No module named _umfpack
Hi- I'm building numpy and scipy according to the directions on the web site (Steve Baum's doc). Everything works fine and compiles properly. When I load scipy:
import scipy
I'm told that there is no module named '_umfpack'. Looking through the archives I see a recent post where someone said that setup.py for linsolve was screwy and was building __umfpack instead of _umfpack. What the post didn't mention is how to fix this error. What needs to be modified? Thanks! Ron
Ron Kneusel wrote:
Hi-
I'm building numpy and scipy according to the directions on the web site (Steve Baum's doc). Everything works fine and compiles properly. When I load scipy:
import scipy
I'm told that there is no module named '_umfpack'.
Looking through the archives I see a recent post where someone said that setup.py for linsolve was screwy and was building __umfpack instead of _umfpack.
What the post didn't mention is how to fix this error. What needs to be modified?
If you actually do want to build the UMFPACK bindings, then you can edit Lib/linsolve/setup.py to replace "__umfpack" with "_umfpack". There are two bugs open and assigned to Robert Cimrman. I hope he can find the time to attend to them soon. http://projects.scipy.org/scipy/scipy/ticket/190 http://projects.scipy.org/scipy/scipy/ticket/191 -- Robert Kern robert.kern@gmail.com "I have come to believe that the whole world is an enigma, a harmless enigma that is made terrible by our own mad attempt to interpret it as though it had an underlying truth." -- Umberto Eco
Robert Kern wrote:
Ron Kneusel wrote:
Hi-
I'm building numpy and scipy according to the directions on the web site (Steve Baum's doc). Everything works fine and compiles properly. When I load scipy:
import scipy
I'm told that there is no module named '_umfpack'.
Looking through the archives I see a recent post where someone said that setup.py for linsolve was screwy and was building __umfpack instead of _umfpack.
What the post didn't mention is how to fix this error. What needs to be modified?
If you actually do want to build the UMFPACK bindings, then you can edit Lib/linsolve/setup.py to replace "__umfpack" with "_umfpack".
you mean Lib/linsolve/umfpack/setup.py? in Lib/linsolve/setup.py there is just config.add_subpackage('umfpack')...
There are two bugs open and assigned to Robert Cimrman. I hope he can find the time to attend to them soon.
Sorry, I missed them...
Well, it should work as it is now - the top-level umfpack module is 'umfpack.py', the swig-generated modules are '_umfpack.py' and '__umfpack.so', so please do not change this. UMFPACK is optional, and it is not included in the scipy SVN, so you have to install it yourself from Tim Davis' homepage if you want (the version 4.4). Then just add [amd] and [umfpack] sections to your site.cfg, as it is described in numpy/site.cfg.example
Done in the SVN. cheers, r.
Robert Cimrman wrote:
Robert Kern wrote:
If you actually do want to build the UMFPACK bindings, then you can edit Lib/linsolve/setup.py to replace "__umfpack" with "_umfpack".
you mean Lib/linsolve/umfpack/setup.py? in Lib/linsolve/setup.py there is just config.add_subpackage('umfpack')...
Umm, yes. That's what I meant.
There are two bugs open and assigned to Robert Cimrman. I hope he can find the time to attend to them soon.
Sorry, I missed them...
I'm working on setting up email notification.
Well, it should work as it is now - the top-level umfpack module is 'umfpack.py', the swig-generated modules are '_umfpack.py' and '__umfpack.so', so please do not change this.
Yes, you're right. I realize now that I wasn't building umfpack at all because I don't have UMFPACK installed. And I probably won't until there is a source distribution of it that doesn't require me to hack makefiles that attempt to build MEX files.
UMFPACK is optional, and it is not included in the scipy SVN, so you have to install it yourself from Tim Davis' homepage if you want (the version 4.4). Then just add [amd] and [umfpack] sections to your site.cfg, as it is described in numpy/site.cfg.example
Done in the SVN.
Thank you! -- Robert Kern robert.kern@gmail.com "I have come to believe that the whole world is an enigma, a harmless enigma that is made terrible by our own mad attempt to interpret it as though it had an underlying truth." -- Umberto Eco
Robert Cimrman wrote:
Done in the SVN.
whoops, it was not fixed. I was too quick and moreover did not counted with the numpy import machinery. I hope this time I got it right... r.
On Tuesday 18 April 2006 04:03, Robert Cimrman wrote:
UMFPACK is optional, and it is not included in the scipy SVN, so you have to install it yourself from Tim Davis' homepage if you want (the version 4.4). Then just add [amd] and [umfpack] sections to your site.cfg, as it is described in numpy/site.cfg.example
For any gentoo users out there, a umfpack ebuild is available here: http://gentooscience.org/browser/overlay/sci-libs/umfpack. I was able to build scipy with umfpack support on an 64bit Athlon machine using this ebuild, which provides version 4.6 with support to build shared libraries. No changes to site.cfg were necessary. scipy.test() and scipy.linsolve.umfpack.test() were both successful. Darren
Darren Dale wrote:
On Tuesday 18 April 2006 04:03, Robert Cimrman wrote:
UMFPACK is optional, and it is not included in the scipy SVN, so you have to install it yourself from Tim Davis' homepage if you want (the version 4.4). Then just add [amd] and [umfpack] sections to your site.cfg, as it is described in numpy/site.cfg.example
For any gentoo users out there, a umfpack ebuild is available here: http://gentooscience.org/browser/overlay/sci-libs/umfpack.
I was able to build scipy with umfpack support on an 64bit Athlon machine using this ebuild, which provides version 4.6 with support to build shared libraries. No changes to site.cfg were necessary.
scipy.test() and scipy.linsolve.umfpack.test() were both successful.
great news! I will try it immediately :) r.
On Tuesday 18 April 2006 08:35, Robert Cimrman wrote:
Darren Dale wrote:
On Tuesday 18 April 2006 04:03, Robert Cimrman wrote:
UMFPACK is optional, and it is not included in the scipy SVN, so you have to install it yourself from Tim Davis' homepage if you want (the version 4.4). Then just add [amd] and [umfpack] sections to your site.cfg, as it is described in numpy/site.cfg.example
For any gentoo users out there, a umfpack ebuild is available here: http://gentooscience.org/browser/overlay/sci-libs/umfpack.
I was able to build scipy with umfpack support on an 64bit Athlon machine using this ebuild, which provides version 4.6 with support to build shared libraries. No changes to site.cfg were necessary.
scipy.test() and scipy.linsolve.umfpack.test() were both successful.
great news! I will try it immediately :)
Oh, let me make clear, the ebuild provides support to build shared libraries, not the umfpack-4.6 package itself. More information here: http://bugs.gentoo.org/show_bug.cgi?id=40255 Darren
Darren Dale wrote:
On Tuesday 18 April 2006 08:35, Robert Cimrman wrote:
Darren Dale wrote:
For any gentoo users out there, a umfpack ebuild is available here: http://gentooscience.org/browser/overlay/sci-libs/umfpack.
I was able to build scipy with umfpack support on an 64bit Athlon machine using this ebuild, which provides version 4.6 with support to build shared libraries. No changes to site.cfg were necessary.
scipy.test() and scipy.linsolve.umfpack.test() were both successful.
great news! I will try it immediately :)
Oh, let me make clear, the ebuild provides support to build shared libraries, not the umfpack-4.6 package itself. More information here: http://bugs.gentoo.org/show_bug.cgi?id=40255
well, it works well for me, moreover I got about 30% speed-up w.r.t the manually built version 4.4 for a simple problem I had at hand. Conclusion: the wrappers seem to work also with the version 4.6; still there might be some extra functionality in 4.6 worth exposing, I will check it out when time permits. r.
Ok, I finally got umfpack built and installed and I get no errors now when I do:
import scipy
Thanks to all those who helped out! However, the gplt and xplt packages do not appear to be installed. What additional modules are required to get them working? I'm using Fedora Core 5. Ron
Ron Kneusel wrote:
Ok, I finally got umfpack built and installed and I get no errors now when I do:
import scipy
Thanks to all those who helped out!
However, the gplt and xplt packages do not appear to be installed. What additional modules are required to get them working? I'm using Fedora Core 5.
Ron
_______________________________________________ SciPy-user mailing list SciPy-user@scipy.net http://www.scipy.net/mailman/listinfo/scipy-user
# Gist-based plotting library for X11 #config.add_subpackage('xplt') It is disabled. See the setup.py file in scipy/Lib/sandbox Nils
Ron Kneusel wrote:
Ok, I finally got umfpack built and installed and I get no errors now when I do:
import scipy
Thanks to all those who helped out!
However, the gplt and xplt packages do not appear to be installed.
They have been removed from the main package. If you really want to use them, they are in the sandbox. You can edit Lib/sandbox/setup.py to enable building them. They can be then accessed as scipy.sandbox.gplt, scipy.sandbox.xplt. They are not supported anymore. gplt, at least, hasn't been ported to numpy at all. The general recommendation is to use matplotlib instead.
What additional modules are required to get them working? I'm using Fedora Core 5.
Each should work alone. Except that gnuplot itself must be installed for gplt. -- Robert Kern robert.kern@gmail.com "I have come to believe that the whole world is an enigma, a harmless enigma that is made terrible by our own mad attempt to interpret it as though it had an underlying truth." -- Umberto Eco
participants (5)
-
Darren Dale -
Nils Wagner -
Robert Cimrman -
Robert Kern -
Ron Kneusel