Machine Learning: Genetic Algorithms Part 1 (Javascript)

The post introduces genetic algorithms as a machine learning approach that is less practical than many others, but useful for showing cost functions, local and global optima, crossover, and mutation. It uses a contrived Javascript example that tries to reproduce “Hello, World!” and explains the basic pieces of a genetic algorithm through that task.

Reading notes

#

• Genetic algorithms generate many answer candidates and score them with a cost or fitness function.
• In the example, the chromosome is a 13-character string that should match “Hello, World!”.
• The cost function compares each character’s ASCII value to the target and squares the difference.
• Mating, or crossover, combines two chromosomes by splitting them at a pivot point and swapping halves.
• Mutation changes a random character so the search does not stay trapped in a local optimum.
• A population is a group of chromosomes that goes through generations of scoring, sorting, killing weaker members, mating stronger ones, mutating, and checking for completion.
• The example starts from random strings and stops when the cost reaches 0.
• The code defines Gene and Population objects with methods for random generation, cost calculation, mating, mutation, sorting, and advancing generations.