Computational experiments: organising different algorithms, their parameters and results

General Usage
21

Hi, just some points I learned from my past experiences with computational modeling:

  • It is often not necessary to calculate all possible parameter combinations. Well selected extreme points will usually tell you a lot in which direction to go, without needing to wait hours/days/months until all the data points have finished computation. You can run the full parameter combinations, while writing the scientific paper.
  • What I also find very usefull is the combination of version control and make-file. Using version control, I know how the overall code worked (functions), whereas the function arguments are stored inside of a make-file. The make-file has the additional benefit, that it can be version controlled / easily backed up and different tools can be very flexibly chained together. Software carpentry has a good tutorial on how to automate such a scientific workflow: Automation and Make
2 Likes
22

Agreed, this requires an interactive design of experiments though, which should be entirely possible with my package.

The make-file tutorial seems interesting, thanks for sharing.

23

Proof of concept:

https://github.com/mohamed82008/ComputExp.jl

There is more work to be done, but further progress might be slow since I am prioritizing GSoC for now and Ramadan is coming up :slight_smile:

4 Likes