[SciPy-user] Ubuntu installation requirements, and wetting my feet.

[William Hunter]

On 16/09/06, Stephen Kelly <steveire at gmail.com> wrote:
> Hi. I emailed this list a while ago about possible subject matter for a
> final year project in engineering
> (http://news.gmane.org/find-root.php?group=gmane.comp.python.scientific.user&article=8498
> ). There was no project resulting directly from the mailing list, but I have
> now started a project concerning the correction of xray diffraction data
> using python. It will mainly be concerned with designing a good user
> interface, but could also involve interpolation and plotting, so I want to
> see what I can do with scipy in a reasonable time to do this.
>
> The installation instructions on the wiki are a bit confusing. I'm inclined
> to get the optimised packages, so:
>
> > To build SciPy using optimized lapack and blas on the ubuntu system you
> should install atlas-3dnow-dev or atlas-sse2-dev or atlas-sse3-dev depending
> on your system.
>
> But those packages are not available in the repos available to me. There are
> others with similar names, but I'm not certain which to choose. What
> characteristics of my system will affect this?
>
> > atlas2            atlas2-sse-dev    atlas3-headers    atlas-dev
> > atlas2-3dnow-dev  atlas3-3dnow      atlas3-sse        atlas-doc
> > atlas2-base-dev   atlas3-3dnow-dev  atlas3-sse2       atlas-test
> > atlas2-dev        atlas3-base       atlas3-sse2-dev
> > atlas2-headers    atlas3-base-dev   atlas3-sse-dev
> > atlas2-sse2-dev   atlas3-doc        atlas3-test

On Dapper, I used atlas3-base-dev: scipy.test() and numpy.test()
didn't report any errors after compilation. If I remember correctly it
puts the libraries in /usr/lib/atlas/, and you have to specify this in
the site.cfg file.

>
> On an unrelated matter, I'm already using the numpy deb package provided at
> astraw.com, and as a learning excercise started making a mechanics of
> materials module for tensor calculations. I'm not sure if numpy/scipy
> already provides such functions, so I don't know if this is reinventing the
> wheel. To calculate the sencond scalar invariant, I thought an easy way
> would be to calculate the trace of the adjoint of the stress tensor I
> define. When I went to do it though, I couldn't find any functions in numpy
> to get an adjoint of a matrix, or get cofactors. Are these functions
> available in numpy or scipy? I've got it working now by multiplying the
> inverse of the matrix by the determinant, but if there's a better way I'd
> like to know. Also, I'd like to assess the validity of such a stress matrix
> by checking if it's symmetric, but I haven't seen any way to do that either.
> I thought there might be a foo.isSymmetric() bool, but there doesn't seem to
> be, so I tried doing a comparison, if foo == foo.T: etc., but comparison of
> matrices doesn't seem to work either. Is the only way to compare them to
> loop over them using regular python methods?

How about this, which should == 0:
abs(A-A.T).max()

> Thanks for help on this. I may be seen a bit more on this mailing list once
> I get into the computation parts of this project. Is there any interest in
> using #scipy on freenode?
>
> Thanks,
>
> Steve
>
> _______________________________________________
> SciPy-user mailing list
> SciPy-user at scipy.org
> http://projects.scipy.org/mailman/listinfo/scipy-user