How to make this text string column program faster?

Question

Leon 2020-3-1

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/508303-how-to-make-this-text-string-column-program-faster

评论： Leon 2020-3-5

Accessions here are text strings with a length between 3 and 8. Here is my current program to process them:

% Accession:
ACCN = [];
c = length(Sta{i}.depth); % 1-column data
if isfield(Sta{i}, 'accession') 
    [ACCN{1:c}] = deal(Sta{i}.accession);
else
    [ACCN{1:c}] = deal('N/A');
end
ACCESSION = [ACCESSION; ACCN'];

I'd want to convert the above program to a more effective way like the below program for processing numerical values:

Latitude = NaN(1000000, 1);
lastIndex = 0;
c = length(Sta{i}.depth); % 1-column data
if isfield(Sta{i}, 'lat')
    Latitude([lastIndex+1:lastIndex+n],1) = ones(c,1) * Sta{i}.lat;
end
Latitude   = Latitude([1:lastIndex],1);

How would I do this?

Thanks.

3 个评论
显示 1更早的评论隐藏 1更早的评论

Leon 2020-3-2

test2.mat

Here is an example data set.

Many thanks!

dpb 2020-3-2

So what are we to make of that? What's the intended output? Every struct has the field so why the test?

Are you just trying to return the values of the field with the indices of the locations that don't have value for the field? Or is the field actually missing in the real dataset? If that's the case, looks like you should fix that instead and initialize the struct with a missing value instead and avoid the whole issue.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

dpb 2020-3-2

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/508303-how-to-make-this-text-string-column-program-faster#answer_418135

在 MATLAB Online 中打开

Dunno how much better; still think to build the struct so all fields are extant would be far better; then you can use an array of struct as well instead of having to dereference the cell array. Remove as many levels of indirection as possible.

for i=1:numel(A)
  d=numel(A{i}.depth);
  try
    ACCN=repmat({A{i}.accension},d,1);
  catch
    ACCN=repmat({'NA'},d,1);
  end
end

If the size is large, the biggest overhead hit yet in both is the dynamic reallocation -- precompute the output size and store into the output array instead.