Xlmhg

~~~~~~~~~~~~~~~~~~
- Improved docstrings and added a user manual (https://xl-mhg.readthedocs.io)

------------------
- Added arguments `exact_pval` and `escore_pval_thresh` to
`get_xlmhg_test_result()`. Also, the `pval_thresh` argument now has a
different meaning.

------------------
- Added binary distributions for Windows (32/64-bit), Mac OS X (10.6+
64-bit), and Linux (32/64-bit), for both Python 2.7 and Python 3.5. This
means that for all of these platforms/environments, the installation of the
`xlmhg` package (`pip install xlmhg`) no longer requires a C compiler to
be present.

------------------
- Changed internal structure used to represent lists, from vector of size N
to vector containing only the indices of the 1's. This saves memory and
avoids storing redundant information.
- Added the `get_xlmhg_test_result()` front-end function, which returns an
`mHGResult` object.

2.1.x updates
-------------
- 2.1.1 (2016-05-01): Fixed readme

------------------
- Added Cython implementation for calculating XL-mHG E-scores
- Added `mHGResult` class for representing test results
- Added tests
- Fixed a few minor issues

2.0.x updates
-------------
- 2.0.7 (2016-04-21): Fixed small problem in setup script
- 2.0.5 (2016-04-19): Fixed an uninstended change introduced in 2.0.4 whereby
a cythonized version instead of the mhg_cython.pyx file was included in the
package
- 2.0.4 (2016-04-19): Added tests/ and CHANGELOG.rst to Manifest.in file
- 2.0.3 (2016-04-18): Including Travis CI build status in Readme
- 2.0.2 (2016-04-18): Integration with Travis CI
- 2.0.1 (2016-04-15): Readme fixes

------------------
Major release with several new features:

- New API (`xlmhg.xlmhg_test()`; see `test.py`).
- Implementation of ``PVAL2`` algorithm for calculating XL-mHG p-values.
This algorithm offers better performance and numerical stability and is
now used by default.
- Implementation of ``PVAL-BOUND`` algorithm for calculating O(N)-bound.
- Implementation of ``PVAL-THRESH`` algorithm for deciding whether the
XL-mHG p-value meets a given signifance level.
- Unit tests to ensure the correctness of all algorithms (using `pytest`).

For details regarding the algorithms, see `Wagner (PeerJ Preprints, 2016)
<https://doi.org/10.7287/peerj.preprints.1962v2>`_.