RuntimeWarning on csgraph.shortest_path when edge lengths are NaN

[michael-s-crouch]

The documentation for scipy.sparse.csgraph says that NaN entries will be treated as non-edges, the same as 0 or inf. But when using csgraph.shortest_path on a graph where some paths are impossible because of NaN edges, irrelevant warning messages are printed. These warning messages don't happen in the other cases.

>>> import numpy
>>> import scipy.sparse.csgraph as csgraph
>>> print csgraph.shortest_path(numpy.array([[0,numpy.inf],[numpy.inf,0]]))
[[  0.  inf]
 [ inf   0.]]
>>> print csgraph.shortest_path(numpy.array([[0,0],[0,0]]))
[[  0.  inf]
 [ inf   0.]]
>>> print csgraph.shortest_path(numpy.array([[0, numpy.nan], [numpy.nan, 0]]))
/Users/creidieki/.virtualenvs/bugtest/lib/python2.7/site-packages/numpy/ma/core.py:2200: RuntimeWarning: invalid value encountered in less_equal
  condition = umath.less_equal(mabs(xnew - value), atol + rtol * mabs(value))
bug.py:4: RuntimeWarning: invalid value encountered in greater
  print csgraph.shortest_path(numpy.array([[0,numpy.nan],[numpy.nan,0]]))
bug.py:4: RuntimeWarning: invalid value encountered in less
  print csgraph.shortest_path(numpy.array([[0,numpy.nan],[numpy.nan,0]]))
[[  0.  inf]
 [ inf   0.]]

Using numpy v1.9.2, scipy v0.16.0, python 2.7.10 in a clean virtualenv on os x 10.10.4.

[rgommers]

thanks @creidieki. Those warnings are harmless but very annoying. They should be easy to avoid.