Evolution Strategies with Parent Replacement and Recombination

Summary

The document explains two population-based optimization methods: (μ,λ)-ES, where offspring replace their parents, and (μ+λ)-ES, where parents and offspring compete for the next generation. Both use mutation and selection to search a real-valued parameter space. The article also modifies the methods to recombine coordinates drawn from different parents, letting offspring inherit from multiple candidates rather than developing independently.

It presents algorithm descriptions and reports experimental comparisons, characterizing both approaches as useful across varied optimization problems while noting a wide spread in results on discrete functions. The discussion is about general numerical optimization implemented in MQL5, not trading strategy performance. Its conclusions are based on the author’s experiments, and the article notes that its implementations modify canonical algorithms, so results should not be treated as a definitive comparison of standard versions.

Key ideas

• In (μ,λ)-ES, only offspring can become parents in the next generation.
• In (μ+λ)-ES, parents and offspring compete for a limited number of survivor places.
• Gaussian mutation changes candidate parameter vectors during the search.
• Recombination can combine coordinates from multiple parents to encourage diversity.
• The reported experimental conclusions are limited to the tested implementations and functions.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.