Nested Sampling posterior sampler

[williamjameshandley]

This is a work-in-progress PR, so is not going to pass tests necessarily.

@bstienen : I would like to check that I have interfaced this correctly before
going on to interface multinest and pybambi.

My main queries are

• At the moment, polychord's temporary files are being stored in a directory
  .polychord_chains. Ideally this would be in a
  hds/<function_name>/polychord/ directory instead. What is the best way to
  get access to that from either a hds.Procedure.__call__, or
  hds.Procedure.__init__
• I'm not sure that I've implemented the reset and is_finished functions
• I have no idea what check_testfunction is supposed to do.
• I updated the hds.functions.Block function so that it has a non-infinite
  range. Will this interfere with anything else?

[bstienen]

Sorry, I misread your comment and thought I had to do nothing with it. Just reread it and came to the conclusion that I was mistaken. Anyway:

• The output folder is created by the Logger object in the logger property of an Experiment object. Procedure objects by design don't make assumptions about the experiment they are in, so they have no access to the Experiment object. I can change this behaviour, but I am not sure if the benefits of this change would outweigh the cost of the time we would have to put in this and the possible bugs it might create. If this is an important feature to you, I will make this change, but I would love your input on this.
• The methods seem to be implemented correctly :-)
• check_testfunction takes a single argument (a TestFunction instance) and returns a boolean. This boolean should reflect if the procedure can work with the supplied test function.
• Changing the Block test function to a bounded function is not a problem, when doing so this has to be changed in the FunctionFeeder as well (specifically in the load_function_group method). Correcting this solves the errors raised in Travis at the moment.

An important thing to note is that in a recent change the manual was moved to the Github wiki, so that we don't need pull requests to update instructions.