pouët.net

Go to bottom

Math library

category: code [glöplog]
 
I usually use very few library and prefer build almost everything from scratch... sort, fft, gui, database and so on. Once I build my own stuff, I could re-use it among my projects. If there is an issue, I just go in my code and fix it.

But, sometimes I run into problems and wishes a simple solution can be available without having to code it myself...

Last week, I run into that problem which seem easy... Consider a vector of N elements (where N is near to hundred of millions!). I wanted to know if some part of that vector is similar than others with at least 92% similar. Hum, I started to code something... launch my software and wait... 2... 3.. hours.. hum, still around 2% done! :(I though, this could be process really fast with other algorithm.. longer to code.

It happen from time to time.. few month ago I wanted to process BIG INT number.. using 4096-bit integer! I even though that today perhaps there is a datatype for a C compiler where you can specify how many N-bit it corresponds... I though.. huh.. no! Wow... I am surprise no one even put that in.

I would like to know if you have some math library that you use for that kind of "exotic" software!

added on the 2011-12-05 21:54:01 by F-Cycles F-Cycles
added on the 2011-12-05 22:09:59 by tobé tobé
Searching for similarities in large vectors sounds like what rzip does for data in files. See http://rzip.samba.org/. It's basically your normal LZ compressor except it works with a block size of 900 MiB rather than 64k. I bet you could borrow stuff from it for your algo.
added on the 2011-12-05 22:19:20 by Tjoppen Tjoppen
I second gmp, that's mainly what is used to do the crypto math on openssl and it has probably had more eyes on it then most other people's big-number libraries.
added on the 2011-12-05 22:30:16 by sigflup sigflup
Maybe python ?
I'm using it on "Big" Problems like finding pattern in Datasets around 5-6Gb.
Speed is not the same as C but with some tweaking it could get as nearly as fast as C.
Handles things like N-Scaling quite nice.
added on the 2011-12-08 08:43:04 by skrebbel skrebbel
it's called correlation.
iftt(fft(vector_a)*fft(vector_b))
added on the 2011-12-08 10:13:24 by rmeht rmeht

login

Go to top