cell indexing, extracting rectangular subset

I have been a MATLAB user for many years, and this is a question that has bugged me for a long time.

I am parsing a feed of text:

[tokens, matches] = regexp(result, template, 'tokens', 'match', 'dotexceptnewline');

Which produce a 700-element cell array, where each cell contains 4 subcells:

>>whos tokens
Name        Size              Bytes  Class    Attributes
  tokens      1x700            392000  cell    

>>tokens{1}
ans =
  1×4 cell array
    {'0'}    {'00'}    {'00.575807000'}    {'some_string'}

Now I want to grab the 700 rows and 3 columns of numerical data and push them into a function that returns a numerical array. If tokens had been a regular array, I would have done:

    tokens(:,1:3)    

Of course, that does not work here. So what would you do except writing loops that seems totally un-MATLAB-esque?

I usually try stuff like:

    tokens{:}(1:3)
    {tokens{:}(1:3)} 
    tokens{:,1:3}

Then I give up, reminding myself that cells are unintuitive and find something else to attack.

Answer 1

[Stephen23]

The simplest (and usually quite efficient) approach is to use a comma-separated list to create one cell array:

tmp = vertcat(tokens{:}); % -> 700x4 cell array
mat = str2double(tmp(:,1:3)) % -> 700x3 double

https://www.mathworks.com/help/matlab/matlab_prog/comma-separated-lists.html

https://www.mathworks.com/matlabcentral/answers/320713-how-to-operate-on-comma-separated-lists

https://www.mathworks.com/company/newsletters/articles/matlab-tips-and-tricks-exploiting-the-comma-separated-list-vectorizing-cell-array-and-structure-references.html

Most of your syntax attempts do not work for the reason give here:

https://www.mathworks.com/matlabcentral/answers/329538-why-do-i-get-this-error-for-this-non-scalar-structure

[dpb]

Yeah, and may be faster than cellfun...

>> str2double(vertcat(tokens{:}))
ans =
         0         0    0.5758       NaN
         0         0    0.5758       NaN
         0         0    0.5758       NaN
         0         0    0.5758       NaN
         0         0    0.5758       NaN
>> 

gets to the same place to just then pull off the first three columns. For some reason I can never remember vertcat when I want it... :(

Answer 2

[dpb]

I just duplicated your example 5 times...

tokens=tokens.';                % reshape to column cell array
vals=cell2mat(cellfun(@str2double,tokens,'UniformOutput',false));   % convert to numeric
vals=ans(:,1:3);                % return the known numeric columns only

The above returns

>> vals =
         0         0    0.5758
         0         0    0.5758
         0         0    0.5758
         0         0    0.5758
         0         0    0.5758
>> 

"Trick" is knowing that str2double returns NaN for non-convertible fields so first step results in an Nx4 double array with the last column being NaN.

Need to first organize by column vector, though, to get the output array in same shape as the input records.