Fix Dask version to avoid new error

• Fix Dask version to 2.30.0. See #144 and #145 for details.

Minor bug fixes

• Minor bug fix in Family.load() where no more than ten subtypes could be recovered.

Updated documentation, easier `kwargs`

• Separate kwargs parameters in edo.DataOptimiser for fitness, stopping, and
  dwindling.
• Updated documentation for v0.3.x.

Remove mutation of metadata parameters

• The effect of mutating metadata distribution parameters does not seem to
  have a massive effect on small examples and confuses the representation
  of individuals. Therefore, it is being removed.

Add `MANIFEST.in`

Add a MANIFEST.in file for requirements.txt and documentation files on PyPI.

Fitness functions must take an individual

• Add missing install requirements to setup.py.
• Fitness functions must take an instance of Individual.
• Fix flaky tests with long runtimes.

Introducing the DataOptimiser, Individual and Family classes

This is the first big improvement in a while and makes the library easier to use
in my opinion. The changes can be summarised as follows:

• The EA has been moved to the edo.DataOptimiser class rather than edo.run_algorithm.
• Individuals are now a fully-stocked class.
• A new class Family for handling the subtypes of a Distribution class
  externally. Note the move from edo.families to edo.distributions to
  reflect this and to avoid confusion in the future.
• The Individual class is more robust now, taking its fitness as an attribute
  and being able to store and recover itself entirely -- including the subtype
  instances in Individual.metadata (a big improvement over the
  dictionaries).
• The pseudo-random number generator framework has been totally decentralised.
  This means that stochastic fitness functions can use np.random.seed if
  necessary without throwing reproducibility out of the window. Individual and
  Family instances are provided their own np.random.RandomState instances to
  use in sampling, and there is a "lead" state used by DataOptimiser.
• Minor changes to README and other documentation files hosted on GitHub.

Addressing Dask aux files and PyYAML warning

v0.2.1 (2019-04-25)

• Ignore Dask aux files if they come up
• Fix warning from PyYAML

Subtypes, parallelisation, writing.

History

v0.2 (2019-04-15)

• Column distributions can now produce independent versions of themselves (#112)
• Fitness computation is parallelised and cached (#117)
• Results written to disk. Output metadata as dictionaries. (#118)

Resetting distributions and doc fixes.

v0.1 (2019-02-05)

• Resetting of columns is now based on the original parameter limits
• Minor fixes in documentation and larger code base