tvanderpol

tvanderpol

Genetic Algorithms in Elixir: First stab at Ch1 algorithm converges just fine? (p28)

I’ve got something of the opposite of the usual problem - my code doesn’t fail in the way the text suggests it should (when it suggests premature convergence is the cause).

I’ve ran the code a fair bit and did some debug print injecting and such but I can’t really see anything out of the ordinary - it converges on the correct answer almost instantly if I run it without any additional debug and it takes a reasonable amount of generations from what I can tell by poking at it.

Now I am clear to continue the chapter as-is, and I understand the argument being made, but I don’t understand how the algorithm is meant to fail (I do in the abstract but I mean the code as written) and that’s bugging me.

Marked As Solved

seanmor5

seanmor5

Author of Genetic Algorithms in Elixir

There was a mistake in my version of the code that forced early convergence with smaller populations that isn’t present in the book’s transcription of the code.

To better demonstrate premature convergence: set chromosome size to 1000 and population size to 100. You’ll notice your version without mutation converges much slower than with mutation. You can continue to decrease the population size further and further and you’ll reach a point where progress completely stops.

Sorry about the confusion!

Also Liked

christhekeele

christhekeele

This was on quite a new Macbook, which may be influencing results from what’s expected.

MacBook Pro (16-inch, 2019)
2.4 GHz 8-Core Intel Core i9
32 GB 2667 MHz DDR4
Erlang/OTP 22 [erts-10.7.2] [64-bit] [smp:16:16]
Elixir 1.10.3

tvanderpol

tvanderpol

Perfect, thank you for responding so fast! That helps my understanding of exactly how it fails a lot.

christhekeele

christhekeele

I noticed this in the B1 edition as well!

Re:

$ elixir one_max.exs
Current Best: 32

But wait, what’s going on here? Why is the algorithm stopping on a best fitness below 42? No matter how many times you run it, the algorithm will almost always certainly stop improving below 42. The problem is premature convergence.

At this point, the chapter’s example code looks like:

Code to Date
population = for _ <- 1..10, do: for _ <- 1..42, do: Enum.random(0..1)

evaluate = fn population ->
  Enum.sort_by(population, &Enum.sum(&1), &>=/2)
end

selection = fn population ->
  population
  |> Enum.chunk_every(2)
  |> Enum.map(&List.to_tuple(&1))
end

crossover = fn population ->
  Enum.reduce(population, [], fn {p1, p2}, acc ->
    cx_point = :rand.uniform(42)
    {{h1, t1}, {h2, t2}} = {Enum.split(p1, cx_point), Enum.split(p2, cx_point)}
    [h1 ++ t2 | [h2 ++ t1 | acc]]
  end)
end

algorithm = fn population, algorithm ->
  best = Enum.max_by(population, &Enum.sum(&1))
  IO.write("\rCurrent Best: " <> Integer.to_string(Enum.sum(best)))
  if Enum.sum(best) == 42 do
    best
  else
    population
    |> evaluate.()
    |> selection.()
    |> crossover.()
    |> algorithm.(algorithm)
  end
end

solution = algorithm.(population, algorithm)
IO.write("\n Answer is \n")
IO.inspect solution

I parameterized it thusly:

Tunable version
-population = for _ <- 1..10, do: for _ <- 1..42, do: Enum.random(0..1)
+problem_size = 42
+population_size = 100
+
+population = for _ <- 1..population_size, do: for _ <- 1..problem_size, do: Enum.random(0..1)

 evaluate = fn population ->
   Enum.sort_by(population, &Enum.sum(&1), &>=/2)
 end

 selection = fn population ->
   population
   |> Enum.chunk_every(2)
   |> Enum.map(&List.to_tuple(&1))
 end

 crossover = fn population ->
   Enum.reduce(population, [], fn {p1, p2}, acc ->
-    cx_point = :rand.uniform(42)
+    cx_point = :rand.uniform(problem_size)
     {{h1, t1}, {h2, t2}} = {Enum.split(p1, cx_point), Enum.split(p2, cx_point)}
     [h1 ++ t2 | [h2 ++ t1 | acc]]
   end)
 end

 algorithm = fn population, algorithm ->
   best = Enum.max_by(population, &Enum.sum(&1))
   IO.write("\rCurrent Best: " <> Integer.to_string(Enum.sum(best)))
-  if Enum.sum(best) == 42 do
+  if Enum.sum(best) == problem_size do
     best
   else
     population
     |> evaluate.()
     |> selection.()
     |> crossover.()
     |> algorithm.(algorithm)
   end
 end

 solution = algorithm.(population, algorithm)
 IO.write("\n Answer is \n")
 IO.inspect solution

In my experimentation, with problem_size = 42, not only did population_size = 100 always converge on the best answer, but even as low as population_size = 8 consistently converged. I started seeing the need for mutation around population_size = 6, which normally gets stuck around 35.

Alternatively, increasing the size of the problem to problem_size = 420 usually converged correctly, but with enough time to watch things work. problem_size = 4200 consistently gets stuck around 2400, as the narrative of the chapter wants it to.

Where Next?

Popular Pragmatic Bookshelf topics Top

iPaul
page 37 ANTLRInputStream input = new ANTLRInputStream(is); as of ANTLR 4 .8 should be: CharStream stream = CharStreams.fromStream(i...
New
mikecargal
Title: Hands-on Rust: question about get_component (page 295) (feel free to respond. “You dug you’re own hole… good luck”) I have somet...
New
herminiotorres
Hi! I know not the intentions behind this narrative when called, on page XI: mount() |&gt; handle_event() |&gt; render() but the correc...
New
patoncrispy
I’m new to Rust and am using this book to learn more as well as to feed my interest in game dev. I’ve just finished the flappy dragon exa...
New
jskubick
I think I might have found a problem involving SwitchCompat, thumbTint, and trackTint. As entered, the SwitchCompat changes color to hol...
New
jskubick
I found an issue in Chapter 7 regarding android:backgroundTint vs app:backgroundTint. How to replicate: load chapter-7 from zipfile i...
New
rainforest
Hi, I’ve got a question about the implementation of PubSub when using a Phoenix.Socket.Transport behaviour rather than channels. Before ...
New
mert
AWDWR 7, page 152, page 153: Hello everyone, I’m a little bit lost on the hotwire part. I didn’t fully understand it. On page 152 @rub...
New
redconfetti
Docker-Machine became part of the Docker Toolbox, which was deprecated in 2020, long after Docker Desktop supported Docker Engine nativel...
New
SlowburnAZ
Getting an error when installing the dependencies at the start of this chapter: could not compile dependency :exla, "mix compile" failed...
New

Other popular topics Top

PragmaticBookshelf
Machine learning can be intimidating, with its reliance on math and algorithms that most programmers don't encounter in their regular wor...
New
PragmaticBookshelf
Learn from the award-winning programming series that inspired the Elixir language, and go on a step-by-step journey through the most impo...
New
PragmaticBookshelf
Write Elixir tests that you can be proud of. Dive into Elixir’s test philosophy and gain mastery over the terminology and concepts that u...
New
brentjanderson
Bought the Moonlander mechanical keyboard. Cherry Brown MX switches. Arms and wrists have been hurting enough that it’s time I did someth...
New
AstonJ
Continuing the discussion from Thinking about learning Crystal, let’s discuss - I was wondering which languages don’t GC - maybe we can c...
New
rustkas
Intensively researching Erlang books and additional resources on it, I have found that the topic of using Regular Expressions is either c...
New
PragmaticBookshelf
Use WebRTC to build web applications that stream media and data in real time directly from one user to another, all in the browser. ...
New
AstonJ
Saw this on TikTok of all places! :lol: Anyone heard of them before? Lite:
New
AstonJ
We’ve talked about his book briefly here but it is quickly becoming obsolete - so he’s decided to create a series of 7 podcasts, the firs...
New
New

Sub Categories: