# how to remove from all fields in the same position of a struct?

1 view (last 30 days)
Christos on 1 Feb 2013
I am sorry for opening a new question, but i cant understand the other solutions here. i have a struct named training_set of size 1x15 with two fields. one is called model and the other samples. samples is 1x3003 cell for each model. Each of these cells is a matrix i.e x. I want to do the following. for each model, i check one by one the 3003 X.( 15x3003 checkings ). I start from training_set(1).samples. if for some of the 3003 X's, abs(det(X'X))<1e-10 then i want to replace these cells with empty cell, but for all the 15 samples. Then for training_set(2) and replace with empty in all other sets. In other words, if training_set(1).samples{5}=[] then so should training_samples(2){5},.., training_set(15).samples{5}. I ve tried doing this with the following code, but it returns the whole training_set.samples to by empty ( i know that the almost the half should remain). Does anyone know why? I am sorry for the long question, i hope i gave you the point of what i am trying. Thank you guys.
for i=1:1:15
for j = 1:1:3003
x = training_set(i).samples{j};
if abs(det(x'*x))<1e-10
for k=1:1:15
training_set(k).samples{j}=[];
end
end
end
end

Jonathan Epperl on 1 Feb 2013
What you did should work, my only idea about a possible error: are your samples "fat" or "tall"? If they are fat, i.e. n columns, m rows, n>> m, then x'*x results in a ( n x n) matrix, which will have maximally rank m, thus be singular and have zero determinant. Long story short: det(x'*x) ~= det(x*x').
Also, this is a little more compact than what you are doing, and maybe it helps you understand other solutions:
% some sample data
training_set(1).samples = {1,2,eps};
training_set(2).samples = {1,eps,3};
training_set(3).samples = {1,1,3};
i0 = cellfun(@(c) det(c.'*c)<1e-10, cat(1,training_set(:).samples) )
for i=1:numel(training_set)
training_set(i).samples(any(i0,1)) = ...
{[]};
% ^^ Curly braces here if you want the cell element to be set to []. If you want to remove the cell element, then no curly braces
end
training_set.samples
Christos on 2 Feb 2013
Thank you Jonathan Epperl. All samples are of size n-by-m with m<=n. I ll see if i can do anything.