Use a different colormap scale for different sections of a matrix

I have a 15-dimensional data set and I would like to view the correlation matrix and the covariance matrix in 1 image. To do this I use the upper half triangle + the diagonal of a 15x15 matrix for the covariance data and the lower half triangle for the correlation data. I would like to apply a heatmap to this matrix, but have the covariance color mapping scale be independent of the correlation color map scale.

%Generate some data, and scale it so that covariance is distinct from
%correlation
exampleData = rand(100,15)*50;

exampleCov = cov(exampleData);
exampleCorr = corrcoef(exampleData);

%Creating and plotting the desired matrix

matrixToPlot = triu(exampleCov) + tril(exampleCorr,-1);
heatmap(matrixToPlot);

However I would like to be able to have two different scales for the colormap, one which is used for the lower correlation data and a different one for the covariance.

I considered plotting exampleCov and exampleCorr seperately, but this just overwrites the latter onto the former:

figure

heatmap(triu(exampleCov));

heatmap(tril(exampleCorr,-1));

%This just yields a plot of exampleCorr

This would be easy if I could set 0 values to NaN, and assign NaN color as 'none', but setting NaN color as 'none' is not supported.

[Adam Danz]

The hold on command with heatmap should give you an error indicating that the use of hold on with heatmap is not supported.

[Sean McKee]

Ah, so it does. I've edited that out.

Answer 1

[darova]

YOu can create your own color matrix

[x,y,z] = peaks(30);
ix = x.^2 + y.^2 < 2^2;
c1 = z*0;
c2 = z*0;
c1(ix) = 1;
c2(~ix) = 1;
C = cat(3,c1,c1*0,c2);
surf(x,y,z,C)

[Sean McKee]

Hmmm I think I see how I could finangle the color matrix to be what I want, but it appears that heat map does not accept a color matrix as input.

Answer 2

[Adam Danz]

> This would be easy if I could set 0 values to NaN, and assign NaN color as 'none', but setting NaN color as 'none' is not supported.

True, but you can set NaN colors to some other color. Setting a color to none with an object that accepts this option just makes the object invisible. For a grid of colors in heatmap, you can set the missing value tiles are any color you want using the MissingDataColor property.

1. Set 0s to NaNs

matrixToPlot(matrixToPlot==0) = NaN;

Or, for floating decimals,

matrixToPlot(abs(matrixToPlot)<max(eps(matrixToPlot(:)))) = NaN; 

2. Create heatmap

hm = heatmap(matrixToPlot);

3. Change NaN color

It can't be changed to none but you can change it to white or any other color.

hm.MissingDataColor = 'w';

[Sean McKee]

Sure I can change the NaN values to white or any other color, but this doesn't get me closer to using different scales for different sections of the matrix. I mentioned setting NaN to "none" because this would allow me to plot the second heatmap over the first, but as it stands this simply overwrites the first.

[Adam Danz]

You can't plot two heatmaps superimposed.

You can only change the values of the heatmap input matrix.

If you use imagesc, then you can superimpose two axes although I wouldbn't recommend doing that.

Answer 3

[Sean McKee]

I'm thinking the way to do this might be to generate the desired color-mapping by rescaling the covariance values appropriately, inputing into heatmap, but then overlaying the original values as text labels, as described on the heatmap examples page:

https://www.mathworks.com/matlabcentral/mlc-downloads/downloads/submissions/24253/versions/7/previews/html/heatmap_examples.html

However, I would still like to be able to display two different colorbars near the plot, and I'm not sure how to get that to work.

[Adam Danz]

I highly suggest you do this with imagesc instead of heatmap. Heatmap is quite difficult to customize.

Here's an answer that creates a "heatmap" using the heatmap function and using imagesc.

https://www.mathworks.com/matlabcentral/answers/722079-generate-a-geographical-heat-map#answer_602491