Bug in coordinate transform with WCSLib 5 #3891

Activity

1. astrofrog

   assigned

   mdboom

   on Jun 26, 2015
2. astrofrog

   added

   Bug

   Affects-devPRs and issues that do not impact an existing Astropy releasePRs and issues that do not impact an existing Astropy release

   wcs

   🔥CriticalCritical

   on Jun 26, 2015
3. mdboom

   added this to the v1.1.0 milestone on Jun 26, 2015
4. mdboom commented on Jun 26, 2015

   

   mdboom

   on Jun 26, 2015

   Contributor

   More actions

   Wow. That's sure something. Looking into it now.
5. cdeil commented on Jun 29, 2015

   

   cdeil

   on Jun 29, 2015

   Member

   More actions

   The change to WCSLib 5 also broke a test in Gammapy:
   https://travis-ci.org/gammapy/gammapy/jobs/68707705#L1115

   Here's the print output corresponding to this test case that shows that the issue is in the WCS transformations, it doesn't round-trip:

   import numpy as np
   from astropy.wcs import WCS
   from gammapy.image import make_header
   
   header = make_header(nxpix=2, nypix=1, binsz=10, xref=0, yref=0, proj='CAR')
   # GLON pixel edges: (+10, 0, -10)
   # GLAT pixel edges: (-5, +5)
   
   EPS = 0.1
   data = [
           ( 5,  5,  1),        # in image[0, 0]
           ( 0,  0 + EPS,  2),  # in image[1, 0]
           ( 5, -5 + EPS,  3),  # in image[0, 0]
           ( 5,  5 + EPS, 99),  # outside image
           (10 + EPS, 0, 99),   # outside image
           ]
   lon, lat, weights = np.array(data).T
   
   wcs = WCS(header)
   origin = 0
   x, y = wcs.wcs_world2pix(lon, lat, origin)
   lon2, lat2 = wcs.wcs_pix2world(x, y, origin)
   print(wcs)
   print()
   print('weights : ', weights)
   print('lon     : ', lon)
   print('lon2    : ', lon2)
   print('lat     : ', lat)
   print('lat2    : ', lat2)
   print('x       : ', x)
   print('y       : ', y)

   Output:

   WCS Keywords
   
   Number of WCS axes: 2
   CTYPE : 'GLON-CAR'  'GLAT-CAR'  
   CRVAL : 0.0  0.0  
   CRPIX : 1.5  1.0  
   PC1_1 PC1_2  : 1.0  0.0  
   PC2_1 PC2_2  : 0.0  1.0  
   CDELT : -10.0  10.0  
   NAXIS    : 2 1
   
   weights :  [  1.   2.   3.  99.  99.]
   lon     :  [  5.    0.    5.    5.   10.1]
   lon2    :  [ 5.   5.   0.   0.1  5. ]
   lat     :  [ 5.   0.1 -4.9  5.1  0. ]
   lat2    :  [ 5.   0.1 -4.9  5.1  0. ]
   x       :  [ 0.    0.    0.5   0.49  0.  ]
   y       :  [ 0.5   0.01 -0.49  0.51  0.  ]
   

   I didn't check if #3893 fixes this case also.
6. astrofrog commented on Jun 29, 2015

   

   astrofrog

   on Jun 29, 2015

   MemberAuthor

   More actions

   @cdeil - if you don't want to try out #3893, you can also just check whether converting the coordinates one by one works (the bug here is related to iteration over coordinate arrays)
7. cdeil commented on Jun 29, 2015

   

   cdeil

   on Jun 29, 2015

   Member

   More actions

   #3893 did fix the issue in Gammapy also. Thanks, @mdboom !
8. mdboom commented on Jun 30, 2015

   

   mdboom

   on Jun 30, 2015

   Contributor

   More actions

   Fixed by #3893.
9. mdboom

   closed this as completedon Jun 30, 2015