Helper Class for Sampling Parameters

[ScepticalRabbit]

Create a helper class called ParameterGenerator - Abstract Base Class - has a 'generate()' method
Make sure this works using the moosherd class and the sweepreader classes for the case where we have MOOSE only or gmsh+MOOSE.
Class returns a list of dictionaries moose_vars for each runner object that can be passed to moosherd.run_para(moose_vars) - see example 5

Allow the user to automatically generate lists of parameter dictionaries to sample a given parameter space (concrete implementations):

• SENSITIVITY ANALYSIS: Perturb a given set of parameters by +/- X% and all combinations of them to do sensitivity analysis. User specifies default value for each parameter and the % perturbation.
• MONTE CARLO: Sample from user specified numpy probability disbtributions e.g. normal distribution specifying mean+standard deviation, user will then ask for N samples from the distribution
• LATIN HYPERCUBES / SURROGATE: User specifies parameters and then class generates N hypercubes.
• MESH REFINEMENT STUDY: User specifies base mesh parameters and is returns a list of dictionaries with double or half the mesh size.

Provide examples for each of the above cases.
Provide an example of a GP surrogate build in scikit learn and pyapprox using the latin hypercube sampler above.