Pyage is agent-based platform for running evolutionary computations, written in python.
Pyage runs on Python 2.X.
Pyage is hosted on pypi, so you can install it using easy_install or pip command-line tool.
pip install pyage
Verify your installation with: python -m pyage.core.bootstrap pyage.conf.test_conf It should run an example config built into pyage. Inspect pyage logfile for any errors.
To run EMAS, use pyage.conf.femas_single as the parameter, but make sure to set AGENTS environment variable to indicate how many aggregate-agents (islands) you want to run.
Configuration module can be located anywhere in your system, so you can write your own config files. Config module should be visible on PYTHONPATH, and passed as argument to pyage.core.bootstrap (see above example).
To create your own configuration, you should start with copy of one of built-in configuration sets, then you can play with parameters.
Example configuration for EMAS would be:
from pyage.core import address
from pyage.core.agent.agent import unnamed_agents
from pyage.core.agent.aggregate import AggregateAgent
from pyage.core.emas import EmasService
from pyage.core.locator import GridParentLocator
from pyage.core.migration import ParentMigration
from pyage.core.stats.gnuplot import TimeStatistics, StepStatistics
from pyage.core.stop_condition import StepLimitStopCondition
from pyage.solutions.evolution.crossover import SinglePointCrossover
from pyage.solutions.evolution.evaluation import FloatRastriginEvaluation
from pyage.solutions.evolution.initializer import float_emas_initializer
from pyage.solutions.evolution.mutation import UniformFloatMutation
agents_count = lambda: 5
agents = unnamed_agents(agents_count, AggregateAgent)
stop_condition = lambda: StepLimitStopCondition(1000)
aggregated_agents = lambda: float_emas_initializer(40, energy=100, size=50, lowerbound=-10, upperbound=10)
emas = EmasService
minimal_energy = lambda: 0
reproduction_minimum = lambda: 90
migration_minimum = lambda: 120
newborn_energy = lambda: 100
transferred_energy = lambda: 40
evaluation = FloatRastriginEvaluation
crossover = SinglePointCrossover
mutation = lambda: UniformFloatMutation(probability=1, radius=1)
address_provider = address.SequenceAddressProvider
migration = ParentMigration
locator = GridParentLocator
stats = StepStatistics
Parameters you may want to change:
- agents_count is the number of islands
- float_emas_initializer(40, energy=100, size=50, lowerbound=-10, upperbound=10) determines initial configuration inside every islands: 40 is size of the genotype, energy is initial value of agents energy, size is equal to number of agents initially placed on each island, and last two parameters denote range for random initial values into genotypes