Research paper code for distributed evolution strategies
Top 26.8% on sourcepulse
This repository provides a distributed implementation of Evolution Strategies (ES) as a scalable alternative to Reinforcement Learning, as detailed in the paper "Evolution Strategies as a Scalable Alternative to Reinforcement Learning." It is targeted at researchers and engineers exploring advanced optimization techniques for complex control problems, offering a robust master-worker architecture for parallel computation.
How It Works
The implementation employs a master-worker architecture. The master node distributes current parameters to multiple worker nodes, which then compute gradients or performance metrics. Workers return these results to the master, enabling iterative parameter updates. This design is advantageous for large-scale parallelization, allowing efficient exploration of the parameter space.
Quick Start & Requirements
scripts/packer.json
and launch experiments with scripts/launch.py
.scripts/dependency.sh
), Packer.Highlighted Details
Maintenance & Community
This project is archived and no longer actively maintained or updated.
Licensing & Compatibility
The repository's license is not explicitly stated in the README. Compatibility for commercial use or closed-source linking is not specified.
Limitations & Caveats
The project is archived and provided as-is, with no expected updates. It requires a user-provided Mujoco license and manual AMI building, indicating a significant setup overhead and potential compatibility issues with modern environments.
5 years ago
1 day