Hi, I am trying to use grpstats on my program and trying to compute median and percentile and skew but somehow i am not getting the program to run.

This is what I write

[  Z]= [WeekNum,C,E,G,I];
  [stat1,stat2]= grpstats(Z,WeekNum,{'mean',@(Z) prctile(Z,50)})

This is the error I am getting---

Error using grpstats>tryeval (line 429) Error computing statistics for group '10'.

Error in grpstats (line 304) z(gnum,:,:) = tryeval(hfun,x(idx,:),glabel{gnum},tsize)';

Error in portfoliocharacterization (line 24) [stat1,stat2]= grpstats(Z,WeekNum,{'mean',@(Z) prctile(Z,50)}) Caused by: Function '@(Z)prctile(Z,50)' returned a result of size [1 1], expected size [1 5]. ------------------------------------------------------------------------

My data has 5 columns,containing financial statistics and I am grouping them based on what week they were calculated.

I would appreciate any help.

Answer 1

[Peter Perkins]

Rishav, I suspect the problem is that your data has a weeknum for which there is only one row. You'll need to specify 1 for the DIM argument to PRCTILE in your anonymous function. GRPSTATS tries to save time by evaluating the functions on all columns or a group of rows all at once, but when it hits the group with one row, PRCTILE without DIM computes along that row, not down the columns, and returns a scalar rather than a 1x5.

All of these summary stat functions have a DIM argument to control just this kind of edge case, it's usually a good idea to use them if the input can be a matrix. In this case, it's probably not obvious that GRPSTATS is passing a matrix to your function, but it is.

BTW, there is a median function, so you could have just used @(x)median(x,1).

[Rishav]

Hey thanks for you help, the dim thing did it :)

Answer 2

[Wayne King]

Hi Rishav, WeekNum should be your grouping variable for the observations in Z so I'm not sure why you are including WeekNum in the matrix Z. For example, suppose I have the following temperature data with month as the grouping variable

     month = [1 1 1 1 6 6 6 6 9 9 9 9]';
     % then the temperature data
     temp_jan = [20 15 21 17]';
     temp_jun = [75 78 80 82]';
     temp_sept = [65 60 70 75]'; 
     temp_data = [temp_jan temp_jun temp_sept];
     temp_data = reshape(temp_data,12,1);
     [stat1,stat2]= grpstats(temp_data,month,{'mean',@(temp_data)  prctile(temp_data,50)});

[Rishav]

Thanks for your answer Wayne, it was really helpful.