Blackbox optimization fitness evolution non-monotonic

General Usage
1

Hi all,

I use the default algorithm (rand/1/bin) in the BlackBoxOptim package to optimize a single-objective function. Here is a plot of the best fitness per generation:

image
image864×576 25.9 KB

I am not sure why the best fitness does not decrease over generations. Here are some guesses:

  1. My code generated a large csv file and I extracted the fitness using shell commands, which might have generated errors along the way.
  2. The algorithm (adaptive differential evolution) does not guarantee that the best fitness decreases monotonically.
  3. There is something wrong with the algorithm, but I don’t know how to verify if this is the case.

Thank you very much for reading this.