Error adding meshes

[fipymigrate]

When I add meshes like
mesh = mesh1 + mesh2
I obtain the error

mesh1 = mesh1 + mesh2
File "/usr/lib/python2.7/dist-packages/numpy/lib/arraysetops.py", line 328, in in1d
mask |= (ar1 == a)
ValueError: operands could not be broadcast together with shapes (2) (2,8140) (2)

The solution would be in abstractMesh.py to flatten the arrays first:

@@ -351,11 +351,11 @@
self_faceVertexIDs = MA.masked_values(self_faceVertexIDs, -1)

want self's Faces for which all faceVertexIDs are in vertexCorrelates

•    self_matchingFaces = numerix.in1d(self_faceVertexIDs,
  

•    self_matchingFaces = numerix.in1d(self_faceVertexIDs.flatten(),
  

  vertexCorrelates[0]).reshape(self_faceVertexIDs.shape).all(axis=0).nonzero()[0]

want other's Faces for which all faceVertexIDs are in vertexCorrelates

•    other_matchingFaces = numerix.in1d(other_faceVertexIDs,
  

•    other_matchingFaces = numerix.in1d(other_faceVertexIDs.flatten(),
  

  vertexCorrelates[1]).reshape(other_faceVertexIDs.shape).all(axis=0).nonzero()[0]

I find it somehow strange I'm the only one hitting this.

Imported from trac ticket #368, created by bm@cage.ugent.be on 10-11-2011 at 10:09, last modified: 09-09-2013 at 18:43

[fipymigrate]

bm@cage.ugent.be attached test.py on 10-13-2011 at 03:31

[fipymigrate]

Note that it is a pitty the array iterator .flat cannot be used with in1d of numpy, so as to avoid the copying of the original array

Trac comment by anonymous on 10-11-2011 at 10:19

[guyer]

We certainly do test mesh concatenation.

Can you provide example meshes that cause the problem?

Trac comment by guyer on 10-12-2011 at 22:29

[fipymigrate]

I have it with this code. Note that I am using scipy trunk and numpy trunk

from fipy import *
from fipy.meshes.numMesh.grid2D import Grid2D
Lval = 2. # half length of a periodic entitiy
lval = 1. # half length of the slug
Hval = 0.1 # half height of the microchannel
hval = 0.01 # distance wall to slug
nxtot = 200 # intervalnr in x direction
nytot = 40 # intervalnr in y direction
dx = Lval / float(nxtot)
dy = Hval / float(nytot)
nx1 = (Lval-lval)/2/dx
nx2 = lval/dx
ny1 = Hval/dy
ny2 = hval/dy
mesh1 = Grid2D(dx=dx, dy=dy, nx=2*nx1, ny=ny1) + [[-(Lval-lval)/2], [0]]
mesh2 = Grid2D(dx=dx, dy=dy, nx=nx2/2, ny=ny2) + [[(Lval-lval)/2],[0]]
mesh3 = Grid2D(dx=dx, dy=dy, nx=nx2/2, ny=ny2) + [[-Lval/2], [0]]
mesh1 = mesh1 + mesh2

Which gives the error:
Traceback (most recent call last):
File "test.py", line 18, in
mesh1 = mesh1 + mesh2
File "/usr/lib/python2.7/dist-packages/FiPy-2.2_dev4804-py2.7.egg/fipy/meshes/abstractMesh.py", line 892, in add
return self._concatenatedClass(**self._getAddedMeshValues(other=other))
File "/usr/lib/python2.7/dist-packages/FiPy-2.2_dev4804-py2.7.egg/fipy/meshes/abstractMesh.py", line 356, in _getAddedMeshValues
vertexCorrelates[0]).reshape(self_faceVertexIDs.shape).all(axis=0).nonzero()[0]
File "/usr/lib/python2.7/dist-packages/numpy/lib/arraysetops.py", line 328, in in1d
mask |= (ar1 == a)
ValueError: operands could not be broadcast together with shapes (2) (2,8140) (2)

With change in bug comment, so .flatten(), things work again as before.

Trac comment by bm@cage.ugent.be on 10-13-2011 at 03:30

[fipymigrate]

Added file test.py, to avoid the bad formatting in the bug submission

Trac comment by bm@cage.ugent.be on 10-13-2011 at 03:31

[wd15]

Works for me with the following config:

fipy version 2.1.3-dev4751
numpy version 1.6.1
pysparse version 1.2-dev
PyTrilinos version 4.4
scipy version 0.9.0
matplotlib version 0.98.1
gist is not installed
mpi4py is not installed
enthought.mayavi is not installed

and python 2.5.2.

Trac comment by wd15 on 10-14-2011 at 12:09

[fipymigrate]

On my laptop at home it also works I see now. Monday I'll see what is different on my work box (newer numpy for sure, but then, it could be due to a numpy bug in trunk).

Trac comment by bm@cage.ugent.be on 10-14-2011 at 12:57

[wd15]

Do you use virtualenv and virtualenv-wrapper?

Trac comment by wd15 on 10-14-2011 at 13:02

[fipymigrate]

No, never heard of that packages before :-)
numpy trunk, scipy trunk, ... normally don't cause problems in those parts that are well established.

Trac comment by bm@cage.ugent.be on 10-14-2011 at 13:13

[wd15]

Replying to bm@…:

No, never heard of that packages before :-)
numpy trunk, scipy trunk, ... normally don't cause problems in those parts that are well established.

I was just pointing out virtualenv because it's extremely useful when you're switching between versions of various packages or doing any python development. It makes life much easier.

Trac comment by wd15 on 10-17-2011 at 09:46

[guyer]

bump

Benny, did you ever figure out what was different about your work configuration?

Our build slaves are all running NumPy 1.6.x, which is current and SciPy 0.9.x or 0.10.x, which is not. Any guess as to which is newer on your system?

Trac comment by guyer on 01-28-2013 at 11:57

[guyer]

We are unable to reproduce the failure. Please reopen if this is still an issue.

Trac comment by guyer on 09-09-2013 at 18:43