A new evolutionary algorithm using the tiger mosquito plague fighthing techniques
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
assets
code
notebook
report
.gitignore
.travis.yml
LICENSE
NK Fitness.md
README.md
Selection.md
_config.yml
algorithm.md
team.md

README.md

The tiger mosquito algorithm

This is the project of one of the teams in the SIGEvo summer school 2018.

This site can be accessed also through the short url https://git.io/wasabitigers

The final presentation at the summer school is available from GDrive

Description of the algorithm

A new evolutionary algorithm inspired from the tiger mosquito plague fighting techniques. The only things that change is the selection mechanism, which uses a pheromone to represent the direction in which search should go.

An algorithm that removes mosquitoes that carry dengue virus, zykova virus, yellow fever virus. You will choose a pheromone to eliminate mosquitoes, a good pheromone and a bad pheromone to make a pheromone that attracts tiger mosquitoes.

  1. Initialize the pheromone.
    • The way to initialize the pheromone fitness is to use the NK_landscape method.
    • A detailed code of nk_landscape is shown on this page.
  2. Measure the fitness of pheromones.
    • After initialization, the fitness function uses PP (Pheromone-Potential).
    • We will use two different update methods for the pheromone value (P) of each individual.
      • Based on the whole inheritance:
      if x_i selected:
        P(x_i) += constant_value
        P(parents[x_i]) += Q^h*P(x_i)
      • Based on their children
      if x_i selected:
        P(x_i) += constant_value + k * size(childrens)
    • The contents of the PP are shown in This page.
  3. Use the tormant selection to get the results.
    • We use the new selection scheme that the individual is given the score stochastically.
      1. The random value r is sampled from the uniform distribution.
      2. Select the N_tournament solutions in X at random.
      3. If r <= sr, the best candidate solution is set as x* = max_x ( f(x_1), f(x_2), ..., f(x_N_tournament) ) otherwise, the best candidate solution is set as x* = max_x ( P(x_1), P(x_2), ..., P(x_N_tournament) ) .
      4. Return the best solution x*
    • A detailed code of selection is shown on this page
  4. Cross over each result to get a new generation and perform mutation.
  5. repeat

A more extended description will be shown in this page

The team

5 students from all over the world, and a tutor.

Other teams

The tigers team tries to model the spread of mosquitos.