Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Already on GitHub? Sign in to your account

TST: sparse/arpack: loosen test tolerances in "numerically bad" cases #159

Closed
wants to merge 1 commit into
from

Conversation

Projects
None yet
3 participants
Owner

pv commented Feb 14, 2012

There are a couple of tests in the ARPACK test suite that still fail on OSX [1].

The failing tests have both following properties: (i) single precision, (ii) using iterative inverse in shift-invert mode. The output in the user reports seems to be between garbage and inaccurate results. I'm guessing the reason is just that the iterative inverse may be too inaccurate in single precision to make the outer eigenvalue computation work well.

The alternatives are either to make the double-precision forcing to include also the iterative inverse, or just disable these tests. The former might be a better solution, but here's a patch for the latter...

[1] http://permalink.gmane.org/gmane.comp.python.scientific.user/31057

Well done. Just tested this on 10.7.3 64bit with clang/++.

    Ran 5103 tests in 46.545s
    OK (KNOWNFAIL=12, SKIP=36)

I vote to include this into 0.10.1.

Owner

rgommers commented Feb 16, 2012

Since computing in double does seem like a better solution, how about just putting this in 0.10.1 and not in master?

Owner

rgommers commented Feb 16, 2012

Oh never mind, your PR is for 0.10.x

Owner

rgommers commented Feb 18, 2012

Merged, thanks Pauli and Samuel. Opened http://projects.scipy.org/scipy/ticket/1599 to fix this for 0.11.

@rgommers rgommers closed this Feb 18, 2012

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment