dorkrawk / darwinning Goto Github PK

1. I am new to genetic algorithms 2) I haven't examined all the code carefully 3) I don't want to sound nitpicky - I am just learning :) but, is there a specific reason in your design for not storing #fitness inside the object rather than computing it each time it is accessed?

Hello Dave,

last year I needed a simple/elegant genetic algorithm library in ruby for one of my student internship. I found yours to be the most adapted to what he needed to do. I still needed to change quite a bit of things in order to get it where I wanted. When I stopped I was not sure a pull request was really feasible anymore or my branch mature enough. More than 30% of the code has been rewritten.

I see that you have received contributions recently, which means that the two branches will diverge from now on. So I though maybe maybe you could be interested in the changes I did before they are irremediably unmergeable.

the fork can be found here:
https://github.com/Nanosim-LIG/darwinning

Don't hesitate if you have questions.

Best regards,

Brice

It won't mutate the last gene because of the bug.

The corresponding line in mutation.rb:33 should be:
random_index = rand(member.genotypes.length)

hi, thanks for sharing your code. How could I run your exemples?

Hi!
I'm just trying to run examples and see this error.

Hey,

Thanks for this project.
However, it is lacking a license which means I am currently unable to use it.

Could you please add a license?
https://choosealicense.com/no-permission
https://help.github.com/articles/adding-a-license-to-a-repository/

Thanks

So I am trying to use this library to do the most basic thing other than the trivial examples provided:

#!/usr/bin/env ruby

require 'darwinning'

class Vehicle < Darwinning::Organism
   @name = "Car"

   @wheels = Darwinning::Gene.new(name: "Wheels", value_range: (0..4))
   @cylinders = Darwinning::Gene.new(name: "Cylinders", value_range: (1..16))
   @tank_size = Darwinning::Gene.new(name: "Tank Size", value_range: (1..16))

   @genes = [ @wheels, @cylinders, @tank_size ]


   def fitness
       return ((@wheels.value - 4) + (@cylinders.value / @tank_size.value)).abs
   end
end

p = Darwinning::Population.new(
    organism: Vehicle, population_size: 10,
    fitness_goal: 0, generations_limit: 100
)
p.evolve!

p.best_member.nice_print

However this doesn't work because Gene.value doesn't get set (its always nil)

I had to do the following to get things to work.

#!/usr/bin/env ruby

require 'darwinning'

class Vehicle < Darwinning::Organism
   @name = "Car"

   @wheels = Darwinning::Gene.new(name: "Wheels", value_range: (0..4))
   @cylinders = Darwinning::Gene.new(name: "Cylinders", value_range: (1..16))
   @tank_size = Darwinning::Gene.new(name: "Tank Size", value_range: (1..16))

   @genes = [ @wheels, @cylinders, @tank_size ]

   def find(name)
       genotypes[self.class.genes.index{ |g| g.name == name}]
   end

   def fitness
       wheels = find("Wheels")
       cylinders = find("Cylinders")
       tank_size = find("Tank Size")

       return ((wheels - 4) + (cylinders / tank_size)).abs
   end
end

p = Darwinning::Population.new(
    organism: Vehicle, population_size: 10,
    fitness_goal: 0, generations_limit: 100
)
p.evolve!

p.best_member.nice_print

Wonderful (and simple) tool that I've been using to try and solve a tricky problem with balancing work teams. I've been wondering if there is any way to seed the population with a few known "decent" solutions. It's taking hundreds of generations until the randomly generated solutions are anywhere near as good as the manually created solutions. So I'd like to feed in one or two of the manually created solutions.

Thanks for your great gem!

-Dave

There is a bug in calculating the best_each_generation. The history for each gen is actually not sorted because the fitness is not yet calculated so the sort_members calls in initialize and in make_next_generation don't have the desired effect.

fix is to remove the sort and adding to history from initialize and for make_next_generation change to:

  def make_next_generation!
    verify_population_size_is_positive!
    sort_members   ####################### sort and add to history before generating new members
    @history << @members
    
    new_members = []

    until new_members.length >= members.length
      m1 = weighted_select
      m2 = weighted_select

      new_members += apply_pairwise_evolutions(m1, m2)
    end

    # In the case of an odd population size, we likely added one too many members.
    new_members.pop if new_members.length > members.length

    @members = apply_non_pairwise_evolutions(new_members)
    @generation += 1
  end

In the example directory there are examples with limited number of genes: Binary -10, Cookies -11, Fifteen - 3 and Triple - 3.

How to implement variabile number of genes or organism with one gene but variabile and unlimited size?

I noticed while working on an unrelated, forthcoming pull request that two of the specs in population_spec.rb fail sometimes in master.

I may look into it if I find the time, but I figured I would at least document the issue.

Is there any way to specify arguments to be passed in to the organism's initialize?

E.g: Consider the example in the README, how could I modify this so I can pass in the target number, instead of having it hard coded to 100?