Astroml

==================

- Add ``fetch_sdss_galaxy_images`` to download sample of SDSS galaxy images
for CNN figure example. [242]
- Fix bug in ``lumfunc.Cminus`` that lead to return NaN values. [237]

==================

- Fix that PyMC3 is an optional dependency. [228]

================

- Added LinearRegressionwithErrors to handle errors in both dependent and
independent variables using pymc3. [206]

- Removed suppport for Python versions <3.5. [174]

- Deprecated function ``savitzky_golay`` in favour of the scipy
implementation. [193]

- Deprecated functions ``check_random_state`` and ``logsumexp`` in favour of
their equivalent in scikit-learn and scipy, respectively. [190]

==================

- Fix syntax for matplotlib.rc usage. [188]

- Various code cleanups and updates to the website.

================

- New utils subpackage, including deprecated decorator and new warning
types. ``astroML.decorators`` has been moved to this subpackage. [141]


API Changes and Other Changes
-----------------------------

- Removed deprecated KDE class. [119]

- Switched to use the updated scikit-learn API for GaussianMixture. This
change depends on scikit-learn 0.18+. [125]

- Minimum required astropy version is now 1.2. [173]

- Deprecated ``astroML.cosmology.Cosmology()`` in favour of
``astropy.cosmology``. [121]

- Deprecated the Lomb-Scargle periodograms from ``astroML.time_series`` in
favour of ``astropy.stats.LombScargle``. [173]

- Deprecated histograms, including Bayesian blocks, as they have been moved
to ``astropy.stats``. [142]

- The book and paper figures has been moved out to their separate
repository (``astroML_figures``).

================

- Add support for Python 3
- Add continuous integration via Travis
- Bug: correctly account for errors in Ridge/Lasso regression
- Add figure tests in ``compare_images.py``