How do I remove all fields of a structure that have at least one NaN?

I have a structure S with n fields, field1, field2 etc.. How do I remove those fields that have at least one NaN?

Thank you

Answer 1

[Walter Roberson]

The question does not permit us to assume that the struct fields are exclusively numeric or that the fields will not be compound data types, or that the initial structure is a scalar structure. We must assume that it might be a non-scalar structure, containing nested data types, and that for any given field at the top level, if there is a nan anywhere inside the object in any of the array dimensions, that the field must be removed from all struct array locations.

Taking this into consideration, you need a real function, and it is probably easiest to make the function recursive.

Under some restricted conditions, this code could be simplified.

function S = structrmnan(S)
    if isstruct(S)
        Sfields = fieldnames(S);
        tc = struct2cell(S);
        cn = cellfun(@containsNan, tc);
        mask = any(cn, 2:ndims(cn));
        tc(any(cn, 2:ndims(cn)),:) = [];
        S = reshape(cell2struct(tc, Sfields(~mask), 1), size(S));
    end
end

function tf = containsNan(V)
    if isnumeric(V)
        tf = nnz(isnan(V))>0;
    elseif ischar(V) || islogical(V)
        tf = false;
    elseif ~isscalar(V)
        tf = any(arrayfun(@containsNan, V(:)));
    elseif iscell(V)
        tf = any(cellfun(@containsNan, V(:)));
    elseif isstruct(V)
        tf = any(structfun(@containsNan, V(:)));
    else
        %for example there could be a NaN inside UserData of an array
        %of line objects
        error('Needs to be extended to handle datatype "%s"', class(V))
    end
end

[Giovanni Barbarossa]

Thank you all. I have attached the structure. The previous suggestions do not seem to work. I will try Mr. Robenson suggestion now. I think the datetime field is complicating things.

[Walter Roberson]

Extended to handle datetime, duration, calendarDuration, categorical, string, table

(I do not promise that all cases with table are handled; for example it would not surprise me if this version failed for a table that contained a nested table.)

function S = structrmnan(S)
    if isstruct(S)
        Sfields = fieldnames(S);
        tc = struct2cell(S);
        cn = cellfun(@containsNan, tc);
        mask = any(cn, 2:ndims(cn));
        tc(mask,:) = [];
        S = reshape(cell2struct(tc, Sfields(~mask), 1), size(S));
    end
end

function tf = containsNan(V)

    if ischar(V) || islogical(V)
        tf = false;
    elseif isnumeric(V) || ...
        isdatetime(V) || isduration(V) || iscalendarduration(V) || ...
        iscategorical(V) || isstring(V) || istable(V) 
        tf = nnz(ismissing(V))>0;
    elseif ~isscalar(V)
        tf = any(arrayfun(@containsNan, V(:)));
    elseif iscell(V)
        tf = any(cellfun(@containsNan, V(:)));
    elseif isstruct(V)
        tf = any(structfun(@containsNan, V(:)));
    else
        error('Needs to be extended to handle datatype "%s"', class(V))
    end
end

[Giovanni Barbarossa]

Thank you very much. It works perfectly.

Answer 2

[darova]

Try rmfield

a = [1 nan 3];
bb = [1 2 3];
S.a = a;
S.bb = bb;
S
nms = fieldnames(S);
for i = 1:length(nms)
    f = getfield(S,nms{i});
    if isnan(sum(f))
        S = rmfield(S,nms{i});
    end
end
S

[James Tursa]

To generalize this, you can loop over the result of the fieldnames( ) function.

[darova]

Agree. Changed that

[Giovanni Barbarossa]

Because I want to remove all fields that have at least one NaN, I would have expected a simpler solution. I thought about using something like rmmissing, which I know does not work for structures. Also something like a combination of isnan and any.

Anyway, it does not work because there is an error using sum: Invalid data type. First argument must be numeric or logical.

Thanks

[darova]

• I would have expected a simpler solution

Im a simple man

• Also something like a combination of isnan and any.

As you wish

if any(isnan(f))

[James Tursa]

@Giovanni: For the invalid variable type, you can test for numeric. E.g.,

if isnumeric(f) && any(isnan(f(:)))

Answer 3

[Image Analyst]

If you want, you can try to use structfun() but it's rather cryptic. darova's solution is much more intuitive and readable.

By the way, I'd have used any() instead of sum(), k instead of i (since i is the imaginary variable), and dynamic field names:

S.a = [1, nan, 3];
S.bb = [1, 2, 3];
S
nms = fieldnames(S);
for k = 1:length(nms)
	thisField = S.(nms{k});
	if any(isnan(thisField))
		S = rmfield(S, nms{k});
	end
end
S

[darova]

thanks bro, appreciate your support

[James Tursa]

Including the the type check here as well:

if isnumeric(thisField) && any(isnan(thisField(:)))