TypeError from purge_vertices()

graph-tool
1

I am trying to use purge_vertices() with a vertex filter to efficiently
remove vertices from a graph. I am seeing different behavior in version
2.2.18 and 2.2.25, both as prebuilt packages for Ubuntu precise (current
"graph-tool" and "python-graph-tool" packages, respectively)

I have a simple example script attached which shows the behavior. If I
run with graph-tool 2.2.18, it works without complaint, e.g.:

$ python graph-tool-play2.py
graph_tool version 2.2.18 (commit 83bb8a49, Tue Nov 13 00:57:29 2012 +0100)
pruning 11
pruning 13
pruning 14
pruning 19
pruning 23
pruning 26
pruning 46
pruning 47
pruning 48
pruning 52
pruning 63
pruning 81
pruning 93
pruning 97
complete

If I run with python-graph-tool 2.2.25, I get a TypeError, e.g.:

$ python graph-tool-play2.py
graph_tool version 2.2.25 (commit feec13d9, Mon Sep 2 00:01:39 2013 +0200)
pruning 10
pruning 12
pruning 13
pruning 18
pruning 22
pruning 25
pruning 45
pruning 46
pruning 47
pruning 51
pruning 62
pruning 80
pruning 92
pruning 96
Traceback (most recent call last):
   File "graph-tool-play2.py", line 59, in <module>
     g.purge_vertices() # MJH this will die in graph-tool 2.2.25 with a
TypeError. Works in 2.2.18.
   File "/usr/lib/python2.7/dist-packages/graph_tool/__init__.py", line
1838, in purge_vertices
     new_g = Graph(self, prune=(True, False, False))
   File "/usr/lib/python2.7/dist-packages/graph_tool/__init__.py", line
1109, in __init__
     vorder.fa = numpy.arange(gv.num_vertices())
   File "/usr/lib/python2.7/dist-packages/graph_tool/__init__.py", line
561, in <lambda>
     lambda self, v: self.__get_set_f_array(v, False),
   File "/usr/lib/python2.7/dist-packages/graph_tool/__init__.py", line
556, in __get_set_f_array
     a[m] = v
TypeError: array cannot be safely cast to required type

Why would this fail in 2.2.25? It's entirely possible that I am missing
something basic about how to fly graph-tool.

Any clues appreciated!

Cheers,

Mark.

graph-tool-play2.py (1.42 KB)

2

[snip]

Update... I built graph-tool from the 2.2.25 source tarball and ran from
that instead of the python-graph-tool 2.2.25-1 Precise deb package, and
my test code ran fine.

Wonder if there's some kind of dependency problem in the 2.2.25-1
Precise deb package?

FWIW I couldn't get 2.2.25 to build from source with the standard
Precise numpy package based on 1.6.1. I replaced that with numpy I built
from their 1.7.1 tarball and then was able to build graph-tool
successfully... after 2 hours of CPU time :wink:

Looking at the present python-graph-tool 2.2.25-1 precise package, it
depends on python-scipy (but no particular version or range specified)
and from there presumably depends indirectly on numpy. Maybe it is
willing to install with an older (1.6) numpy but is partially broken
with it?

Cheers,

Mark.

3

This is strange... I definitely was able to build it with numpy 1.6,
otherwise how could I have made the packages?

I do see the same problem on precise, but nowhere else. This all does
sound like a nasty numpy compatibility thing, but it is not at all clear
what is going on. I'll have to take a closer look.

Cheers,
Tiago

4

Thx Tiago. A little more detail might help: when I tried to compile
graph-tool 2.2.25 on Precise with the standard numpy 1.6.1 package, it
died while building graph_closeness, complaining of undefined
NPY_ARRAY_ALIGNED and brethren:

$ make
make all-recursive
make[1]: Entering directory `/home/markh/Code/graph-tool-2.2.25'
Making all in src
make[2]: Entering directory `/home/markh/Code/graph-tool-2.2.25/src'
Making all in graph
make[3]: Entering directory `/home/markh/Code/graph-tool-2.2.25/src/graph'
Making all in centrality
make[4]: Entering directory
`/home/markh/Code/graph-tool-2.2.25/src/graph/centrality'
   CXX graph_betweenness.lo
   CXX graph_centrality_bind.lo
   CXX graph_closeness.lo
In file included from graph_closeness.hh:29:0,
                  from graph_closeness.cc:23:
./../numpy_bind.hh: In function 'boost::python::api::object
wrap_vector_owned(std::vector<ValueType>&)':
./../numpy_bind.hh:78:34: error: 'NPY_ARRAY_ALIGNED' was not declared in
this scope
./../numpy_bind.hh:78:54: error: 'NPY_ARRAY_C_CONTIGUOUS' was not
declared in this scope
./../numpy_bind.hh:79:25: error: 'NPY_ARRAY_OWNDATA' was not declared in
this scope
./../numpy_bind.hh:79:45: error: 'NPY_ARRAY_WRITEABLE' was not declared
in this scope
./../numpy_bind.hh:79:64: error: there are no arguments to
'PyArray_ENABLEFLAGS' that depend on a template parameter, so a
declaration of 'PyArray_ENABLEFLAGS' must be available [-fpermissive]
./../numpy_bind.hh:79:64: note: (if you use '-fpermissive', G++ will
accept your code, but allowing the use of an undeclared name is deprecated)
./../numpy_bind.hh: In function 'boost::python::api::object
wrap_vector_not_owned(std::vector<ValueType>&)':
./../numpy_bind.hh:96:33: error: 'NPY_ARRAY_ALIGNED' was not declared in
this scope
./../numpy_bind.hh:96:53: error: 'NPY_ARRAY_C_CONTIGUOUS' was not
declared in this scope
./../numpy_bind.hh:97:25: error: 'NPY_ARRAY_WRITEABLE' was not declared
in this scope
./../numpy_bind.hh:97:44: error: there are no arguments to
'PyArray_ENABLEFLAGS' that depend on a template parameter, so a
declaration of 'PyArray_ENABLEFLAGS' must be available [-fpermissive]
./../numpy_bind.hh: In function 'boost::python::api::object
wrap_multi_array_owned(boost::multi_array<ValueType, Dim>&)':
./../numpy_bind.hh:116:34: error: 'NPY_ARRAY_ALIGNED' was not declared
in this scope
./../numpy_bind.hh:116:54: error: 'NPY_ARRAY_C_CONTIGUOUS' was not
declared in this scope
./../numpy_bind.hh:117:25: error: 'NPY_ARRAY_OWNDATA' was not declared
in this scope
./../numpy_bind.hh:117:45: error: 'NPY_ARRAY_WRITEABLE' was not declared
in this scope
./../numpy_bind.hh:117:64: error: there are no arguments to
'PyArray_ENABLEFLAGS' that depend on a template parameter, so a
declaration of 'PyArray_ENABLEFLAGS' must be available [-fpermissive]
./../numpy_bind.hh: In function 'boost::python::api::object
wrap_multi_array_not_owned(boost::multi_array<ValueType, Dim>&)':
./../numpy_bind.hh:130:34: error: 'NPY_ARRAY_ALIGNED' was not declared
in this scope
./../numpy_bind.hh:130:54: error: 'NPY_ARRAY_C_CONTIGUOUS' was not
declared in this scope
./../numpy_bind.hh:131:25: error: 'NPY_ARRAY_WRITEABLE' was not declared
in this scope
./../numpy_bind.hh:131:44: error: there are no arguments to
'PyArray_ENABLEFLAGS' that depend on a template parameter, so a
declaration of 'PyArray_ENABLEFLAGS' must be available [-fpermissive]
./../numpy_bind.hh: In function 'boost::multi_array_ref<ValueType, dim>
get_array(boost::python::api::object)':
./../numpy_bind.hh:173:30: error: 'NPY_ARRAY_C_CONTIGUOUS' was not
declared in this scope
At global scope:
cc1plus: warning: unrecognized command line option
"-Wno-unused-local-typedefs" [enabled by default]
make[4]: *** [graph_closeness.lo] Error 1
make[4]: Leaving directory
`/home/markh/Code/graph-tool-2.2.25/src/graph/centrality'
make[3]: *** [all-recursive] Error 1
make[3]: Leaving directory `/home/markh/Code/graph-tool-2.2.25/src/graph'
make[2]: *** [all-recursive] Error 1
make[2]: Leaving directory `/home/markh/Code/graph-tool-2.2.25/src'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/markh/Code/graph-tool-2.2.25'
make: *** [all] Error 2
$

Perhaps your build machine has a numpy 1.7 installed in
/usr/local/lib/python2.7/dist-packages (or elsewhere) in addition to the
standard Precise package in usr/lib/python2.7/dist-packages, and your
graph-tool ./configure is finding that one to compile against?

Cheers,

Mark.

5

I am trying to use purge_vertices() with a vertex filter to efficiently
remove vertices from a graph. I am seeing different behavior in version
2.2.18 and 2.2.25, both as prebuilt packages for Ubuntu precise
(current
"graph-tool" and "python-graph-tool" packages, respectively)

[snip]

Perhaps your build machine has a numpy 1.7 installed in
/usr/local/lib/python2.7/dist-packages (or elsewhere) in addition to the
standard Precise package in usr/lib/python2.7/dist-packages, and your
graph-tool ./configure is finding that one to compile against?

Update... building numpy 1.7.1 from source and installing in default
(/usr/local/lib) location is enough to get my test code to work with
your python-graph-tool 2.2.25-1 deb package for Precise.

I've left the Precise numpy 1.6.1 installed in /usr/lib so as not to
wrestle with other deb package dependencies. The py environment finds
the /usr/local/lib (1.7.1) one first.

Cheers,

Mark.

6

Hi,

Please try the new debian / ubuntu packaged for the new version. This
issue should be fixed now.

Cheers,
Tiago