Karaboga's Artificial Bee Colony in Rust (and in parallel!)

Documentation, including sample code.

There's also a demo, which animates the algorithm for your viewing pleasure. It also uses a couple useful techniques for applying the algorithm to a complex, persistent state.

Available from crates.io as abc.

The Artificial Bee Colony is an optimization algorithm. It considers a set of solution candidates by sending conceptual "bees" to work on those solutions. There are three kinds of bee:

• One worker bee is dedicated to each solution. Each time a worker bee runs, it looks at a solution near the current one, keeping each improvement.
• Observer bees are like workers, but are not dedicated to a single solution. Instead, they look at all of the active solutions and choose one randomly to work on. Observers usually prefer to work on higher-fitness solutions.
• A worker or observer whose solution appears to be a local maximum becomes a scout. Scouting entails generating a new, random solution to break out of a rut.

The ABC algorithm can be -- and has been -- applied to a variety of applications. As with many such algorithms, the logic is fairly agnostic about the domain that it works in. In fact, the prerequisites for using the algorithm are:

• a data structure with a solution,
• a way of generating new, random solutions,
• a way of generating solutions "near" an existing solution, and
• a fitness function to score solutions.

A solution could be a game-playing AI, a blueprint for a building, or just a point in space, and the abc crate treats them all alike. Simply implement the Context trait for a type of your choice, construct a Hive, and start running.

Speaking of running,abc supports two run modes:

• running for a fixed number of rounds and returning the best solution,
• running continuously until stopped, or
• running continuously in the background and sending each improved solution over a Rust channel.

The abc crate takes advantage of Rust's excellent concurrency support to explore the same space. This means that heavy computation can be distributed across multiple CPU cores, or I/O-bound evaluation can run without blocking. The hive maintains a queue of bees, and the threads each take bees from the queue and apply the bees' logic to the solutions. So, at a given moment, there is a different bee working in each thread.