Genetic Algorithms

[JietseV]

This is the pull request of the final project in the STMO course 2021-2022.
Contributors: Jordi Verbruggen and Jietse Verweirder

[MichielStock]

Hi Jordi and Jietse,

Your project looks to be in very good shape!

Small remarks:

• I personally think a heatmap is more clear than a 3d surface plot, but this is personal taste,
• the first figure is a bit grainy?
• you might allow me to change the objective?
• you can leave your code blocks visible. Suppress the docs with a ; at the end
• in rand_init_pop, by using . or other smart things you can make sure pop is not an Any vector. Concrete types greatly improve performance. For example pop = [ ... ] always yields a concrete type. Fix this for your other functions as well
• minima_himmelblau => would give the objective here, could easily work for arbitrary 2D functions, no?

solving the Any[] vectors seems to be the only 'critical thing'. Good job!

[FaustSchotte]

Hi Jordi and Jitse,

I think your project looks really good. I don't have any remarks on the actual coding part of your project.
However, I did spot some typos in your project which are easy to clean up:

• In the section initial population: I think there is a typo at permutation size. Should "higher change" not be "higher chance"?
• In the termination criteria section: "Herefore" is used instead of "Therefore"
• Under Roulette Wheel Selection: you typed "an thus resulting" instead of using "and"
• Under the crossover section: for "random parings of parents" this should be "pairings", and in the same paragraph "devided" should be "divided"
• The complete algorithm: "sart" should be "start"

[natclaret]

https://github.com/natclaret/Ant_Colony_Optimization.jl/blob/master/geneticalgorithms.jl

Awesome notebook! I especially like your interactive parameter sliders / gif element. I also appreciate the paragraph you include after, that explains what we’re seeing in the graphic.

Linked notebook: Sorry, couldn’t get the pull request working right so I saved a local copy of your project, edited it, and put it in my repo – so that's what the link is for.

• For the most part I fixed some typos. In a very few spots I made subjective changes, like slightly rewording a sentence or changing the boldface. Since that is mostly dependent on someone’s personal taste I only did it if I was convinced that it was a good/helpful change. 😊
• I did not alter your code at all. Everything looked good to me.

Other remarks / feedback: To be clear, your content is great as is. Obviously I was looking for things to offer feedback on, so some of this is kind of picky stuff. See below:

• roulette_wheel_selection(population fitness): the formula you give for the function has variable names that don’t seem to exactly match the formula names you use in the function. I suggest you update the formula to match the function or vise versa.
• “Defining the initial population” section: you say “diversity” must be taken into account. Maybe add another sentence to explain that in more detail. What does it mean for the population to be diverse?
• “Fitness” section: Is it common that the fitness function can bottleneck the algorithm? Is this a drawback of using GAs or would you say that other comparable optimization techniques would suffer just as much from an inefficient cost function?
• “The code goes as follows” statements before each function: sometimes you have them italicized and sometimes in normal face. Change them all to a matching style.
• “Selection” section: The sentence “Note that an individual can be selected more than one time, otherwise the population would not change.” It’s not obvious to me why this is true?
• “Roulette Wheel Selection”: an image of a Roulette wheel might be helpful here.
• “Rank selection”: This sentence is confusing – “In rank selection, the fitness values can even be negative, as it is their rank that counts and not their actual value.”. I don’t think it fits well there.
• “Crossover”: I would leave out “This is a specific case of a N-point Crossover technique” unless you explain what N-point crossover technique means.
• Maybe add a comment to let the reader know when we’ve stopped talking about various selection techniques and are moving onto another aspect of GAs.
• “Mutation” section: What is a Gaussian deviate?
• “Mutation” section: You mention “the change is getting smaller”- changes in what? Is it that the evolution rate is slowing and the offspring population more closely resembles the parent population?
• “Mutation” section: Before giving the function code, can you mention which of the given mutation strategies you are going to exemplify?

Again, really nice work, Jordy and Jietse! I enjoyed reading it.