Calculating correlations across multiple time series

I have two matrices with data consisting from 324 trials, and 102 channels as follows:

alphapower [324×102 double]

betapower [324×102 double]

I would like to calculate a matrix of correlation and beta (linear) across all channels. So as an output i need a matrix which is [102 x 102 double] which holds the correlations across all channels. The following code i believe achieves this:

% These two variables holds the original data
% which is here replaced by random numbers
alphapower = randn(324,102);
betapower = randn(324,102);

% Instantiation and loop across both dimensions - must be a better way of
% doing this
betas = zeros(102,102);
rsquare = zeros(102,102);
whichstats = {'beta','rsquare'};
for i=1:102
    for j=1:102
        stats = regstats(alphapower(:,i),betapower(:,j),'linear',whichstats);
        betas(i,j) = stats.beta(2);
        rsquare(i,j) = stats.rsquare;
    end
end

% Lets visualize our results
figure
heatmap(betas)
figure
heatmap(rsquare)

Does anyone know a better way to write this maybe using vectorization? The code above is surprisingly fast but i need to do this operation many times. I am also frustrated i havent found a nicer way of solving this.

Answer 1

[Sarvesh Kale]

Hi Jlil,

If you are trying to compute the correlation between columns of a matrix or between columns of two different matrices then you should take a look at corr and xcorr function inbuilt in MATLAB https://in.mathworks.com/help/stats/corr.html

following code should do the trick

x=randn(324,102); 
y=randn(324,102);
Rxy=corr(x,y); % replace the inputs with your matrices 
figure;
title('correlation Rxy')
heatmap(Rxy)   % show Rxy 

I hope this answers your query, please accept the answer if it does.

Thank you.

[Jlil]

Hi Sarvesh that got me on the right track - thanks!