Lab 2 Review

[Peipi98]

Pros and Observations

• I've appreciated the use of a mask with True and False values corresponding to the ALL_LISTS' indexes rather than a tuple of tuples as in my case, which is more efficient and less memory consuming than my solution;
• Code is well organized and richly commented, it helped me a lot to move through your code.
• You have tuned accurately your model in order to have a good tradeoff between time and solution (I've also tested your hyperparameters in my code and it works better than mine);
• The strategy to "cut out" some weight with the negation function is appreciated, but maybe there is an error here (please answer me if I'm wrong):

for number_iteration in range(N_MAX_ITERATIONS):
    offspring = list()
    for i in range(OFFSPRING_SIZE):

        parent1 = select_parent(population, tournament_size = tournament_size)
        parent2 = select_parent(population, tournament_size = tournament_size)
        new_individual = crossover(parent1.genome, parent2.genome) 

        if np.random.uniform() <= mutation_rate: #Mutation perfomed 50% of the times
            if number_iteration < N_MAX_ITERATIONS/negation_rate: #First mutations are rweigth loss aimed
                new_individual = negations(random.choice([parent1.genome, parent2.genome, new_individual])) 
            else: #single flip mutation 
                new_individual = single_mutation(random.choice([parent1.genome, parent2.genome, new_individual]))

In the internal if there is a condition in which the mutation_rate is 0.5 and number_iteration is minor than N_MAX_ITERATIONS, so your condition will always be true since number_iteration will always be less than the double of his N_MAX_ITERATIONS (I've checked with a print inside the else, and it never prints out).

Suggestions

• Rather than using (number_of_covered_elements, -weigth) inside the fitness_function, I strongly suggest you to use only the weight and setting a control on the covered elements in order to give a strong penalty to the individuals which haven't already reached the set covering (check my code to have an example).

[AlessiaLeclercq]

Hi :) Thank you for your review. I fixed the internal if to N_MAX_ITERATIONS*negation_rate and now it works faster, however it still returns the same results (see below). I implemented the fitness as you suggested (and also changed the code accordingly from a maximization to a minimization problem) as follows: Here I report some statistics of your version (in green) and mine (yellow) and they are pretty similar.

Thank you again :)