [2017-01-01t02:15, boston, 30.663333333333338] [2017-01-01t02:30, boston, 66.55333333333333] 

i leave sorting you.

you may want think twice though. class rows may have advantage of better modelling, impact entire application, not average calculation, though on other hand require model each of entities separately, ever data entity holds. more locally, class, auxiliary class holding array , method getting quarter of hour, might still make above code more readable.