Go Blas implementation in biogo.matrix about blas HOT 7 CLOSED

Thanks for the input, though the main developer of this package is Dan Kortschak, who is also the developer of biogo. He's porting it here to make it more accessible.

On Nov 1, 2013, at 12:10 PM, Malcolm [email protected] wrote:

I want to point out that there's a Go wrapper for BLAS already written. It's in biogo.matrix. Here's my SO on it: http://stackoverflow.com/questions/16536029/go-matrix-library/19418804#19418804

—
Reply to this email directly or view it on GitHub.

from blas.

Awesome! I just discovered biogo.matrix a few weeks ago and wanted this to happen. It looks like he just got started on this, right? Any idea when it'll be finished or should I just ask Dan?

from blas.

I'm not exactly sure what you're asking, but it's probably better to move the discussion to the dev list https://groups.google.com/forum/#!forum/gonum-dev . Loosely, we think it works, though we are still working on extensive testing. We are also still working on the API, so it is not stable. If you're okay with that, we would love more people using and helping to develop the tool suite.

from blas.

Thanks for pointing me to the dev list. I'll mark this as closed.

from blas.

Malcolm,

Are there particular things that you think are missing. mat64 is a continuation of biogo.matrix with design improvements added in and gonum's cblas is the actively developed project. A number of things have been left out intentionally (sparse for example) and the API is simpler and (hopefully) more Go like.

A vector type has been asked for by David Neumann and there is some of that included, but I have stalled on that, for reasons. Stack and Augment need implementation, but this is complicated by a design decision to support row and column major formats. I'm sort of waiting on noise from the nd slice proposals before I make a lot more effort there unless that drags on.

As Brendan pointed out testing is a need. Also, an implementation of at least the D BLAS routines in Go that matches blas.Float64 would be good. I'm not going to do that.

Dan

from blas.

Hi Dan,

My own personal desires are to stop using java and start using go for my
own research in machine learning. I'm mainly interested in having a speedy
and reliably matrix library so that I can implement speedy and reliable
classifiers :) I'm quite enamored with go and would be happy to donate a
little time to writing tests and try to get the D BLAS routines working.
I'm a grad student, so time is tight, but I really want to see go break
into some python territory here (looking at numpy/scipy).

• Malcolm

On Fri, Nov 1, 2013 at 3:53 PM, kortschak [email protected] wrote:

Malcolm,

Are there particular things that you think are missing. mat64 is a
continuation of biogo.matrix with design improvements added in and gonum's
cblas is the actively developed project. A number of things have been left
out intentionally (sparse for example) and the API is simpler and
(hopefully) more Go like.

A vector type has been asked for by David Neumann and there is some of
that included, but I have stalled on that, for reasons. Stack and Augment
need implementation, but this is complicated by a design decision to
support row and column major formats. I'm sort of waiting on noise from the
nd slice proposals before I make a lot more effort there unless that drags
on.

As Brendan pointed out testing is a need. Also, an implementation of at
least the D BLAS routines in Go that matches blas.Float64 would be good.
I'm not going to do that.

Dan

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/1#issuecomment-27595983
.

from blas.

Happy to take pull requests. I was thinking more though about specific needs that you might want implemented or be prepared to add. Are there notable deficits in gonum/matrix that you would like to see addressed?

I wish my time was only as tight as a grad student's.

Dan

from blas.