On 10/1/2011 2:36 PM, Joel Berger wrote:
Thanks Chris,
I missed that announcement as I wasn't on the mailing list at that
point in time. I like the idea of improving the IO, it makes PDL more
usable for day-to-day actions.
I needed the methods I present in PDL::Util because I needed them, but
wanted to contribute them back to the community. They seemed natural
to be outside the Core since they use Perl level actions rather than
PP (which is still rather terrifying to me). You are more than welcome
If you look at the source, much of PDL functionality
is implemented from perl with the performance critical
parts implemented in PDL::PP.
to implement something like `unroll` in the core, it will not offend
Thanks, but it would more likely be something like
moving your unroll into the PDL distribution with
any necessary massaging along the way. :-)
me, quite the contrary. If instead, you would like to pass other
little functions to me for PDL::Util (my inspiration was the
List::MoreUtils module, or more specifically its mission) for non-Core
but useful PDL routines, I would be happy to include them, and thereby
help keep the Core clean.
A bigger issue for PDL as a distribution is not to
add external dependencies, especially ones that are
not available for all the main perl/PDL platforms:
unix/linux/*bsd/solaris, macosx, and windows.
Did you ever get anywhere with the Moose+PDL stuff?
--Chris
On Sat, Oct 1, 2011 at 11:54 AM, chm<[email protected]> wrote:
Hi Joel-
The new (in PDL git) support for 2-D pdls in
wcols appears to match the convention used by
your PDL::Util::export2d routine as well.
The current rcols support for 2-D arrays was
the result of some performance optimization
work last spring which lead to the new functionality
as well. The announcement thread is here:
http://mailman.jach.hawaii.edu/pipermail/pdl-porters/2010-March/001884.html
The idea was to make rcols more useful for
mixed string/perl column data and numeric/pdl
data files. Reworking wcols to work the same
has been on the to-do list for a while... just
no time to get to it.
It turns out that the rcols implementation
evolved a backwards convention for dimensions
compared with wcols.
On 10/1/2011 11:40 AM, Joel Berger wrote:
what do you mean "Added support for 2D pdls"? do you mean when you
call wcols on a 2D pdl it will write it out automatically without
having to slice it first? I've always wondered why that simple
operation took so much work. Its been part of my (renamed) PDL::Utils
module, but I guess I would rather it be in the core.
There are *many* enhancements to PDL that have
been proposed, it just takes volunteers to make
things happen. If I had known of your interest
in the problem, I would have suggested the wcols
fix directly.
I think it would make sense to fold unroll() into
the PDL core at some point. I'm just not sure
the best way to do so. Given that PDL users
actually do "real work" with the module one
requirement of new additions to PDL core is that
they *not* break back compatibility if possible
or reasonable (e.g., maintaining broken/buggy
functionality going forward is not reasonable).
As a project, I would like to see a clean way
to coordinate and involve external PDL development
activity with the PDL core evolution. No good
ideas so far but for 2.4.10 I would like to add
a list of external PDL modules to the available
docs. Searching for PDL on http://search.cpan.org
doesn't give a vey helpful list. I only
accidentally discovered your PDL::Util and its
predecessor module.
Cheers,
Chris
On Sat, Oct 1, 2011 at 8:58 AM, chm<[email protected]> wrote:
rcols/wcols users:
I just added support to wcols for 2-D pdls and
discovered that it was not the inverse operation
to rcols---which I think it should be.
The rcols implementation uses dim(0) for input
columns for the 1-D inputs, but dim(1) for the
multi-column inputs.
The wcols implementation uses dog on the 2-D
input pdls and then does the original wcols
output so effectively, both cases use dim(0).
My thought is to modify the rcols 2-D support
so that the columns in the file always to to
dim(0) in the pdls. That would be consistent
with wcols usage and with the general idea of
things happening on dim(0) and threading on
higher dims---not that we thread with [rw]cols
at this time.
This also makes the rcols consistent with the
original get_data() routine by Maggie which
motivated the development of 2-D support.
Cheers,
Chris
_______________________________________________
Perldl mailing list
[email protected]
http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
