- Take a numerical optimisation problem and a GA that is suited to solve it, i.e., uses the appropriate representation. (You can write your own code, or download it from the Web.) Select 3 different values for each of the parameters population size μ, mutation rate
*p*_{m}, and crossover rate *p*_{c} . Execute 30 runs with each of the 27 different GA instances and for each run save the best fitness at termination, the number of fitness evaluations and the CPU time needed to complete the run. Perform a simple statistical analysis on the spread of the outcomes, e.g., calculate the minimum, the maximum, the average, the standard deviation, etc. Use all 27 setups as the basis of your statistics first, then fix one parameter at one of its values and do the same analysis for the 9 corresponding runs. How does this change your results? Summarise your observations in a short report.
- Same as above. Now compare the results for the 27 GA instances and try to draw conclusions about good parameter values. Are the good parameter values for the highest solution quality and the ones for the fastest GA the same?
- Same as the first exercise, but now for an evolution strategy and the parameters μ and λ.
- Same as the second exercise with the evolution strategy.
- How many EA instances were used to generate the plots for the Figure below?
- How many EA instances were used to generate the plots for the Figure below?

## The on-line accompaniment to the book Introduction to Evolutionary Computing