Dr Robin Wilson from Geography at Southampton gave a research seminar concentrating on the importance of reproducibility for satellite imaging.
Dr Wilson explained the importance of satellite imaging for a broad range of applications, from disaster relief, government management and commercial exploitation. This computationally complex problem is challenging to automate, with many steps often using graphical interfaces. Even when the workflow has been automated, the number of steps, and places where steps can be “invisibly” changed, make it difficult to reproduce results.
Dr Wilson then discussed the recipy code, a python package that stores metadata for reproducibility purposes. He demonstrated how easy it is to use, discussed how it links with version control, and illustrated best practices in its use. The metadata is linked to the outputs of runs and analyses (for example, the plots output from a script), so that the commands and software needed to re-create the plot from scratch.
In the questions, Dr Wilson went into more detail on how recipy works internally, so that users can wrap their own code. By monkey-patching functions that read and write data, recipy is able to keep track of the steps in the workflow.
Posted by Ian Hawke