Disable qz sort keyword #303

Merged
merged 2 commits into from Sep 13, 2012

Conversation

Projects
None yet
5 participants
@rgommers
Owner

rgommers commented Sep 8, 2012

See ticket 1717 (http://projects.scipy.org/scipy/ticket/1717) for details. Since I haven't been able to find the issue yet and can't spend more time on this, disabling the sort kw is the most reasonable option imho. This function is new in 0.11.0, so backwards compatibility is not an issue.

This PR also includes some docstring fixes and adds an example.

@pv

This comment has been minimized.

Show comment Hide comment
@pv

pv Sep 8, 2012

Owner

+1, this is the easier way around the release blocker.

There's something strange going on with this in any case, since cgohlke reported also MKL binaries failing the tests (apparently, with something more than just numerical precision issues).

Owner

pv commented Sep 8, 2012

+1, this is the easier way around the release blocker.

There's something strange going on with this in any case, since cgohlke reported also MKL binaries failing the tests (apparently, with something more than just numerical precision issues).

rgommers added a commit that referenced this pull request Sep 13, 2012

@rgommers rgommers merged commit 8651e2e into scipy:master Sep 13, 2012

@rgommers

This comment has been minimized.

Show comment Hide comment
@rgommers

rgommers Sep 13, 2012

Owner

OK, merged.

Owner

rgommers commented Sep 13, 2012

OK, merged.

@ricardomayerb

This comment has been minimized.

Show comment Hide comment
@ricardomayerb

ricardomayerb Nov 25, 2013

Mmm not sure if this is the right place to ask but ... are the any news regarding this issue? One particular form of sorting () is required for a relatively common algorithm used in (macro)economics.

I'm currently writing some notebook examples to convince people that python is a viable alternative to matlab when doing macroeconomics, but this feature is somewhat critical. My workaround has been calling an R function that does this work, but rpy2 isn't the friendliest of packages to install on Windows, so it would be much, much nicer to have it all in python.

thanks in advance
R

(sadly , I have no experience in interfacing with LAPACK, or I would volunteer to work on this particular issue)

Mmm not sure if this is the right place to ask but ... are the any news regarding this issue? One particular form of sorting () is required for a relatively common algorithm used in (macro)economics.

I'm currently writing some notebook examples to convince people that python is a viable alternative to matlab when doing macroeconomics, but this feature is somewhat critical. My workaround has been calling an R function that does this work, but rpy2 isn't the friendliest of packages to install on Windows, so it would be much, much nicer to have it all in python.

thanks in advance
R

(sadly , I have no experience in interfacing with LAPACK, or I would volunteer to work on this particular issue)

@ricardomayerb

This comment has been minimized.

Show comment Hide comment
@ricardomayerb

ricardomayerb Nov 25, 2013

BTW, by that "particular form of sorting" I was referring to the "'iuc" option in linalg.qz documentation.

BTW, by that "particular form of sorting" I was referring to the "'iuc" option in linalg.qz documentation.

@jseabold

This comment has been minimized.

Show comment Hide comment
@jseabold

jseabold Nov 25, 2013

Contributor

Yes, I added qz and enabled the sort function originally for macro applications (replacing matlab), but unfortunately I have no idea why this could be the case that it isn't working for windows.

Contributor

jseabold commented Nov 25, 2013

Yes, I added qz and enabled the sort function originally for macro applications (replacing matlab), but unfortunately I have no idea why this could be the case that it isn't working for windows.

@jseabold

This comment has been minimized.

Show comment Hide comment
@jseabold

jseabold Nov 25, 2013

Contributor

I'm also not set up to build numpy on windows to do any more debugging of this. Is the only easy way to build numpy on windows with mingw-32 with the old ATLAS binaries from numpy vendor?

Contributor

jseabold commented Nov 25, 2013

I'm also not set up to build numpy on windows to do any more debugging of this. Is the only easy way to build numpy on windows with mingw-32 with the old ATLAS binaries from numpy vendor?

@ricardomayerb

This comment has been minimized.

Show comment Hide comment
@ricardomayerb

ricardomayerb Nov 25, 2013

Bummer. Is it too naive to believe that --given that works well in windows-- looking how R is built and how packages geigen and QZ interface with lapack would give us any useful hint?

Bummer. Is it too naive to believe that --given that works well in windows-- looking how R is built and how packages geigen and QZ interface with lapack would give us any useful hint?

@jseabold

This comment has been minimized.

Show comment Hide comment
@jseabold

jseabold Nov 25, 2013

Contributor

Maybe. My suspicion at the time was it was simply the LAPACK version had a bug, though some sort of problem showed up on MKL since then as well. Unless one of the windows compilation / C expert wizards weighs in, I don't see myself solving this though.

Contributor

jseabold commented Nov 25, 2013

Maybe. My suspicion at the time was it was simply the LAPACK version had a bug, though some sort of problem showed up on MKL since then as well. Unless one of the windows compilation / C expert wizards weighs in, I don't see myself solving this though.

@ricardomayerb

This comment has been minimized.

Show comment Hide comment
@ricardomayerb

ricardomayerb Nov 25, 2013

I'll keep my fingers crossed for the experts to weigh in, using rpy2 from the time being. Thanks, @jseabold .

I'll keep my fingers crossed for the experts to weigh in, using rpy2 from the time being. Thanks, @jseabold .

@argriffing

This comment has been minimized.

Show comment Hide comment
@argriffing

argriffing Nov 25, 2013

Contributor

looking how R is built

If I remember, we aren't allowed to look too closely at R when developing scipy, because of licenses.

Contributor

argriffing commented Nov 25, 2013

looking how R is built

If I remember, we aren't allowed to look too closely at R when developing scipy, because of licenses.

@jseabold

This comment has been minimized.

Show comment Hide comment
@jseabold

jseabold Nov 25, 2013

Contributor

The QZ package is MPL, which I think is BSD-compatible, but in any event I don't think looking at their C wrapper code and trying to square f2py with some weird nuance could cause any problems. It would be a fishing expedition though.

Contributor

jseabold commented Nov 25, 2013

The QZ package is MPL, which I think is BSD-compatible, but in any event I don't think looking at their C wrapper code and trying to square f2py with some weird nuance could cause any problems. It would be a fishing expedition though.

@jseabold

This comment has been minimized.

Show comment Hide comment
@jseabold

jseabold Nov 25, 2013

Contributor

Ah. QZ ::ordqz just avoids using the selctg function. It does the reordering after using [zd]tgsen or [zd]trsen as appropriate.

I can try to look at this sometime this week.

http://www.netlib.org/lapack/explore-html/d0/d58/dtgsen_8f.html

Contributor

jseabold commented Nov 25, 2013

Ah. QZ ::ordqz just avoids using the selctg function. It does the reordering after using [zd]tgsen or [zd]trsen as appropriate.

I can try to look at this sometime this week.

http://www.netlib.org/lapack/explore-html/d0/d58/dtgsen_8f.html

@ricardomayerb

This comment has been minimized.

Show comment Hide comment
@ricardomayerb

ricardomayerb Nov 25, 2013

That would be awesome. Incidentally, the dtsegen-after-dgegs strategy seems to be the same used by paul klein in http://people.bu.edu/rking/REmodels/k02jedc.pdf

That would be awesome. Incidentally, the dtsegen-after-dgegs strategy seems to be the same used by paul klein in http://people.bu.edu/rking/REmodels/k02jedc.pdf

@rgommers rgommers deleted the rgommers:disable-qz-sort branch Nov 25, 2013

@jseabold

This comment has been minimized.

Show comment Hide comment
@jseabold

jseabold Dec 2, 2013

Contributor

See #3107. Another set of eyes would be welcome.

Contributor

jseabold commented Dec 2, 2013

See #3107. Another set of eyes would be welcome.

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