Creating a loop to use all columns in a dataset array

Hello , i have a data set with 10 columns.The first column contains the names of the variables and the 9 next contain the data for the variables across 9 periods. I want , for each period ,to take the data which are lower than the data's median .I use the following command for the first period: s1=data(data.mv1<median(data.mv1),{'Name','mv1',}), where mv1 is the header of the first period's column and s1 a new dataset which contains only the variables i want. My question is , can i write a (for) loop that will automatically do this for the whole 9 periods, thus giving me s1,s2,...,s9?

[Jan]

There must be one or two typos in "s1=data(data.mv1<median(data.mv1),{'Name','mv1',})"

Please fix them by editing your question.

Answer 1

[Oleg Komarov]

data = dataset({('a':'z').','names'},{rand(26,2),'mv1','mv2'});

EDIT

% Retrieve column names/variables (from second onwards)

varnames = data.Properties.VarNames(2:end);
nV = numel(varnames);

% Preallocate

C = cell(1,nV);

% Loop per each column (variable)

for v = 1:nV
    idx = data.(varnames{v}) < median(data.(varnames{v}));
    C{v} = data.(varnames{v})(idx);
    C{v} = data(idx,[1 v]);
end

[Giorgos]

Thank you for your answer. A few brief questions: If i use C{v} i get a 'Cell contents assignment to a non-cell array object.' , whereas if use a new letter (for example s{v}) it works fine. Is there any way though for the s{v} to display the names of the variables along with the data? Also, your code creates one cell array which contains all the new data sets i wanted. How do i split them so i have different elements for each s{v}? Again , thank you for your time.

[Oleg Komarov]

C{v} gave error cause I forgot to change preallocation to cell. Now it's fixed.

I added another line in the loop but don't know if it does what you're asking. Choose the first or the new line.

Can you elaborate on "how do I split..."?

[Giorgos]

The line you added does display the names like i wanted but the variable n is not defined. I assumed you wanted to write 'v+1' insted of 'n' so i tried that and it works. The end result of the code is a cell array with dimensions '1*nV' meaning each cell of the first row (up to the nV cell) contains a dataset ,namely C{v}. My question is, how do i extract these datasets to the matlab workspace so i can use them as variables?

[Oleg Komarov]

Sry for the distraction, it should be v (why v+1?).

Why do you want it to be extracted to the workspace, I suggest to keep it in the cell array, easier to reference. Read http://matlab.wikia.com/wiki/FAQ#How_can_I_create_variables_A1.2C_A2.2C....2CA10_in_a_loop.3F

[Giorgos]

If i use v ,the first dataset doesn't show the first data but instead repeates the names of the variables. Agreed, it's much easier if i keep them in the cell ,it came to me right after i posted it unfortunately. Thanks again for all your help.

Answer 2

[Jan]

It would be easy, if you do not use symbols like 's1' and 'mv1', which have an index inside the name. Better use and index as index: s{1}, s{2}, ... and mv{1}, mv{2}, ...

See also: Matlab Wiki: create variables A1,A2,...

Answer 3

[Titus Edelhofer]

Hi,

I guess this leads to the question, how to get to data.mv1 where mv1 is given in a variable?

Note, that data.mv1 is the same as data.('mv1'). So if you have e.g.

header = {'mv1', ...};

then you could do

for i=1:length(header)
  col = data.(header{i});
  % do your median thing
  dataNew = data(col<median(col));
end

Hope, this helps,

Titus

[Giorgos]

Is your <header> a cell array? When trying to run your code i get a 'Dataset array subscripts must be two-dimensional.' error. Also , shouldn't there be a suscript to dataNew like dataNew{i}?

[Titus Edelhofer]

Yes, header is a cell array containing the names. And yes, there should be some subscript depending on what further you want to do with the reduced data ...