3 ms·
It's terrible in the CS field, too—how often on HN do we see papers about some new algorithm, with a fancy demo reel, and no code? A CS paper without source co
by mcphage 12y ago
It's terrible in the CS field, too—how often on HN do we see papers about some new algorithm, with a fancy demo reel, and no code?
A CS paper without source code isn't science, it's show-and-tell. If all you're looking to do is get hired by Adobe/Microsoft/Google/Apple, fine, but don't insult our intelligence by pretending you're doing science.
- nmrm2 12y agoWow. I'll go let the theorists know that their theorems and proofs don't count.
- mcphage 12y agoI'm speaking about claims of execution (we ran our image processing algorithm, or GA, or whatever, and got the following results). If your claim is a theorem, then sure, the proof of the theorem is sufficient.
- jimmaswell 12y agoWhy would a published algorithm have to include runnable sample code unless it was part of a language specification or something?
- mcphage 12y agoHow else is it repeatable? If the claim they're making is a proof—then yes, the proof is sufficient. But if they're making claims about execution "we churned through data set X with results Z", then without someone else able to run the same process, how else could someone verify your claims?
- tzs 12y agoThat's a fairly weak form of reproduction. Running their code on their data and seeing the same result just shows that they didn't make a clerical error when preparing their publication, and that the code is portable to your hardware. When someone else implements from the published algorithm, ideally without even looking at the code, and gets the same results, then we're getting into interesting reproduction.
- mcphage 12y ago> Running their code on their data and seeing the same result just shows that they didn't make a clerical error when preparing their publication, and that the code is portable to your hardware. Or fraud, or any larger mistake. You can run it with other data to make sure the claims hold up. You can verify what they're saying is accurate, and not just a fluke in their data.