Pyusid

Major user-facing changes
* `pyUSID`
* `io`
* `hdf_utils`
* `reshape_to_n_dims` now sorts dimensions from slowest to fastest instead of fastest to slowest in positions and spectroscopic.
* `write_ind_val_dsets` now allows dimensions to be ordered either by fastest to slowest (default) or vice versa. Default will be swapped in next release
* `write_main_dataset` now assumes by default that the dimensions are arranged from fastest varying to slowest varying. Dimensions can now be specified in opposite order as well by setting `slow_to_fast` to `True`
* `USIDataset`'s `sort_dims` now sorts dimensions from slowest to fastest varying
* `Translator`
* `is_valid_file`- implemented new function that returns the path of the input file that must be provided to `translate()`
* `generate_dummy_main_parms()` removed
* `processing`
* `parallel_compute` now uses `multiprocessing` backend by default
* `process`
* `Process` now requires `process_name` and optionally `parms_dict` to be passed to `__init__` for automated checking of existing results
* `Process` can now return dask.array object from `read_data_chunk()`
* `parallel_compute` - empty copy removed from this legacy position

Major Internal changes
Features
* New function - `pyUSID.io.USIDataset.__slice_n_dim_form()` accelerates and simplifies slicing of datasets having an N-dimensional form (most popular use-case)

More robust checking
* `pyUSID.io.hdf_utils.reshape_to_n_dims`
* ``pyUSID.processing.Process``
* ``pyUSID.io.USIDataset``
* ``pyUSID.io.hdf_utils.simple.check_and_link_ancillary()``

Bug fixes:
* ``pyUSID.Process``:
* Now able to specify which partial group to continue computation from
* Corrected identification of existing partial and complete results
* `pyUSID.io.hdf_utils.get_unit_values` now catches dimensions in spectroscopic indices that do not start from `0`
* Explicitly stating mode of opening hdf5 files per new h5py versions
* Better forward compatible import statements
* `Process` cookbook now fixed to use

Testing infrastructure:
* Standard `BEPS` dataset
* Each dimension's values and the N-dimensional form also written to HDF5 for quick and accurate testing in both `hdf_utils` and `USIDataset`
* Results now are compound and complex-valued to cover more ground in testing
* Now being used to test `USIDataset` as well
* Far more extensive tests on reshaping with varying dimension ordering, `numpy` vs `dask.array` etc.
* Built configurable base tests that whose arguments can be changed to simulate different unit tests in `hdf_utils/model`

Added unit tests for following:
* ``io``
* ``io_utils``
* ``hdf_utils``
* ``print_tree``
* ``reshape_to_n_dims``
* ``USIDataset``
* ``processing``
* ``Process``
* ``comp_utils``

[Code coverage](https://coveralls.io/builds/28104310) rose from 51% to 64%. `viz` is the last remaining sub-package that needs significant testing

* Bug fixes associated with `pyUSID.io.dtype_utils.stack_real_to_target_dtype`
* Bug fixes associated with latest `numpy` and `h5py` versions breaking `pyUSID.USIDataset`
* Improvements to `MPI` portions of the `pyUSID.Process` class
* Parts of `pyUSID.viz.plot_utils` now work with `dask.array` inputs

Bug fix - Added back deprecated skeletons of functions that were moved out from hdf_utils into reg_ref

Mainly a bug fix to ensure that HDF5 datasets are consistently and correctly loaded as Dask arrays

* Significant changes to requirements
* Lowered numpy to 1.10.0
* Moved unittest2 to tests
* Added Dask
* Several operations now work with Dask
* Reshape to and from N dimensions
* Writing to Main dataset
* direct function
* NumpyTranslator - renamed to ArrayTranslator
* Data transformations in dtype_utils
* USIDataset with improved functionality
* Reduce
* Slice to dataset
* Improvements to tests
* Test classes now share functions in data_utils
* Reogranized tests and functions - region references
* LOT more unit tests
* Testing modules broken down into multiple classes
* hdf_utils reorganized into three submodules:
* Base - USID agnostic HDF5 helper functions
* Simple - USID specific but simple helper functions
* model - model dependent (n-dimensional, sparse, incomplete) functions
* Reusable string argument validation functions added
* Moving towards working with datasets without N-dimensional form:
* Added new class - DimType to account for different kinds of dimensions
* Patrik Marschalik contributions for interpolating spiral / other scans to cartesian coordinates
* far more robust get_unit_values() works with incomplete dimensions, etc.

Minor changes and mainly bug fixes over 0.0.5
* Reduced requirements - scikit-image replaced by simpler pillow, pyqtgraph removed since it was unused, pyQT5 now optional requirement
* Process class now correctly identifies partial datasets
* USIDataset.visualize() now correctly works for 2D images