@mpf -- I discussed with @tinatorabi the implementation of a second test set where the solution is not sparse but has decaying coefficients, e.g., something like
A = randn(N, M)
x = permute( randn(M) ./ (1:M).^2 )
b = A * x
and then to test whether the first so-and-so-many largest coefficients are correctly recovered.
This is the use-case I am most interested in. Do you think this is feasible? Would you suggest any changes or refinements to this protocol?
@mpf -- I discussed with @tinatorabi the implementation of a second test set where the solution is not sparse but has decaying coefficients, e.g., something like
and then to test whether the first so-and-so-many largest coefficients are correctly recovered.
This is the use-case I am most interested in. Do you think this is feasible? Would you suggest any changes or refinements to this protocol?