I've added the test and fixed inv(). I don't think we need the
mod to PDL::Fit::Polynomial. This does point out that PDL really
could use a clean, supported linear algebra framework rather
that the roll-our-own patchwork...
On 3/5/2015 11:38, Chris Marshall wrote:
A couple of points-
(1) please open a bug and add a test for the inv() failure
(2) modify PDL::Fit::Polynomial to use matinv() in preference to inv()
*if available*
(3) I'll take a look at inv(), I remember it threading correctly. It
definitely needs fixing.
The problem may be a result of something changed elsewhere in
perl/PDL.
Thanks,
Chris
On 3/4/2015 16:04, Karl Glazebrook wrote:
Hi Derek
Yes it seems the switch from PDL::Slatec::matinv to
PDL::MatrixOps::inv was at fault! Clearly the latter does not thread
correctly.
Further investigation suggests that for fitpoly1d($x,$y) with 2D $y
then it fails when
- $x is absent
- $x is 2D
But it WORKS when $x is 1D, with $y 2D which is probably the most
common case and why it was not caught to now! (That and me not
writing a test)
Who made the change? Would anyone mind if we switched it back? The
alternative is to make inv() thread correctly. I’d kind of like to
correct this in 2.008. It’s my module :-)
Karl
On 4 Mar 2015, at 8:05 am, Derek Lamb <de...@boulder.swri.edu
<mailto:de...@boulder.swri.edu>> wrote:
Hi Karl,
Maybe if we had a test for threaded fitpoly1d we would have caught
this 4.5 years ago!
Looks like the problem is with the change from PDL::Slatec::matinv
to PDL::MatrixOps::inv [8adb0b]
<http://sourceforge.net/p/pdl/code/ci/8adb0b9e6e88ded9153563509794ed253d4f50e1/>.
Running your example through the debugger I get:
DB<34> p inv($C)
[
[ 0.168 -0.064]
[ -0.32 0.16]
]
DB<35> p matinv($C)
[
[
[ 0.6 -0.4]
[-0.4 0.4]
]
[
[ 0.6 -0.4]
[-0.4 0.4]
]
]
Wolfram Alpha suggests that matinv is correct. When running your
single-dimension example through, inv and matinv return identical
results.
My guess is PDL::MatrixOps::inv is not threadable. It certainly
isn't advertised as such in the docs, though other subs in MatrixOps
are advertised as such.
cheers,
Derek
On Mar 4, 2015, at 12:59 AM, Karl Glazebrook
<karlglazebr...@mac.com <mailto:karlglazebr...@mac.com>> wrote:
fitpoly1d() no longer seems to thread over extra dimensions:
pdl> $x = pdl( [101,103,104,102,109] )
pdl> p $x
[101 103 104 102 109]
pdl> ($xf, $c) = fitpoly1d($x, 2)
pdl> p $xf
[100.8 102.3 103.8 105.3 106.8]
pdl>
pdl> $x = pdl( [101,103,104,102,109],[101,103,104,102,109] )
pdl> p $x
[
[101 103 104 102 109]
[101 103 104 102 109]
]
pdl> ($xf, $c) = fitpoly1d($x, 2)
pdl> p $xf # WTF?!
[
[ 53.496 12.576 -28.344 -69.264 -110.184]
[ 53.496 12.576 -28.344 -69.264 -110.184]
]
I took a look at the Polynomial.pm and could no longer figure it
out! I swear it used to work - anyone have any ideas?
Karl
------------------------------------------------------------------------------
Dive into the World of Parallel Programming The Go Parallel
Website, sponsored
by Intel and developed in partnership with Slashdot Media, is your
hub for all
things parallel software development, from weekly thought
leadership blogs to
news, videos, case studies, tutorials and more. Take a look and
join the
conversation now.
http://goparallel.sourceforge.net/_______________________________________________
pdl-devel mailing list
pdl-devel@lists.sourceforge.net
<mailto:pdl-devel@lists.sourceforge.net>
https://lists.sourceforge.net/lists/listinfo/pdl-devel
------------------------------------------------------------------------------
Dive into the World of Parallel Programming The Go Parallel Website, sponsored
by Intel and developed in partnership with Slashdot Media, is your hub for all
things parallel software development, from weekly thought leadership blogs to
news, videos, case studies, tutorials and more. Take a look and join the
conversation now.http://goparallel.sourceforge.net/
_______________________________________________
pdl-devel mailing list
pdl-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/pdl-devel
------------------------------------------------------------------------------
Dive into the World of Parallel Programming The Go Parallel Website, sponsored
by Intel and developed in partnership with Slashdot Media, is your hub for all
things parallel software development, from weekly thought leadership blogs to
news, videos, case studies, tutorials and more. Take a look and join the
conversation now. http://goparallel.sourceforge.net/
_______________________________________________
pdl-devel mailing list
pdl-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/pdl-devel
