liveBook · Manning

Selection

As explained in Self-Improving Agents

An evolutionary search operator that evaluates candidates and chooses among them according to their fitness.

Selection scores candidates by fitness. In the framework, a Signal performs this role, making scoring part of the existing improvement loop. section 4.1.2

In a typical evolutionary search, selection can:

  • Score every candidate in a generation. section 4.3.1
  • Choose the two highest-scoring candidates as parents for crossover. section 4.3.1
  • Check whether the target score has been reached and end the search if it has. section 4.3.1

A toy fitness score

A toy function scores a genome as -abs(genome - TARGET), so candidates closer to TARGET score higher. In the framework equivalent, a Signal measures a mutated artifact. section 4.1.2

Figure 4.2 The same four mechanics, in two vocabularies. Evolution runs down the left column and the framework down the right: variation maps to mutate(), heredity to the parent pointer, selection to the Signal, and time to the search round. Reading across each row is how the evolutionary loop becomes the real primitives, with nothing added but new names.

The figure maps selection to Signal scoring alongside the other mechanics of the evolutionary loop. section 4.1.2

How Self-Improving Agents covers it

How deep it goes
Explains candidate scoring, choosing parents for crossover, and fitness-based termination without specifying a broader selection policy.
Examples
Uses a toy genome fitness function and a Signal measuring a mutated artifact.
What you can do after
Apply selection by scoring candidates and choosing high-scoring parents for crossover.
How it connects
Connects fitness and Signals to evolutionary search, mutation, crossover, and search rounds.

Also explained in

Compare how 6 books explain selection →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Self-Improving Agents. The section links point to the parts of the book it draws on.