What’s New in Astropy 2.0?


Astropy 2.0 is a major release that adds significant new functionality since the 1.3.x series of releases.

On this page, you can read about some of the big changes in this release:

In addition to these major changes, Astropy 2.0 includes a large number of smaller improvements and bug fixes, which are described in the Full Changelog. By the numbers:

  • 696 issues have been closed since v1.3
  • 316 pull requests have been merged since v1.3
  • 232 distinct people have contributed code

New unit support for most models

Most Astropy models now can handle inputs with units, and produce the appropriate outputs with units as well. Some models cannot support this due to their definitions (e.g., Legendre, Hermite, etc), while some will have this capability added in a future release. Example usage:

>>> from astropy import units as u
>>> from astropy.modeling.models import Gaussian1D
>>> g = Gaussian1D(amplitude=1*u.J, mean=1*u.m, stddev=0.1*u.m)
>>> g([3, 4, 5.5] * u.cm)
<Quantity [  3.70353198e-21,  9.72098502e-21,  4.05703276e-20] J>

For more information, see Support for units and quantities.

New image class CCDData added

A new class, CCDData, has been added to the astropy.nddata package. It can read from/write to FITS files, provides methods for arithmetic operations with propagation of uncertainty, and support for binary masks. For examples of how to use this class, see the Getting started section of the astropy.nddata documentation.

Experimental velocity support in astropy.coordinates

Astropy coordinate frame objects now contains experimental support for storing and transforming velocities. This includes, among other things, support for transforming proper motion components between coordinate frames and transforming full-space velocities to/from a local standard of rest (LSR) frame and a Galactocentric frame. For example, to transform a set of proper motions from the Galactic frame to the ICRS frame:

>>> from astropy.coordinates import Galactic, ICRS
>>> gal = Galactic(l=8.67*u.degree, b=53.09*u.degree,
...                pm_l_cosb=-117*u.mas/u.yr, pm_b=13*u.mas/u.yr)
>>> gal.transform_to(ICRS)
<ICRS Coordinate: (ra, dec) in deg
    ( 226.45743375,  8.3354549)
 (pm_ra_cosdec, pm_dec) in mas / yr
    (-77.61973364, -88.50523685)>

Or, for example, to transform a 3D velocity from the ICRS frame to a Galactocentric frame with custom values for the sun-galactic center distance and solar velocity vector:

>>> icrs = ICRS(ra=11.23*u.degree, dec=58.13*u.degree,
...             distance=213.4*u.pc,
...             pm_ra_cosdec=9*u.mas/u.yr, pm_dec=3*u.mas/u.yr,
...             radial_velocity=-61*u.km/u.s)
>>> v_sun = coord.CartesianDifferential([10, 244, 7.])*u.km/u.s
>>> gc = icrs.transform_to(coord.Galactocentric(galcen_distance=8*u.kpc,
>>> gc.x, gc.y, gc.z
(<Quantity -8112.928728515727 pc>,
 <Quantity 180.22175948399217 pc>,
 <Quantity 9.781203623025618 pc>)
>>> gc.v_x, gc.v_y, gc.v_z
(<Quantity 34.40211035247248 km / s>,
 <Quantity 187.80653073084486 km / s>,
 <Quantity 14.74171285614737 km / s>)

The velocity support works by adding support for “differential” objects which contain differences of representations. For more details, see Working with velocities in Astropy coordinates. This functionality will likely be added to the SkyCoord class in future.

In addition, the SkyCoord class now has a radial_velocity_correction method which can be used to compute heliocentric and barycentric corrections for radial velocity measurements. While in the future this may use the mechanisms described above, currently it uses a simpler algorithm for numerical stability. A simple example of using this functionality might be:

>>> from astropy.coordinates import SkyCoord, EarthLocation
>>> from astropy.time import Time
>>> obstime = Time('2017-2-14')
>>> target = SkyCoord.from_name('M31')
>>> keck = EarthLocation.of_site('Keck')
>>> target.radial_velocity_correction(obstime=obstime, location=keck).to('km/s')
<Quantity -22.363056056262263 km / s>

New functionality in astropy.stats

New sigma-clipping class

A new SigmaClip class has been added as an object-oriented interface for sigma clipping:

>>> from astropy.stats import SigmaClip
>>> data = [1, 5, 6, 8, 100, 5, 3, 2]
>>> sigclip = SigmaClip(sigma=2, iters=5)
>>> print(sigclip)  
    sigma: 3
    sigma_lower: None
    sigma_upper: None
    iters: 10
    cenfunc: <function median at 0x108dbde18>
    stdfunc: <function std at 0x103ab52f0>
>>> sigclip(data)
masked_array(data = [1 5 6 8 -- 5 3 2],
             mask = [False False False False  True False False False],
       fill_value = 999999)

Note that once the sigclip instance is defined above, it can be applied to other data, using the same, already-defined, sigma-clipping parameters.

New robust statistical functions

New biweight_midcovariance() and biweight_midcorrelation() functions were added to astropy.stats. The biweight_midcovariance() function computes the robust covariance between two or more variables, and biweight_midcorrelation() computes a robust measure of similarity between two variables.

For example:

>>> import numpy as np
>>> from astropy.stats import biweight_midcovariance
>>> from astropy.stats import biweight_midcorrelation
>>> # Generate two random variables x and y
>>> rng = np.random.RandomState(1)
>>> x = rng.normal(0, 1, 200)
>>> y = rng.normal(0, 3, 200)
>>> # Introduce an obvious outlier
>>> x[0] = 30.0
>>> # Calculate the biweight midcovariances between x and y
>>> bicov = biweight_midcovariance([x, y])
>>> print(bicov)  
[[ 0.82483155 -0.18961219]
 [-0.18961219 9.80265764]]
>>> # Print standard deviation estimates
>>> print(np.sqrt(bicov.diagonal()))  
[ 0.90820237  3.13091961]
>>> # Compute the biweight midcorrelation between x and y
>>> print(biweight_midcorrelation(x, y))  

New statistical estimators for Ripley’s K Function

New statistical estimators for Ripley’s K Function, RipleysKEstimator, in astropy.stats. For example:

import numpy as np
from matplotlib import pyplot as plt
from astropy.stats import RipleysKEstimator
z = np.random.uniform(low=5, high=10, size=(100, 2))
Kest = RipleysKEstimator(area=25, x_max=10, y_max=10, x_min=5, y_min=5)
r = np.linspace(0, 2.5, 100)
plt.plot(r, Kest.poisson(r), label='poisson')
plt.plot(r, Kest(data=z, radii=r, mode='none'), label='none')
plt.plot(r, Kest(data=z, radii=r, mode='translation'), label='translation')
plt.plot(r, Kest(data=z, radii=r, mode='ohser'), label='ohser')
plt.plot(r, Kest(data=z, radii=r, mode='var-width'), label='var-width')
plt.plot(r, Kest(data=z, radii=r, mode='ripley'), label='ripley')
plt.legend(loc='upper left')

(png, svg, pdf)


Easier use of efficient bytestring Table columns in Python 3

Working with bytestring Table columns (numpy 'S' dtype) in Python 3 has been made more convenient because it is now possible to compare and set array elements with the natural Python string (str) type. Previously one had to use the Python bytes type and bytestring literals like b'hello'. This change allows working with ASCII data columns in Python 3 using only 1-byte per character instead of the default 4-bytes per character for the numpy 'U' unicode dtype. For large datasets this improves memory performance.

Please see Bytestring columns in Python 3 for details. Note that no change has been made to behavior for Python 2.


This introduces an API change that affects comparison of bytestring column elements in Python 3.

If comparison with str instead of bytes is a problem (and bytes is really more logical), please open an issue on GitHub.

New way to instantiate a BinTableHDU directly from a Table

A new way to instantiate a FITS BinTableHDU directly from a Table object. For example:

>>> from astropy.io import fits
>>> from astropy.table import Table
>>> tab = Table([[1, 2, 3], ['a', 'b', 'c'], [2.3, 4.5, 6.7]],
...             names=['a', 'b', 'c'], dtype=['i', 'U1', 'f'])
>>> hdu = fits.BinTableHDU(tab)

New printdiff convenience function for FITS

A new printdiff() convenience function was added for comparison between FITS files or HDUs. For example:

>>> from astropy.io import fits
>>> hdu1 = fits.ImageHDU([1, 2, 3])
>>> hdu2 = fits.ImageHDU([1, 2.1, 3])
>>> fits.printdiff(hdu1, hdu2)

Headers contain differences:
  Keyword BITPIX   has different values:
     a> 64
     b> -64
      ? +

Data contains differences:
  Data differs at [2]:
       (int64) a> 2
     (float64) b> 2.1000000000000001
  1 different pixels found (33.33% different).

New molar_mass_amu unit equivalency

A new equivalency named molar_mass_amu has been added to convert between g/mol unit to atomic mass unit (amu). For example:

>>> from astropy import constants as const
>>> from astropy import units as u
>>> x = 1 * (u.g / u.mol)
>>> y = 1 * u.u
>>> x.to(u.u, equivalencies=u.molar_mass_amu())
<Quantity 1.0 u>
>>> y.to(u.g/u.mol, equivalencies=u.molar_mass_amu())
<Quantity 1.0 g / mol>

Store astropy core object types in ASCII ECSV table file

It is now possible to store the following mixin column types in an ASCII ECSV table file: Time, TimeDelta, Quantity, Latitude, Longitude, Angle, Distance, EarthLocation, SkyCoord. The table file can then be read back into astropy with no loss of object data or attributes.

Improvements to astropy.convolution

Convolution has undergone a significant overhaul to make fft and direct convolution consistent. They keyword arguments have changed and the behavior of convolve is no longer the same as in versions prior to 2.0 (although convolve_fft’s behavior remains unchanged). The details are given on the astropy convolution.

No relativistic species by default in cosmological models

For all of the built in cosmological model types (e.g., FlatLambdaCDM) the default CMB temperature at z=0 is now 0K, which corresponds to no contributions from photons or neutrinos (massive or otherwise). This does not affect built in literature models (such as the WMAP or Planck models). The justification is to avoid including mass-energy components that the user has not explicitly requested. This is a non-backwards compatible change, although the effects are small for most use cases.

Renamed/removed functionality

Several sub-packages have been moved or removed, and these are described in the following sections.


The bundled version of pytest has now been removed, but the astropy.tests.helper.pytest import will continue to work properly. Affiliated packages should nevertheless transition to importing pytest directly rather than from astropy.tests.helper. This also means that pytest is now a formal requirement for testing for both Astropy and for affiliated packages.


The cone search module has been moved to Astroquery (0.3.5 and later) and will be removed from Astropy in a future version. The API here will be preserved as the “classic” API in Astroquery, however some configuration behavior might change; See the Astroquery documentation for new usage details.


The SAMP (Simple Application Messaging Protocol) module, formerly available in astropy.vo.samp, has now been moved to astropy.samp, so you should update any imports to this module.

Full change log

To see a detailed list of all changes in version v2.0, including changes in API, please see the Full Changelog.