Two bugs in quiver?

I think there may be two bugs in quiver.

1. Quiver doesn't seem to accept 1D arrays X and Y if they are
different sizes -- they need to be turned into 2d arrays with meshgrid
for the program to work. With line 13 uncommented I get the following
error message:

  File "quivtest.py", line 13, in <module>
    ax.quiver(XT,YT,uu,vv) #fails
  File "/Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/site-packages/matplotlib/axes.py",
line 5533, in quiver
    q = mquiver.Quiver(self, *args, **kw)
  File "/Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/site-packages/matplotlib/quiver.py",
line 343, in __init__
    self.XY = np.hstack((X[:,np.newaxis], Y[:,np.newaxis]))
  File "/Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/site-packages/numpy/lib/shape_base.py",
line 488, in hstack
    return _nx.concatenate(map(atleast_1d,tup),1)
ValueError: array dimensions must agree except for d_0

2. Masking is not working properly with quiver, while it works fine
with contour -- see the attached quivtest.png

quiver.py:

import numpy as np
import numpy.ma as ma
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
XT = np.arange(10.)
YT = np.arange(20.)
uu = np.zeros([20,10])+7.
vv = np.zeros([20,10])+3.

fig = plt.figure()
ax = fig.add_subplot(121)
# ax.quiver(XT,YT,uu,vv) #fails
XX,YY = np.meshgrid(XT,YT)
ax.quiver(XX,YY,uu,vv) #Ok

ax = fig.add_subplot(122)
msk = XX**2+YY**2 >50
uumsk = ma.masked_array(uu,mask=msk)
vvmsk = ma.masked_array(vv,mask=msk)
ax.contour(XT,YT,uumsk*XX) # works
ax.quiver(XX,YY,uumsk,vvmsk) # gives strange shapes
plt.show()
fig.savefig('quivtest.png',dpi=100)

George Nurser.

quivtest.png

George Nurser wrote:

I think there may be two bugs in quiver.

Thanks for the report. I can reproduce both problems, and I will try to solve them.

Eric

···

1. Quiver doesn't seem to accept 1D arrays X and Y if they are
different sizes -- they need to be turned into 2d arrays with meshgrid
for the program to work. With line 13 uncommented I get the following
error message:

  File "quivtest.py", line 13, in <module>
    ax.quiver(XT,YT,uu,vv) #fails
  File "/Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/site-packages/matplotlib/axes.py",
line 5533, in quiver
    q = mquiver.Quiver(self, *args, **kw)
  File "/Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/site-packages/matplotlib/quiver.py",
line 343, in __init__
    self.XY = np.hstack((X[:,np.newaxis], Y[:,np.newaxis]))
  File "/Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/site-packages/numpy/lib/shape_base.py",
line 488, in hstack
    return _nx.concatenate(map(atleast_1d,tup),1)
ValueError: array dimensions must agree except for d_0

2. Masking is not working properly with quiver, while it works fine
with contour -- see the attached quivtest.png

quiver.py:

import numpy as np
import numpy.ma as ma
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
XT = np.arange(10.)
YT = np.arange(20.)
uu = np.zeros([20,10])+7.
vv = np.zeros([20,10])+3.

fig = plt.figure()
ax = fig.add_subplot(121)
# ax.quiver(XT,YT,uu,vv) #fails
XX,YY = np.meshgrid(XT,YT)
ax.quiver(XX,YY,uu,vv) #Ok

ax = fig.add_subplot(122)
msk = XX**2+YY**2 >50
uumsk = ma.masked_array(uu,mask=msk)
vvmsk = ma.masked_array(vv,mask=msk)
ax.contour(XT,YT,uumsk*XX) # works
ax.quiver(XX,YY,uumsk,vvmsk) # gives strange shapes
plt.show()
fig.savefig('quivtest.png',dpi=100)

George Nurser.

------------------------------------------------------------------------

------------------------------------------------------------------------

------------------------------------------------------------------------------
This SF.net email is sponsored by:
SourcForge Community
SourceForge wants to tell your story.
http://p.sf.net/sfu/sf-spreadtheword

------------------------------------------------------------------------

_______________________________________________
Matplotlib-users mailing list
Matplotlib-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/matplotlib-users

George Nurser wrote:

I think there may be two bugs in quiver.

Actually just 1...

1. Quiver doesn't seem to accept 1D arrays X and Y if they are
different sizes -- they need to be turned into 2d arrays with meshgrid
for the program to work. With line 13 uncommented I get the following
error message:

This is a bug. It is fixed now in svn trunk and 98.5 maintenance branch.

2. Masking is not working properly with quiver, while it works fine
with contour -- see the attached quivtest.png

This is a numpy.ma bug that has been fixed in numpy. I don't know exactly when it was fixed, but after I tracked it down to numpy.ma, I updated my copy of numpy from svn, and the problem went away. Somehow, when a multiplication was done with broadcasting, the mask was getting handled and propagated incorrectly. I suspect this bug was present only for a moderate interval of time, because quiver with masked points worked for a long time before you ran into the bug.

Eric

Downloaded latest svn numpy and matplotlib and it works perfectly now.
Thanks for sorting it out so quickly.

George.

2009/1/20 Eric Firing <efiring@...202...>:

···

George Nurser wrote:

I think there may be two bugs in quiver.

Actually just 1...

1. Quiver doesn't seem to accept 1D arrays X and Y if they are
different sizes -- they need to be turned into 2d arrays with meshgrid
for the program to work. With line 13 uncommented I get the following
error message:

This is a bug. It is fixed now in svn trunk and 98.5 maintenance branch.

2. Masking is not working properly with quiver, while it works fine
with contour -- see the attached quivtest.png

This is a numpy.ma bug that has been fixed in numpy. I don't know exactly
when it was fixed, but after I tracked it down to numpy.ma, I updated my
copy of numpy from svn, and the problem went away. Somehow, when a
multiplication was done with broadcasting, the mask was getting handled and
propagated incorrectly. I suspect this bug was present only for a moderate
interval of time, because quiver with masked points worked for a long time
before you ran into the bug.

Eric