Astroidal

[TroyGustke]

This pull request includes commits that add the sobol sampler and astroidal sampler. I would highly recommend looking over the algorithm for astroidal sampling. I understand that we have established the exponential factor to be the number of parameters, but this makes any system greater than 4 dimensions extremely centric concentrated. Also, I think that we should make the sobol choice the primary choice for multistart_initialPointsDistributionType since it is overall better than uniform sampling. This should be user friendly, but the option to be completely random still exists.

[TroyGustke]

The sobol and uniform gave very similar results. Sobol was changed to the default initial distribution rather than uniform.

[AdityaSavara]

All unit tests passed. For the astroidal and higher dimensions resulting in a smaller / tighter sampling: I think this is okay. One way to look at it is that the "95% prior probability volume" is simply a smaller portion of the volume in the higher dimensional space. Consider it like this: (a,b) each have 90% probability of being between 0.05 and 0.95. Now how likely is this point: (0.05,0.05). It is less than 5% likely. That was just going from 1 to 2. In higher dimensions, being near corners becomes less likely. So astroidal (as we've done it) will naturally follow that trend -- and should. If someone wants a more even sampling they should use sobol or uniform. That being said, I'm not against adding a variable that lets people make this tightening less extreme. There should be a continuum where uniform and astroidal are just two limits. I don't mind if we add a variable that allows people to push astroidal towards uniform (and even beyond). That solves our "what is opposite of astroidal?" problem. Just figure out how to make astroidal move towards the limit of it becoming the same as uniform, add that factor, and add a comment. We can discuss.