Pyspeckit

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

* MAJOR bugfix 125: Discovered a major issue introduced in 122 in which a
typo in the radiative transfer equation results in an incorrect model

~~~~~~~~~~~~~~~~~~~~~~~~~~~

Major change: this is the first version of pyspeckit fully compatible with
python3.

* Bugfixes: 92, 93 related to fiteach, parameter validation,
spectroscopic axis array handling, in_range checks, and the ammonia fit
and plot wrapper. Also includes an update to the versioning scheme such
that dev versions (like 0.1.18.dev) will include a git devstring when
possible.
* Bugfix 91: Offset of 0.13 km/s between ammonia frequency used and that
in the LOVAS catalog
* Bugfix 90: Allow fitting of multiple peaks interactively (again)
* Enhancement 89: Allow data to be passed to Spectrum as a quantity
* Bugfix 88: line_ids needed to handle quantities
* Enhancement 85: Remove hard-coded T_A* label
* Bugfix 82: Fix a loophole in which parameters were not checked for being
in bounds when passed as `guesses`
* Addition 99: fiteach example with NIR data cubes, specfit example using
user-created model, and a guide to installing pyspeckit via GitHub.
* Enhancement 100: Display EQW highlighting at appropriate location
* Addition 102: New N2D+ model
* Bugfix 105: Ammonia thin works (though it is still not recommended)
* Bugfix 106: Baselines were dependent on the X-axis coordinate unit.
* Enhancement 108, 110, 111, 112: Python 3 compatibility
* Bugfix 113: Improved plotting & robustness of measure_fwhm, especially
when baseline is not subtracted
* Bugfix 115: py3ify steppify
* Enhancement 117/119: Better error messages when using
``SpectralCube.load_model`` and more robust figure (re-)opening (if you
closed a figure created by spectral cube and then tried to plot again, it
would result in a crash. This behavior was never supported, so it is not
a bugfix, but now the behavior should work)
* Bugfix 118: Re-fitting within a given window sometimes failed, possibly
due to how matplotlib handles event handlers. Refactoring should make
this much more robust.
* Enhancement 120: astropy-helpers + travis-ci related infrastructure
update
* Bugfix 121: Use correct variable name in N2D+ degeneracy
* Bugfix 122: Informative errors on file reading & exact RT equation in
``hyperfine`` model
* Addition 124: LTE model

~~~~~~~~~~~~~~~~~~~~~~~~~~~

* Bugfixes: 67, 69, 71, 74, 75 related to fiteach
* Bugfix for error bar plotting (PR 76, issue 78)
* Documentation cleanup and enhancement (77)

~~~~~~~~~~~~~~~~~~~~~~~~~~~

* Major refactor: use astropy's Quantity and units to replace pyspeckit's
minimalist unit framework. You may see deprecation warnings as a result!
(https://github.com/pyspeckit/pyspeckit/pull/26)
* The refactor led to many subsequent bugfixes: 61, 55, 51, and others
* Removal of the `peakbgfit` default method: instead, the default is to treat
all fits as multifits. Changes came from 32, 25
* New ammonia models (28, 50): absorption against a background source and
treating each line independently without a temperature connecting them

~~~~~~~~~~~~~~~~~~~~~~~~~~~
* Bugfix: write_fit is part of Cube, not CubeStack
* Bugfix: excludefit must occur after selectregion if fit_plotted_area is True
* API change: For the fitter & baseliner, data selection is end-inclusive
if specified in world coordinates
* Bugfix: numpy 1.8 added a "writeable" flag that broke units; that is now
corrected
http://docs.scipy.org/doc/numpy/reference/generated/numpy.ndarray.flags.html
* Baseline: add a `fit` method that is independent from `button2action` to
make masking and fitting independent processes
* Added `astropy <https://www.astropy.org>`_ as a dependency
* Converted to astropy-helper template for setup
* Changed 'units' -> 'unit' in Spectrum
* Baseline: add spline fitting
* Add Zenodo badge
* Allow `parinfo` to be used in place of `guesses` in specfit

~~~~~~~~~~~~~~~~~~~~~~~~~~~
* Bugfix: integral(direct=True) double-subtracted the baseline if it was
already subtracted
* New Feature: Models now include analytic integrals (only implemented for
Gaussian so far)
* New Feature: hyperfine models added that allow varying amplitudes & widths