[racket] Blog post about Racket

On Tue, May 13, 2014 at 9:35 AM, Konrad Hinsen
<konrad.hinsen at fastmail.net> wrote:
> I have doubts about some points on that list:
>
>   - machine learning

For what it's worth, I implemented Pegasos awhile back:
https://gist.github.com/jkominek/1275886

It's intended for SVMs on large datasets (like, read everything once
and only once large). Between it and Racket, you can use anything you
want for the kernel function trivially. The example at the end of the
file uses a string kernel, and I've toyed briefly with graph kernels
(for comparing molecules).

> If there are Racket libraries for any of those, I'd like to hear about
> them.  There is some statistical stuff in the math library, but a
> modern data scientist needs a lot more than that. Plus libraries to
> read and write common data formats, which are missing from that
> article's list, probably because the author took them for granted.

Perhaps you could add a page to the Racket wiki on github, like the
intro projects page
(https://github.com/plt/racket/wiki/Intro-Projects) but which
covers... "Scientific Racket projects"? Divide it up like "machine
learning", "statistical analysis", "file formats" and then either list
the things you'd like to see it capable of, or just point to the
best-in-class from other languages and say "beat that".

> I see Racket's strength for scientific computing in a very different
> aspect: the possibility to define languages tailor-made for expressing
> computational models in some application domain. Scientists generally
> don't want to "write programs", and when they do, the results are
> often not pretty. I'd like to have scientists do science and
> programmers write programs. Racket could become the meeting point for
> the two professions.

I've personally watched a number of projects where that could've saved
significant time, money and frustration. I'm not optimistic about it
coming to pass, but it'd sure be nice.

-- 
Jay Kominek