Re: Good starting points/project/bugs

[Viral Shah]

Hi Sonal,

Welcome. It is exactly for this purpose that the "up for grabs" tag has been introduced in the Issues. These are typically features that can be implemented by people as starter projects, which are largely independent.

https://github.com/JuliaLang/julia/issues?labels=up+for+grabs&page=1&state=open

Is there a particular area that excites you, where you would like to contribute?

-viral

On Thursday, June 21, 2012 7:56:42 PM UTC+5:30, Sonal Chhawal wrote:

Hi Everyone,

I have been experimenting with Julia language for a while now. Now, I want to contribute to the project. Can someone guide me to a good point to start off? As a background I have several years of coding experience with C/C++ and currently, I am familiarizing myself with the Julia code base.

Cheers,

Sonal

[Sonal Chhawal]

Hey,

Thanks for replying and for pointing out in the right direction. It
took some time to figure out my interests and this would be my first
open source work. I have understanding of statistical methods and I
am looking to start with something which is easily deliverable, so
that I can get into pace for development of statistical features and
general developement of the language. A couple of issues look as good
starting point:
a. Reading mat file (#805)
b. Implement HDFS Support(#811 )
c. working on a performance feature "use copy_to in assign() and ref()
where possible"(#190)
d. Implementing better summation algorithm. ..

I guess I will just pick the first one in the list and start from there.

Cheers

> --

[Stefan Karpinski]

That would be a great contribution. Be sure to check out libmatio. Providing appropriate bindings to this C library using ccall is almost certainly the way to go here.

--

[Jeffrey Sarnoff]

If you choose matio, do not hesitate to post questions you may have about how to form the ccall for a given C function.
The answers benefit others -- like me. 

[Tim Holy]

Hi Sonal,

This sounds great. What do you mean by a "better summation algorithm"? I'm
intrigued...

FYI, item c is already underway in the "arrays" branch, but there's still
quite a lot of work to be done. Your efforts are welcome, on this issue or on
any of the others you mentioned (they're all great choices).

Best,
--Tim

[Sonal Chhawal]

Hi,

Thanks , I will surely ask whenever I ll encounter any major issues. I am starting with the feature for mat file support and let's go from there.

Tim- i was referring to issue #199 which discusses a potential requirement for a better summation algorithm.

cheers,
- Sonal

> --
>
>
>

[Stefan Karpinski]

Yes. Fun fact: since floating-point arithmetic isn't associative, accumulating a set of numbers in a sum gives potentially very different results depending on the order. Kahan summation helps; the most accurate but slowest algorithm is to sort all your numbers and then add them in order; recursive binary summation is somewhere in between (and easily distributed).

> --
>
>
>

[Jeffrey Sarnoff]

TOMS 37,3 (Sep 2010) Article: 37
for	Online Exact Summation of Floating-Point Streams
by	Yong-Kang Zhu and Wayne B. Hayes

http://www.netlib.org/toms/ (click and save 908) or use http://www.netlib.org/toms/908

[if it is delivered as 908, add the .zip extension]

Siefried M. Rump, Takeshi Ogita, and Sin'ichi Oishi

  Fast High Precision Summation 

       http://oishi.info.waseda.ac.jp/~oishi/papers/RuOgOi08.pdf

  Accurate Floating-Point Summation 

       http://www.ti3.tu-harburg.de/paper/rump/RuOgOi07I.pdf

       http://oishi.info.waseda.ac.jp/~oishi/papers/RuOgOi07II.pdf 

Laurent Fousse and Paul Zimmermann 

  Accurate Summation (Proofs for the Demmel & Hida algorithm)

      http://webloria.loria.fr/~zimmerma/papers/rnc5.pdf

      http://www.cs.berkeley.edu/~demmel/AccurateSummation.ps (Demmel & Hida paper)

[Stefan Karpinski]

Like so many things floating-point, summation is shockingly complicated. Don't even get started on printing!

--