The read_out_to_table function uses a field that appears to not exist on the Linux version of the 'dir' function in Matlab. Specifically, if a = dir(output_directory), then a.folder does not exist. Need to workaround this issue somehow.

Because ga_interim.mat is just checked for existence in ga_output.m, there is the potential for a carry-over effect between runs of the GA. Solution: delete ga_interim between each call of the ga function.

The proposal distribution of the M-H algorithm is probably incorrect as-is. I suggest changing this line in mh_generate_pps:

q = max(p_bnds(:,1),min(p_bnds(:,2),p + (p_bnds(:,2)-p_bnds(:,1))/10.*(2*rand(num_p,1)-1)));

to:

for ii = 1:numel(q0)
    q(ii) = icdf(truncate(makedist('Normal', 'mu', p(ii), 'sigma', sigma_proposal), 0, 1), rand());
end

where sigma_proposal is some pre-defined width of the proposal distribution and we assume every parameter runs 0 --> 1 (LB & UB stored somewhere else).

Hard-wiring of directory structures with "/" probably breaks the code on Windows machines. The fix is probably straightforward using the "isunix" function built into Matlab to switch the directory structuring from "/" --> "".

Check GoF calculations for plot_gof_vp_vs_pp versus plot_gof_vs_npps. The numbers should be same to machine precision. Currently, very close but definitely different enough to worry about a bug.