Replace a missing string in a table

I want to replace all missing strings in a table with a string of my choice, say 'unknown'. I use R2016a (without an upgrade option), so functions like fillmissing are not available to me, in case they could be of help. Eg:

 dblVar = [NaN; 3; 7; 9];
 cellstrVar = {'one'; 'three'; ''; 'nine'};
 categoryVar = categorical({''; 'red'; 'yellow'; 'blue'});
 A = table(dblVar, cellstrVar, categoryVar)
A = 
    dblVar    cellstrVar    categoryVar
    ______    __________    ___________

    NaN       'one'         <undefined>
      3       'three'       red        
      7       ''            yellow     
      9       'nine'        blue

I would like to end up with this:

 A = 
    dblVar    cellstrVar    categoryVar
    ______    __________    ___________

    NaN       'one'         unknown
      3       'three'       red        
      7       'unknown'     yellow     
      9       'nine'        blue

Note I also replaced the categorical '<undefined>' as well, if you can please include in your answer.

Is there a way to do this without changing A's structure, eg from table to cell, in the process? The reason I want to avoid the transformation is my table is large, and transformation may cause memory issues.

Edit to add: the location of the missing string value has to be identified as well, there may be several such columns in the table.

Many thanks.

Answer 1

[George]

This should work:

dblVar = [NaN; 3; 7; 9];
cellstrVar = {'one'; 'three'; ''; 'nine'};
categoryVar = categorical({''; 'red'; 'yellow'; 'blue'});
cellstrVar2 = {'four'; 'none'; '7'; ''};

A = table(dblVar, cellstrVar, categoryVar, cellstrVar2);
varNames = A.Properties.VariableNames;

for ii = 1:numel(varNames)

     if iscellstr(A{1,varNames{ii}})
         undefloc = strcmp(A.(ii), '');
         A{undefloc, ii} = cellstr('unknown');
     end

     if iscategorical(A{1, varNames{ii}})
         undefloc = isundefined(A{:,ii});
         A{undefloc, ii} = categorical(cellstr('unknown'));
     end

end

A = 

    dblVar    cellstrVar    categoryVar    cellstrVar2
    ______    __________    ___________    ___________

    NaN       'one'         unknown        'four'     
      3       'three'       red            'none'     
      7       'unknown'     yellow         '7'        
      9       'nine'        blue           'unknown'

You can use the undefloc variable to find where things were undefined or empty strings.

[Ory]

Thanks George. I was hoping to get something that does not include for loops. It does not seem to be possible in Matlab.

[George]

If you know what categories are going to be cell / categorical arrays you don't need to use the loop.

dblVar = [NaN; 3; 7; 9];
cellstrVar = {'one'; 'three'; ''; 'nine'};
categoryVar = categorical({''; 'red'; 'yellow'; 'blue'});
cellstrVar2 = {'four'; 'none'; '7'; ''};
A = table(dblVar, cellstrVar, categoryVar, cellstrVar2);
undefloc = strcmp(A(:, {cellstrVar, cellstrVar2}), '');
A{undefloc, {cellstrVar, cellstrVar2}} = cellstr('unknown');
undefloc = isundefined(A{:,categoryVar});
A{undefloc, categoryVar} = categorical(cellstr('unknown'));

[Ory]

Awesome! If not known, one can get to the classes index by, say for cell strings:

 catInd = strcmp(varfun(@class, A, 'OutputFormat', 'cell'), 'cell');

[George]

There you go. If you do that and slam the cells and categoricals into cell arrays you can use the curly brace syntax on the cell array, rather than the variable name.

Answer 2

[Peter Perkins]

George's loop seems fine to me although you could tweak it a bit as

for name = varNames
    var = A.name;
    if iscellstr(var)
       var(strcmp(var,'')) = {'unknown'};
    elseif iscategorical(A.(name))
       var(isundefined(var) = 'unknown';
    end
    A.Name = var;
end

If you're willing to write a couple of small functions, you can do this:

theCellStrs = varfun(@iscellstr,A);
A(:,theCellStrs) = varfun(@replaceEmptyString,A(:,theCellStrs));

function c = replaceEmptyString(c)
c(strcmp(c,'') = {'Unknown'};

(and similarly for categorical) but varfun uses a loop underneath.