perform matrix operation without for loop

I have a matrix (exp_data) of size 25000*45000. I need to perform the following operations as in the code, however with the for loop it takes ~4 hours to operate,

beta is a set of 3 values, for which I need to perform the operations. beta=[0.1, 0.04, 0.6]

any help is much appreciated

for m=1:3
for i=1:25000
  for j=1:45000
if exp_data(i,j)>1.5;
xes(i,j)=exp(beta(m).*(exp_data(i,j)));
else xes(i,j)=1;
end;
  end
end;
xe(:,:,m)=xes;
end;

Answer 1

[madhan ravi]

xe = exp(reshape(beta, 1, 1, []) .* exp_data) .* (exp_data > 1.5) + 1 * (exp_data <= 1.5);
% use bsxfun() for implicit expansion in older versions
xe = exp(bsxfun(@times, reshape(beta, 1, 1, []), exp_data)) .* (exp_data > 1.5) + 1 * (exp_data <= 1.5);

[Ameer Hamza]

Not sure if MATLAB realizes that when (exp_data > 1.5) is 0, it should avoid the corresponding calculations in exp(reshape(beta, 1, 1, []); otherwise, that is just a waste of computation. 🤔

Answer 2

[Ameer Hamza]

Following code is equivalent

xe = zeros([size(exp_data) numel(beta)]);
mask = exp_data > 1.5;
for m = 1:numel(beta)
    xes = ones(size(exp_data));
    xes(mask) = exp(beta(m).*(exp_data(mask)));
    xe(:,:,m) = xes;
end

Not sure if there are any speed improvements since JIT compiler is already claimed to be working very well.

[madhan ravi]

I belive your last line should be:

xe(:,:,m) = xes;

[Ameer Hamza]

Yes!! Thanks for correction.