Scipy install on linux: cannot find object lapack sgges_

I am trying to install scipy and am running into problems. Comparing my results on the FAQ page, it looks like my ATLAS is incomplete. The page says:

The LAPACK library provided by ATLAS is incomplete You will notice this when you receive import errors like ImportError: ... / flapack.so: undefined symbol: sgesdd_ To make sure NumPy / SciPy is built against full LAPACK, check the size of the liblapack.a file - it should be around 6MB. The location of liblapack.a is shown by running python numpy / distutils / system_info.py lapack To fix: follow the instructions at http://math-atlas.sourceforge.net/errata.html#completelp Build the complete LAPACK library to build the complete liblapack.a. Then copy liblapack.a in the same location where libatlas.a is installed and try again with scipy build.

The actual object I'm missing is "sgges_". I've tried two solutions.

1) I followed the ATLAS instructions to create a complete LAPACK assembly, rev. 3.4.2. (BTW the LAPACK received is about 10MB, more than 6MB in the FAQ.)

2) I found a precompiled binary for LAPACK-rev. unclear - and just downloaded it and dropped it in place. (It's close to 6MB in size.)

I am getting the same problem regardless.

"nm" shows sgges_ defined in liblapack.a as day. System_info.py finds my shortcut in / usr / local / lib.

Is this a critical failure? Is there anything else?

By the way, I am running python 2.7 on a 64 bit CentOs 5.x machine and gfortran is my FORTRAN compiler.

Thank.

+3


source to share


1 answer


I had to compromise scipy from scratch on a slackware 13 linux distro which is a very basic linux system; I have listed all the steps and results here:

SCIPY install from source on slackware 13



In my installation, I was trying to install dynamic libraries to the local / home / andrew 3 / .local / lib directory because it was not a system installation. Similar errors in what you are encountering were caused by the ldd linker not being able to find the libraries.

You might be able to isolate your problem by replicating my steps and seeing where it fails, if anywhere.

+1


source







All Articles