Clustering using Gower's Distance

Hello all, I have a dataset that includes both categorical and numerical features, and I'm looking to perform clustering on it. I've read that Gower's Distance (code is available) is suitable for handling mixed data types. However, I am getting an "isnan" error. How can I fix the problem? Thanks for the help.

DataSet = readtable("Test.xlsx", 'ReadVariableNames', true);
GowerDst = gower(DataSet);
[Idx, C] = kmedoids(DataSet, 2, 'Distance', GowerDst);

Error using isnan
Invalid data type. Argument must be numeric, char, or logical.
Error in kmedoids (line 220)
wasnan = any(isnan(X),2);
             ^^^^^^^^
Error in Gower_Distance (line 9)
[Idx, C] = kmedoids(DataSet, 2, 'Distance', GowerDst);

[the cyclist]

Can you upload the data, or a representative sample that illustrates the problem? You can use the paper clip icon in the INSERT section of the toolbar.

[MB]

The dataset is pretty big, so I only uploaded part of it.

Answer 1

[Torsten]

To use a distance that is not implemented, you have to define a function handle. Since I guess that GowerDst is not a function handle, MATLAB errors.

Look at the documentation for "kmedoids" for more details:

@distfun

Custom distance function handle. A distance function has the form

function D2 = distfun(ZI,ZJ)

% calculation of distance

...where

• ZI is a 1-by-n vector containing a single observation.
• ZJ is an m2-by-n matrix containing multiple observations. distfun must accept a matrix ZJ with an arbitrary number of observations.
• D2 is an m2-by-1 vector of distances, and D2(k) is the distance between observations ZI and ZJ(k,:).

If your data is not sparse, you can generally compute distance more quickly by using a built-in distance instead of a function handle.

[MB]

Looks like I have to modify the "Gower's Distance" function as well and that’s exactly what I was worried about.

[Torsten]

See Edward Barnard's answer here:

https://uk.mathworks.com/matlabcentral/answers/632969-how-to-performe-k-medoids-with-pre-calculated-distance-matrix

I suggest you test whether it's correct for implemented distances once by supplying the distance matrix as below, second by using the 'Distance','...' option and comparing the results.

Or take a look at

https://uk.mathworks.com/matlabcentral/fileexchange/28860-kmedioids

DataSet = readtable("Test.xlsx", 'ReadVariableNames', true);

Warning: Column headers from the file were modified to make them valid MATLAB identifiers before creating variable names for the table. The original column headers are saved in the VariableDescriptions property.
Set 'VariableNamingRule' to 'preserve' to use the original column headers as table variable names.

GowerDst = gower(DataSet);
K = 2;
N = 18;
[idx, C, sumd] = kmedoids((1:N)', K, 'Distance', @(ZI, ZJ) GowerDst(ZJ, ZI));

function D = gower(data)
[n, p] = size(data);
D = zeros(n, n);

for i = 1:p
    column = data{:, i};
    if isnumeric(column)
        range = max(column) - min(column);
        if range == 0
            continue;
        end
        d = abs(column - column') / range;
    elseif iscell(column) || iscategorical(column) || isobject(column)
        d = zeros(n, n);
        for j = 1:n
            for k = 1:n
                d(j,k) = ~isequal(column{j}, column{k});
            end
        end
    else
        warning('Skipping column %d: unsupported data type', i);
        continue;
    end

    D = D + d;
end

D = D / p;

end