Evolution Strategy (ES) is an optimisation technique based on ideas of adaptation, evolution, mutation, and breeding. It tries to simulate the natural process of selection, and encompasses the survival of the fittest ideaolgy to train an artificial neural network (a fundamentally different appraoch than Gradient Descent) using reinforcement learning.
You can read more about it in a paper by openai.
This implementation can be used to train any model built using Keras api, and openai/gym like environments.
From the Github Repository:
$ pip install git+https://github.com/ViRu-ThE-ViRuS/Evolve.git
-
use python3
-
install dependencies from requirements.txt:
$ pip install -r requirements.txt
-
can be found in example.py:
$ python example.py
- documentation is in progress
- run setup:
$ ./scripts/setup.sh $ source venv/bin/activate
- run checks:
$ ./scripts/check.sh - run automated fixes:
$ ./scripts/fix.sh