Use arrayfun and cellfun to avoid for-loops

Hey there,

I am currently getting used to all the functions provided by MATLAB, so I am unfortunately not entirely sure how to fully utilize cellfun and arrayfun.

I was wondering, whether there are faster workarounds for for-loops, e.g.

    for i= 1:length(Categories)-1
        for j= i+1:length(Categories)
            % separated_data is a 3D array of data, I want to get the p-Values for the data matching
            % several numerated categories here
            p=vertcat(p,[i,j,ranksum(separated_data(:,1,i),separated_data(:,1,j)),...
                ranksum(separated_data(:,2,i),separated_data(:,2,j)),ranksum(separated_data(:,3,i),separated_data(:,3,j))]);
        end
    end

Another one would be a routine I am trying to write to more conveniently import my data (I know there are a lot of workarounds, but I am writing my own for practice and to meet the special formatting of my data). Here's my plan:

• Get the file name (done)
• Obtaining the header titles and enumerating them, then ask the user to give a vector with the columns they want to import (done)
• Create a cell array by repmat({'%*f'},1,max(input_vector)) (done)
• Now I want to remove the asterisks for all the elements given in the input_vector

I have tried using cellfun and arrayfun, but once again I am not familiar enough with them to get it to work without a for-loop. Therefore, here is my current code for that issue:

        columns = input('Enter which columns you''d like to import as a vector:\n');
        format = horzcat(repmat({'%*f'},1,max(columns)),{'%*[^\n]'});
        for i = 1:length(columns)
            format{i} = erase(format{i},'*');
        end

I'd be grateful for any advice related to the given problems, but also if you know a more detailed description of cellfun, arrayfun and these things (helping to optimize resources, runtime, etc.) than provided by

        doc cellfun
        doc arrayfun

Thanks in advance!

Answer 1

[Walter Roberson]

    for i= 1:length(Categories)-1
        for j= i+1:length(Categories)
            % separated_data is a 3D array of data, I want to get the p-Values for the data matching
            % several numerated categories here
            p=vertcat(p,[i,j,ranksum(separated_data(:,1,i),separated_data(:,1,j)),...
                ranksum(separated_data(:,2,i),separated_data(:,2,j)),ranksum(separated_data(:,3,i),separated_data(:,3,j))]);
        end
    end

That code has to expand p every time through the loop, which is inefficient.

    nC = length(Categories);
    pto = cell(nC-1,1);
    for i= 1:nC-1
        nleft = nC - i;
        pt = cell(nleft, 1);
        for j= i+1:length(Categories)
            % separated_data is a 3D array of data, I want to get the p-Values for the data matching
            % several numerated categories here
            pt{j-i} = [i,j,ranksum(separated_data(:,1,i),separated_data(:,1,j)),...
                ranksum(separated_data(:,2,i),separated_data(:,2,j)),ranksum(separated_data(:,3,i),separated_data(:,3,j))];
        end
        pto{i} = vertcat(pt{:});
    end
    p = vertcat(pto{:});

The above does not need to expand any arrays.

[Walter Roberson]

        columns = input('Enter which columns you''d like to import as a vector:\n');
        clear fmt
        fmt(1:max(columns)) = {'%*f'};
        fmt(columns) = {'%f'};
        fmt{end+1} = '%*[^\n]';
        fmt = horzcat(fmt{:});

I changed variable names to avoid conflicting with the format command.

Answer 2

[Jannis Korn]

Thanks for your answer! I rarely think about predefining the size of my elements, that is indeed a very helpful remark. Also I wasn't aware that such things as

fmt(columns) = {'%f'};

work, so once again thanks for that.

Do you think it would be faster to actually prelocate the entire size of p and then just fill in the values or would the additional code that I'd need to find the line to write in just slow the process?

[Walter Roberson]

    nC = length(Categories);
    nP = nC * (nC-1) / 2;
    p = zeros(nP, 5);
    pidx = 0;
    for i= 1:nC-1
        sdi = separated_data(:,1:3,i);
        for j= i+1:length(Categories)
            % separated_data is a 3D array of data, I want to get the p-Values for the data matching
            % several numerated categories here
            pidx = pidx + 1;
            sdj = separated_data(:,1:3,j);
            p(pidx,:) = [i, j, ranksum(sdi(:,1),sdj(:,1)), ...
                         ranksum(sdi(:,2),sdj(:,2)), ranksum(sdi(:,3),sdj(:,3))];
        end
    end