[methods.partition DirectedSphereExclusion]

[FarnazH]

To do:

• [x] Add test to reach 100% coverage
• [x] Add comments in the code
• [x] Add formulas to docstring and polish docstring (add reference).

[FarnazH]

To be discussed in the next meeting:

• [ ] The algorithm never attempts to select exactly the requested size subset. Review when selected is returned.
• [ ] Add an argument so the user can specify whether they would like to optimize r, or just use r0 (guess) for selection.
• [x] random_seed argument is not used. Remove?
• [x] argument cluster_ids is not used! This needs to be fixed, but probably we need to change it across different methods.
• [x] The DirectedSphereExclusion is a distance-based method (similar to OptiSim); see the reference paper. So, it should be moved to methods.dissimilarity module (the name of this module should change to distance probably).

Based on the paper:

1. We are not addressing ties when sorting the samples based on their distance to the reference. In the paper, it is recommended to use the closest point to reference to break the tie.
2. This implementation only works for Minkowski p-norm measure, not any desired similarity measure.