How do I use a 2D logical index matrix to change a subset of a 3D array?

I have data in a 11x100x11 array. I'm looking for outliers within the subsets along the third dimension, i.e. data(:,:,1), and want to replace them with NaNs. I use the following line:

cond = abs(data(:,:,1)) > threshold;

which produces a 11x100 matrix, cond. Now I'm trying to figure out how to appropriately use cond to index the values I want to change. If I use the following line:

data(cond,1) = NaN;

I get a 1100x100x11 array. How do I index with cond to cover two dimensions of my array?

Answer 1

[Star Strider]

I’m not exactly sure what you want. Consider a 3D array to be a cube with the third dimension as ‘slices’, or with 11 (11x100) matrices along the third dimension of the cube. So ‘data(:,:,1)’ is the first ‘page’, ‘data(:,:,2)’ the second ‘page’, and so forth through ‘page’ 11.

To illustrate:

data = randi(100, 5, 6, 5);             % Create Smaller ‘data’ Matrix
threshold = 75;
Q1_data = data(:,:,1)
cond = abs(data(:,:,1)) > threshold;
Q2_data(cond,1) = NaN;
Q2_data(~cond,1) = Q1_data(~cond);
Q2_data = reshape(Q2_data, size(data,1), size(data,2))

So you probably have to loop through all 11 ‘pages’ and do the same procedure on each one.

[Debra]

This works, but I would simplify it like this:

data = randi(100, 5, 6, 5);             % Create Smaller ‘data’ Matrix
threshold = 75;
Q1_data = data(:,:,1)
cond = abs(data(:,:,1)) > threshold;
Q1_data(cond) = NaN;
data(:,:,1)=Q1_data;

Answer 2

[James Tursa]

data(cond(:)) = nan; % Use linear indexing since target is 1st plane

[Debra]

That works fine for the 1st plane, but Star Strider is right that I want to look at all planes, which will each have their own set of conditions. I'm not sure how linear indexing would work for the subsequent planes.

[James Tursa]

My Answer was based on this line, data(cond,1) = NaN, where it looked to me like you only wanted to change the 1st plane. If you wanted to have different thresholds for each plane and apply them separately, then I would have suggested:

thresholds = 11-element threshold vector for each plane
cond = bsxfun(@gt,abs(x),reshape(thresholds,1,1,[]));
data(cond) = nan;

But, frankly, I am still not sure what you are after since I am not sure what you mean by "... each have their own set of conditions ..."

[Debra]

Having different thresholds for each plane is precisely what I meant by "...each have their own set of conditions." This looks interesting, I'll have to play with it.

Answer 3

[Sean de Wolski]

Why not just use a for loop over slices...?

[Debra]

I'm going to, but that doesn't address the indexing problem I was curious about. The variable cond is supposed to represent the first and second dimensions of the 3D array, but Matlab treats it as only the first dimension in the line:

data(cond,1) = NaN;

[Rogier Westerhoff]

I am struggling with the same. It also makes the whole loop quite slow if you have to do this over many pages. Hopefully Mathworks will find a solution for this soon?