Using cellfun to remove nans from uniform cell

Hi Guys, I have a bunch of cell values continaing 17 unifrom sized matrices that have NaN values in them. I need to replace these NaNs with any large place-holder number. I am trying to use a solution that has shown up in other posts:

https://www.mathworks.com/matlabcentral/answers/440856-replacing-nans-with-zero-in-a-matrix-within-a-cell-array

https://www.mathworks.com/matlabcentral/answers/459133-how-to-replace-nan-values-with-any-other-values-in-a-cell-array

Here is my code

RvarNnan1 = cellfun(@(M) subsasgn(M, substruct('()', {isnan(M)}), 123000), RvarNnan1, 'UniformOutput', 123000);

I am getting the following errors, prior to running:

"output must be assigned to a variable"

and after running get :

"Error using cellfun

Non-scalar in Uniform output, at index 1, output 1.

Set 'UniformOutput' to false."

I think i don't understand the "M" variable. I've attached my cell variable(this is a sliced version {5x1}, of the actual variable {17x1} due to upload size restraits. Also of note, different from the other posts, I am replacing the NaNs with a meaningless large number. I have tried switiching around M with a predifenedvariable as well as using the actual VarNnan1.

Any help here is greatly appreciated!

[James Spalding]

Resolved it myself. I needed to change the final field to 'false' and it ran correctly.

here's what the code looks like now.

RvarNnan1 = cellfun(@(M) subsasgn(M, substruct('()', {isnan(M)}), 123000), RvarNnan1, 'UniformOutput', false);

Answer 1

[dpb]

If they are and always will be the same size, I'd use subterfuge...

[ra,ca]=size(R{1});             % save the array sizes first...
[r,c]=size(R);                  % and the cell array size
R=cell2mat(R);
R(inan(R))=0;
R=mat2cell(R,r*ones(ra,1),c*ones(ca,1));    % turn back to cell array

[dpb]

ADDENDUM: Same trick will work if they're not all the same number of rows but consistent columns, but not if both sizes change.

I'd have to dig into the bowels to figure out the one-liner for that case; I'd probably just use the straightforward loop; it would probably out perform the complex cellfun form besides.

ADDENDUM SECOND: It is possible (probable?) that one doesn't really need to do this at all but would be better off keeping the missing value indicators instead, and writing code to deal with it.

[James Spalding]

Thanks for the feedback!

And yeah I would prefer to keep the nans but the next step for me is using Interp2 which can't have NaN values, but i need to maintain the dimensions of the arrays for the cordinates to match to the cordinates variables. I get around wildly wrong interpolated values using the 'nearest' option.

[dpb]

Oh. In that case I'd look at fillmissing instead...you could use one of the interpolation methods there instead and have "ready for prime time" data array for interp2. Or, if it is just using interp2 to fill the missing rather than actually interpolating to a finer grid, you've saved a step entirely.

Athough fillmissing only operates over one dimension; it can interpolate either by row or by column (default) but, won't/can't do a 2D interpolation over the surrounding values.