Open ptizei opened 3 years ago
thanks @ptizei , I like these comments a lot!
I updated Elitism. Great catch on the PROB_CROSS_OVER, PROB_MUTATION calculation!
I'm leaving this open at the moment because I need to think more about the usefulness of applying PCA or t-SNE. My first intuition is that it would bring in yet more stochasticity to the problem (perhaps best explore on an exhaustive search space.
I searched a bit and it seems like the important point when implementing Elitism is to ensure that this fraction of best genotypes doesn't get destroyed by mutation/crossover. Some comments I saw mention that some implementations still consider those "Elites" as possible parents in the breeding process, but in that case they don't delete the parents like your current implementation does.
Maybe replace it with something like to make it more clear:
In the second markdown cell of "6 Knapsack Optimisation", you have the following equation 'replicate_prob= 1 - PROB_CROSS_OVER - PROB_MUTATION', which would give 'PROB_CROSS_OVER + PROB_MUTATION = 1' after setting 'replicate_prob' to 0.
Could dimensionality reduction methods like PCA or t-SNE also be useful for visualisation here? You probably wouldn't be able to pick out the pareto fronts visually after the points got remapped, but I guess they could provide some overview of how the populations behave over time.