MATLAB: Accumarray with tall arrays

accumarraygatherlooptalltall array

Accumarray is not in the list of functions that support tall arrays. Is there a way to do what it does with tall arrays?
For example, suppose I have a tall array of dates tt.DATE and a tall array of corresponding values tt.VAL. How can I sum tt.VAL for each unique date in tt.DATE?
uniqDate = gather(unique(tt.DATE);
sumVal = zeros(length(uniqDate),1);
for i = 1:length(uniqDate)
 thisInd = tt.DATE == uniqDate(i);
 thisSum = gather(tt.VAL(thisInd));
 sumVal(i, 1) = thisSum;
end
This approach works except that it requires a call to gather at each step so it is far too slow. If I could write the gather statement outside of the loop somehow, I imagine that would help, but I can't figure out how to do it.

Best Answer

You can use findgroups and splitapply to do this, like so:
% Make some example data
tt = tall(table(datetime(2017, 03, randi([1 31], 100, 1)), ...
    rand(100, 1), 'VariableNames', {'Date', 'Value'}));
% group by date
g = findgroups(tt.Date);
% Call splitapply to find the sum of values on each unique date
sumAndDate = splitapply(@(v, d) {sum(v), d(1)}, tt.Value, tt.Date, g)
Note that because findgroups for tall arrays doesn't support the second output argument, I've concocted a slightly unusual function for the splitapply stage which returns both the sum and the datetime corresponding to that group.
Related Question