# Calculate monthly averages without reshaping

1 view (last 30 days)
aaron Harvey on 18 Feb 2016
Edited: Stephen23 on 24 Feb 2016
i have a time series of data and i would like to calculate monthly averages but keeping the vector lengths the same. (the variables also contain many nans)
For example if i have a date vector (yyyymm):
date=
199901
199901
199901
199905
199905
200107
200107
200107
variable=
1
2
3
4
5
5
6
7
How can i get
variable_montlyav=
2
2
2
4.5
4.5
6
nan (or 6 i dont mind)
6
so that i can use my new montly averaged variable with some existing codes i have.
Stephen23 on 22 Feb 2016
Edited: Stephen23 on 24 Feb 2016
What exactly do you mean by a "date vector"? In MATLAB the term "date vector" has a specific meaning, which is a numeric vector consisting |y,m,d,H,M,S.F] values. If your data really was a date vector, it would looks like this:
1999 1 1 0 0 0
1999 1 1 0 0 0
1999 1 1 0 0 0
1999 5 1 0 0 0
1999 5 1 0 0 0
2001 7 1 0 0 0
2001 7 1 0 0 0
2001 7 1 0 0 0

Stephen23 on 22 Feb 2016
Edited: Stephen23 on 24 Feb 2016
Note: this code works if date is a cell array of strings, and also if date is a numeric vector of values representing year and month.
date = {'199901','199901','199901','199905','199905','200107','200107','200107'};
variable = [1,2,3,4,5,5,6,7];
[~,~,idx] = unique(date);
tmp = accumarray(idx(:), variable, [], @mean);
tmp(idx)
creates this output:
ans =
2.0000
2.0000
2.0000
4.5000
4.5000
6.0000
6.0000
6.0000
aaron Harvey on 23 Feb 2016
worked perfectly! i dunno how it works but it does :)

Roger Wohlwend on 22 Feb 2016
A = accumarray(date, variable, [], @mean);
MeanValues = A(date);
Stephen23 on 22 Feb 2016
This is not a very memory-efficient solution, as it will create temporary arrays with more than 200000 elements in them. See my answer for a more memory-efficient solution.