Loss of best fitness using mu + lambda and NSGA II

[Liz Wachs]

Hello,

I have a code that solves the multi-objective knapsack problem using the mu + lambda and NSGAII for selection. I ran experiments and then plotted my minimum fitness values to check for convergence. On 9 of my experiments I could see the correct trend. On the 10th, one of my 20 experiments moved up at the end. The individual with the lowest fitness (minimization) disappeared. I am trying to understand how this could be. I am getting back into DEAP after a while. Thanks for any leads! I am wondering if the best individual could have mutated and thus disappeared. Perhaps if I had more generations (stopped at 50) this would have escaped my attention and I would see the type of curve I expected.