Avoiding Excess Computation in Asynchronous Evolutionary Algorithms

This repo contains code for part of the experiments described in our 2021 paper,

• Eric O. Scott, Mark Coletti, Catherine D. Schuman, Bill Kay, Shruti R. Kulkarni, and Maryma Parsa. "Avoiding Excess Computation in Asynchronous Evolutionary Algorithms," 20th UK Workshop on Computational Intelligence.

The project consists of a Python module, async_sim.components, which provides objects that are composed into experiment applications found under examples.

Our algorithms are built atop v0.7dev of the LEAP library (see leap_ec on PyPI).

The scripts here will generate live telemetry plots (these differ from the plots in the paper, which were computed offline from the outputted CSV data):

Setup

Optionally set up a virtual environment (ex. via python -m venv ./venv && source venv/bin/activate).

Then install the package:

pip install -e .

Ensure the tests pass:

pip install pytest
pytest

Run Experiments

python examples/async_simulation.py
python examples/takeover_times.py