culebra.trainer module

Implementation of some trainers.

This module is composed by:

  • The abc module, which defines some abstract base classes to implement several types of trainers.

  • The ea module, which implements several evolutionary algorithms-based trainers.

  • The aco module, which supports several ant colony optimization approaches.

  • The topology sub-module, which provides several tolpologies for distributed trainers.

Attributes

DEFAULT_NUM_SUBTRAINERS = 1

Default number of subtrainers.

DEFAULT_REPRESENTATION_SIZE = 5

Default value for the number of representatives sent to the other subtrainers.

DEFAULT_REPRESENTATION_FREQ = 10

Default value for the number of generations between representatives sending.

DEFAULT_REPRESENTATION_SELECTION_FUNC = <function selTournament>

Default selection policy function to choose the representatives.

DEFAULT_REPRESENTATION_SELECTION_FUNC_PARAMS = {'tournsize': 3}

Default parameters for the representatives selection policy function.

DEFAULT_ISLANDS_REPRESENTATION_TOPOLOGY_FUNC = <function ring_destinations>

Default topology function for the islands model.

DEFAULT_ISLANDS_REPRESENTATION_TOPOLOGY_FUNC_PARAMS = {}

Parameters for the default topology function in the islands model.