11/08/2010

11-08-10 - Name your techniques!

When you create an algorithm, give it a god damn name. I want to be able to refer to "PPMC" or "ARC" or "lookup2" or something and have a well defined name for a specific algorithm.

It's really fucking annoying when people write a paper and call it "A novel DCT compressor based on pattern segmentation" and then within the paper they refer to the new algorithm as "the method in this work". Then if you are referencing them you have to call it something and it's a fucking disaster. (to get back at them you should refer to it as "The work by the morons who didn't name their algorithm").

Furthermore, when you publish revisions, fucking use a new name or a version number or something. It sucks when people publish a series of papers where they introduce changes and the whole time keep referring to it as "VDP" . No no, it's "VDP" the first time and then after that it's "VDP-2" or "VDP-M" or whatever. Don't make me have to specify "the PPMD from the first paper in 2004".

I know a lot of authors are timid and think it's not cool to just name a technique after themselves, but that's what they really want, so they intentionally don't give their algorithm a name in the hope that the community will call it "The Wang Method" or whatever. Don't be a pussy, just name it after yourself if that's what you want, it's fine.

Of course another sin that I have been guilty of myself is to release source code for an algorithm ("LZP2" for example in my case) that's different from that paper, so you refer to "LZP2" and describe it in detail in the paper and then your release an exe called "LZP2" that's actually different. This is just constantly done and is very sloppy ; at minimum you should make it clear and call the source code "LZP2a" or something; almost every time I see source code released with a paper, it differs from the description in the paper. (of course I am still happy that there was at least source code released).

No comments:

old rants