How to turn NaN values in only numerical columns into -999?

I have some data with both numerical and string columns. See attached for an example (aa.xlsx).

It has four columns like this:

Header1  Header2  Header3  Header4 
  1,       4,       -9,    ABC123 
  2,      NaN,       0,    NaN
  5,       6,       NaN,   ABC789 

My goal is to convert any NaN values that are in only numerical columns into -999, while leaving the NaN values in string columns intact. The end results should look like something like this:

Header1  Header2  Header3  Header4 
  1,       4,       -9,    ABC123 
  2,      -999,      0,    NaN
  5,       6,      -999,   ABC789 

Here is the code I know will work, if all of my columns are numerical:

%convert any NaN into -999
T1 = readtable ('aa.xlsx', 'PreserveVariableNames',true)
Ind_table = isnan(T1{:,:}); 
T1{:,:}(Ind_table) = -999;

How should I modify it so that it won't do the conversion for columns that are made up of strings?

Many thanks!

Answer 1

[Adam Danz]

When you create your table, the missing values in the Header4 column will not appear as NaNs since that column contains character arrays. Instead, they will just be an empty char array. A very annoying thing with tables is that they do not support subscript indexing. So the solution below converts the table to a cell array, replace the NaN values in numeric columns, and then puts the cell array back into a table with matching properties as your original table.

T1 = readtable ('aa.xlsx', 'PreserveVariableNames',true);
T1cell = table2cell(T1); 
isnum = varfun(@isnumeric,T1,'output','uniform'); % ID columns that are numeric
ismiss = ismissing(T1);   % find missing values
T1cell(ismiss & isnum) = {-999}; 
T1New = cell2table(T1cell); 
T1New.Properties = T1.Properties; % your new table with NaN replacement

Result

T1New =
  3×4 table
    Header1    Header2    Header3     Header4  
    _______    _______    _______    __________
       1           4         -9      {'ABC123'}
       2        -999          0      {0×0 char}
       5           6       -999      {'ABC789'}

[Leon]

The code works flawlessly.

Thank you so much! This is amazing.

[Adam Danz]

Glad I could help!

Answer 2

[Sebastian Bomberg]

You can have fillmissing apply only to the numeric variables directly:

fillmissing(T1,"constant",-999,"DataVariables",@isnumeric)

[Leon]

Very cool and it works!

Thank you very much!

Answer 3

[Walter Roberson]

fillmissing(T1,'constant',{-999,-999,-999,'NaN'})

Note that this will use the character vector 'NaN' (three characters) in place of the numeric NaN entries in column 4, as it is not possible to have numeric entries in a column devoted to character vectors.

[Leon]

Many thanks! This is a much simpler way of dealing with this specific example!

The only problem is that in the real world, I have no idea which column is numeric and which one is a string. I won't even know how many columns the file will have.

[Walter Roberson]

char_vars = varfun(@iscell, T1, 'OutputFormat', 'uniform');
constants = num2cell(-999*ones(1,width(T1)));
constants(char_vars) = {'NaN'};
new_T1 = fillmissing(T1, 'constant', constants);

[Leon]

Mind blowing!

I love this solution even better, because it also allows me to fill empty cells within string columns with my preferred string 'N/A'.

Thank you so much! So sorry for unable to accept another answer.

[Adam Danz]

You can always unaccept / accept; I don't mind :)

[Leon]

Weird that the code works for some of my files but not for all of them. I now get an error.

Here is the code I use:

% replacing any NaN values with -999 for numerical columns, and
% any NaN values with 'N/A' for string columns.
char_vars = varfun(@iscell, T1, 'OutputFormat', 'uniform'); % identify the columns that are not numerical (strings)
constants = num2cell(-999*ones(1,width(T1)));               % create a one-row matrix of -999
constants(char_vars) = {'N/A'};                             % replace -999 with 'NaN' for string columns
new_T1 = fillmissing(T1, 'constant', constants);            % fill the empty cells within each respective columns
T1 = new_T1;

Here is the error:

Error using fillmissing/checkArrayType (line 548)
Invalid fill constant type.

Error in fillmissing/fillTableVar (line 182)
[intConstVj,extMethodVj] = checkArrayType(Avj,intMethod,intConstVj,extMethodVj,x,true);

Error in fillmissing/fillTable (line 160)
        B.(vj) = fillTableVar(indVj,A.(vj),intMethod,intConst,extMethod,x,useJthFillConstant,useJthExtrapConstant,mavj); 

Error in fillmissing (line 138)
        B = fillTable(A,intM,intConstOrWinSize,extM,x,dataVars,ma);

Error in FunctionLoading (line 32)
new_T1 = fillmissing(T1, 'constant', constants);            % fill the empty cells within each respective columns