Iterating in a dataset encompassing a structure

Hello all,

I have an issue accessing an array in a dataset table which emcompasses a struct in it.

I want to build an array which contains eigenvectors from different rank during the Johansen test.

I have the following :

basevar = evalin('base', x)
[c,n] = size(basevar)
arrcoef = zeros(n)
[h,pvalue,stat,cValue,mles] = jcitest(basevar,'model','H1', 'lags', 1)

Here i get, a dataset called mles.

Inside this dataset I have :

mles = 

            r0              r1          
      t1    [1x1 struct]    [1x1 struct]

and when i do mles.r0 I have

   paramNames: {5x1 cell}
       paramVals: [1x1 struct]
             res: [1498x2 double]
          EstCov: [2x2 double]
          eigVal: 0.0106
          eigVec: [2x1 double]
             rLL: -1.2002e+03
             uLL: -1.1902e+03

Here I want to get the eigVec array.

So i open a loop :

for i = 1:n
  coeff = double(pvalue(:,i))
  if 0.05 - coeff && 0.05 - coeff > 0
  arcoeff(i) = mles.r(i).eigVec
  end
  end

I can see the issue it comes from this part : mles.r(i), it takes r as a variable which is normal, I want to know how to make this .r0 reactive to an iteration. I tried with {} and putting the name in something like this

    a = mles.Properties.VarNames{i}
    mles.a.eigVec

But this does not work.

Do you know how can make this reactive to my iterative value i?

thanks

D

Answer 1

[Matt Tearle]

If I understand the problem correctly, you're getting a dataset with variables r0, r1, foo, potato, etc, each one of which contains a struct, and you want to extract the eigVal field from each variable. But the problem is that you can't find a way to loop over the variable names programmatically.

Assuming that's the issue...

Firstly, if you can use 13b or later, you can use tables rather than datasets, which are a bit nicer and allow either named or numeric indexing with {}:

for i = 1:n
   arcoeff(i) = mles{:,i}.eigVec;
end

But, either way, you can use dynamic variable referencing in the same way you can with structs:

x = dataset(rand(5,1),rand(5,1))
y = 'Var2'
z = x.(y)  % Equivalent to z = x.Var2

So now

vnames = mles.Properties.VarNames;
for i = 1:n
   arcoeff(i) = mles.(vnames{i}).eigVec;
end

[Side note: given that the eigenvectors are, well, vectors, I think your indexing will have to be something like arcoeff(:,i) = ...]

[Second side note: the condition if 0.05 - coeff && 0.05 - coeff > 0 isn't well formed: the first part 0.05 - coeff is numeric, so will be converted to logical, which will make it equivalent to 0.05 - coeff > 0 (so you're testing the same thing twice -- "if X && X").]

[Davin]

Thanks Matt for yr answer. You are also right on the last comment.

cheers

Answer 2

[Guillaume]

Use dynamic field names for accessing your rank variables:

arcoeff{i} = mles.(sprintf('r%d', i)).eigVec;