Finding a maximum period

Hey Guys,

I'm trying to wrap my brain around how to find n maximum or minimum values in a row. I've used grpstats to generate mean values, but need to find the three highest or lowest in a row, so I can't use a sort function:

[m, g]=grpstats(data(:, 2), data(:, 1), {'mean', 'gname'})

m =

   16.4605
   19.6645
   19.0127
   20.0246
   21.0003
   20.0367
   15.7379
   22.7125
   19.9485
   20.4098
   21.6320
   17.6872

g =

    '1'
    '2'
    '3'
    '4'
    '5'
    '6'
    '7'
    '8'
    '9'
    '10'
    '11'
    '12'

Any ideas or functions to suggest? Thanks and all the best.

[Walter Roberson]

Are you looking for a moving maximum, where every point is replaced by the maximum of the point and the point before and the next point? Are you looking for a single group of three values in a row that has the highest average out of all of the sliding possibilities?

[Kate]

The later, I'm looking for the 3 highest (or lowest) consecutive values. This is a metric called Mean Temperature of the Warmest Quarter. Thanks Walter.

Answer 1

[Image Analyst]

If you're looking for the group of 3 elements that have the highest sum (and mean) then you can use conv() or movmean()

threeElementSum = conv(data, [1,1,1], 'same');
[maxSum, indexOfMax] = max(threeElementSum);

This will tell you where the center of the 3 element window is where the values inside that 3 element window have the highest sum of any window location in the data. Is that what you're looking for?

[Kate]

Perfect, convolution is exactly what I was looking for. I appreciate it.

Answer 2

[Alexandra Harkai]

If you're looking for the n highest and lowest entry for each row of input data, you can still use the sort function:

[sorted, row_idx] = sort(data, 2);

sorted will contain the sorted data, row_idx will contain the row-wise indices, so:

sorted(:,1:n)            % n smallest elements in each row
sorted(:,(end+1-n):end)  % n largest elements in each row
row_idx(1,1:n)           % indices in rows for the n smallest elements in each row
row_idx(1,(end+1-n):end) % indices in rows for the n largest elements in each row

[Kate]

It's a great crack, but essentially I can't sort the values, because then the months are split up. Here's an example result of your code:

row_idx((end+1-n):end, 1)

ans =

     5
    11
     8

Indicating May, November and August, which aren't proximal in time. I appreciate the help though.