geppy.algorithms package

Submodules

geppy.algorithms.basic module

This basic module provides fundamental boilerplate GEP algorithm implementations. After registering proper operations into a deap.base.Toolbox object, the GEP evolution can be simply launched using the present algorithms. Of course, for complicated problems, you may want to define your own algorithms, and the implementation here can be used as a reference.

geppy.algorithms.basic.gep_simple(population, toolbox, n_generations=100, n_elites=1, stats=None, hall_of_fame=None, verbose=True)[source]

This algorithm performs the simplest and standard gene expression programming. The flowchart of this algorithm can be found here. Refer to Chapter 3 of [FC2006] to learn more about this basic algorithm.

Note

The algorithm framework also supports the GEP-RNC algorithm, which evolves genes with an additional Dc domain for random numerical constant manipulation. To adopt gep_simple() for GEP-RNC evolution, use the GeneDc objects as the genes and register Dc-specific operators. A detailed example of GEP-RNC can be found at numerical expression inference with GEP-RNC. Users can refer to Chapter 5 of [FC2006] to get familiar with the GEP-RNC theory.

Parameters:
  • population – a list of individuals
  • toolbox – Toolbox, a container of operators. Regarding the conventions of operator design and registration, please refer to Conventions of genetic operator design and registration in geppy.
  • n_generations – max number of generations to be evolved
  • n_elites – number of elites to be cloned to next generation
  • stats – a Statistics object that is updated inplace, optional.
  • hall_of_fame – a HallOfFame object that will contain the best individuals, optional.
  • verbose – whether or not to print the statistics.
Returns:

The final population
Returns:

A Logbook recording the statistics of the evolution process

geppy.algorithms.copt module

This copt module provides advanced routines for numerical constant optimization in gene expression programming. It is known that both genetic programming (GP) and GEP can usually find a near-optimal solution quite quickly. However, they both have difficulty in optimizing the numerical constants integrated in the model, especially when the constants are continuous real numbers. To optimize the constants efficiently, a local optimizer is often incorporated, i.e., GEP mainly aims to formulate the true structure of the model, while the embedded local optimizer is dedicated to improving the constants existent in the individuals.

To reduce computational cost, such local optimization is applied every k generations and often only applied to selected individuals. Accordingly, two arguments opt_period and opt_selector are required by the algorithms.

Such kind of algorithms combining evolutionary algorithms and local improvement procedures are often referred to as memetic algorithm in literature.

A common interface for a local optimizer is optimize(population, selector, generation, **kwargs) -> population.

geppy.algorithms.copt.gep_hill_climbing(population, toolbox, n_generations=100, n_elites=1, stats=None, hall_of_fame=None, verbose=True, opt_period=30, opt_selector='best 10', max_step=1, rnc_gen=None)[source]
geppy.algorithms.copt.gep_opt(population, toolbox, n_generations=100, n_elites=1, stats=None, hall_of_fame=None, verbose=True, optimizer=None, opt_period=None, opt_selector=None, **kwargs)[source]

Module contents