{"id":22,"date":"2016-07-29T05:20:17","date_gmt":"2016-07-29T05:20:17","guid":{"rendered":"http:\/\/blog.orkan-tief.de\/?p=22"},"modified":"2022-09-27T10:08:27","modified_gmt":"2022-09-27T10:08:27","slug":"unit-testing-in-scientific-programming","status":"publish","type":"post","link":"http:\/\/www.martin-rdz.de\/index.php\/2016\/07\/29\/unit-testing-in-scientific-programming\/","title":{"rendered":"Unit Tests in Scientific Programming"},"content":{"rendered":"

Testing your code seems like a sane idea, at least from a programmers perspective [eg 1<\/a>]. As a scientist when it comes to crunching numbers, in the majority of cases, you hack together a small script and hope that it runs as intended. If strange behavior\u00a0(aka bug) occurs while processing a gigabyte large dataset, you give it a quick fix, pray for the best and start computation all over. After nights of debugging it gets clear that this is not the best approach (or earlier\u00a0 – i have learned it the hard way).<\/p>\n

Assuring the correct execution is rather easy for small parts of code. It is obvious what the following (exaggerated simple)\u00a0function does:<\/p>\n

def larger_zero(number): \r\n    return number > 0<\/pre>\n
Problems develop with increasing length of the function or multiple functions\/objects interacting with each other. In the real world you are more likely to have a ~1000-lines-of-code data analysis script. It is reading from different files, doing computations and plotting and finally writes the results out. How to assure correct execution in this case?<\/p>\n
When your scientific software is modularized (which it ideally is, but this topic is too broad to discuss it here), testing seems pretty straightforward:<\/p>\n
    \n
  1. starting at the bottom: test independent functions on their own (the core of unit testing)<\/li>\n
  2. make sure, that the communication between functions works correct<\/li>\n
  3. examine if the whole script does what you intended<\/li>\n<\/ol>\n
    Some people argue, that you should write the tests before<\/strong> you start to write your software. In science where you might not know at the beginning what your program should look like at the end, this is not the perfect solution. My current way to go, looks approximately like this (I’m not claiming that there is no room improvements):<\/p>\n