Evolutionary Algorithms

For various reasons, this week’s quiz answers will be discussed next week. Additionally, given the mishaps in formulation, we have extended the deadline until Sunday for those that haven’t finished yet.

The next (3rd) assignment will be still be available from the 6th of May onwards.

We have noticed that some of you have been including the mutation of strategy parameters into the solution to the assignment, and have taken note of the ambiguity / error due to why this is happening.

The primary focus on this assignment is on the problem variables, requiring no alteration / mutation of the strategy parameters. This was additionally confusing, as the assignment originally used “genotype”, which during the lecture was explained explicitly to be both the problem variables and strategy parameters. The description of the assignment has been altered accordingly to clarify this.

Solution & tests have been kept identical: if you have completed the assignment already, you do not need to change anything.

An error has been found in the tests for “Adapting strategy parameters in the single-variance case”. \(\tau^2\) was used as a std. deviation instead of a variance. This resulted in incorrect solutions passing all tests, while the correct solution did not pass all tests.

We have corrected for the error while accounting for people who have already passed the assignment.

We would like to make two clarifications for this weeks assignment. First, we used the term “self-mutability” in the first question of the quiz, but this should be “self-adaptation” to reflect the newer terminology. The content/meaning of the question remains unchanged. Second, there may be some confusion around variances and standard deviations, and which of these are encoded in an individuals strategy parameters. This is always standard deviation, even if some of the terminology (like l-variances) uses variance. In the practical coding assignments, assume therefore that standard deviations are encoded.

It has come to our attention that using the ‘np.choice’ function to randomly select individuals from the population may lead to a time-out of the spec tests for assignment 1.3. Instead of directly choosing individuals from the population, we advise you to randomly generate indices \(i \in [0,\ell-1] \) and then select the individuals with \( \texttt{self.population[i]} \).

Due to the limited amount of time left, we advise you to use this workaround for now, while we will make a permanent fix for next year’s course.