Dimod

New Features
------------
• `assert_consistent_bqm` to `dimod.testing` for testing different BQM implementations
• Testing is now done with parameterized package - this does not affect installed packages
• FileView version 2.0 with improved docs

New Features
------------
* `pack_samples` kwarg for `SampleSet.to_serializable()` method that allows the user to _not_ pack samples. This means bumping the sample set serialization version to 3.1.0

Fixes
-----
* `SampleSet.info` is now correctly propagated through the `ClipComposite`
* All `.pyx` and `.pxd` files are included in the MANIFEST which means that `cimport dimod` will now work correctly

New Features
------------
* BQM C++ headers are now in `dimod/include/dimod/`. Packages that need access to the headers can get the path using the new `get_dimod_headers()` function. Note that the C++ code is still considered unstable and does not respect semantic versioning.
* `Initialized` ABC and accompanying `parse_initial_states` method for samplers that accept initial states.
* `IdentitySampler` that returns passed-in initial states.
* `ising_to_qubo` function now accepts an underspecified `h` arg.

Fixes
-----
* General performance and structural improvements to C++ bqms.
* cyBQMs are now pickleable and deepcopyable.

Changes
-------
* Removed deprecated kwarg in `SpinReversalTransformComposite`

Fixes
-----
* `RoofDualityComposite`, `ConnectedComponentComposite` and `FixedVariableComposite` now all work with the new BQM types

New Features
------------
* A sample method testing framework for Samplers
* Significant documentation update
* A unified location for header files

Fixes
-----
* Fix numpy version range
* `BQM.get_quadratic` now has consistent/correct behaviour when a default is provided
* Restored some removed methods, `BQM.to_coo`, `.from_coo`, `.to_numpy_matrix`, `.from_numpy_matrix`, `.to_networkx_graph`, `.from_networkx_graph`
* `BQM.energies` now raises the correct exception when given bad samples

New Features
------------
* Binary quadratic models can now be constructed from linear/quadratic iterators
* `BQM.base` attribute to help when constructing from views
* `BQM.energies` mixin now uses a faster numpy-based energy calculation