|
|
|
|
|
|
by ArbitraryLimits
5170 days ago
|
|
|
> Of course anyone who runs the same program gets the same results, so ordinary replication is pointless That would be true, if anyone actually distributed their actual code. Pick a journal article at random in any field that describes results from a computational model, and 99% of the paper will describe the results and not the model. The paper will never contain the complete code (which is fair enough, since it would be too long); 1 paper out of 100 will have excerpts of the code, and another 10/100 will have a URL that claims to have the code. If you actually follow that link, you'll find that 2-3 times out of 10 the code won't actually compile or run, and 9 times out of 10, the figures in the paper were generated by tweeking some parameters not defined in the paper whose particular values the author never recorded, and so even the author couldn't reproduce what he actually published, even if he wanted to. Some people are trying to make institutional change here, see http://www-stat.stanford.edu/~wavelab/Wavelab_850/wavelab.pd... But really it's a pretty sad state of affairs. |
|
|