Adds the combinatorial optimisation feature as an optional dependency

[willu47]

Includes:

• corrected bug Error in compile_output if maximum_combo is not in ascending order #49 and Morris groups with optimal trajectories not correctly computing distance matrix #48 and refactored morris_optimal code to integrate combinatorial optimisation (Optimised trajectories #17) and brute force methods
• new tests for combinatorial optimisation are skipped unless the gurobipy module is present
• amended setup.py to include gurobipy as extra_requirement
• few other small admin changes

NB, supersedes #50 which should be closed without merging

[coveralls]

Coverage decreased (-5.75%) to 87.24% when pulling dc7d53b on willu47:feat_gurobi into 3362b90 on jdherman:master.

[coveralls]

Coverage decreased (-5.75%) to 87.24% when pulling dc7d53b on willu47:feat_gurobi into 3362b90 on jdherman:master.

[jdherman]

Really nice!! So if I understand this right:

• The option for optimization doesn't affect the current (non-optimized) Morris API
• gurobipy will not be installed as a dependency automatically, for example when running pip install SALib. People using it will need to install it separately.
• Travis CI will skip the Gurobi tests

Is this right? Again this is great, it's nice to have the option for optimized trajectories.

[willu47]

Yep, that's all correct. Unfortunately, gurobi has to be installed manually. If use is for academic research, it's easy to get a free academic license.

[jdherman]

That's fine, it's better than having no optimizer at all, and it's clearly optional. Plus this addresses some of the long-standing issues, so thanks again!!