Skip to content

The Implementation of Reinforcement Learning Algorithms with Negatively correlated search framework. It contains four algorithms: NCS-C, NCSCC, NCSRE, NCNES. It based on OpenAI CES API and details will be showed on the readme.md.

Notifications You must be signed in to change notification settings

Desein-Yang/NCS-RL

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

README

This is a library of Evolutionary Reinforcement Learning(ERL) algorithms. The library contains a series of evolutionary algorithms based on negative correlation algorithms applied to reinforcement learning strategy optimization.

  • NCS(Negative Correlated Search) Framework
    • NCS-C
    • NCSCC
    • NCNES
    • NCSRE

We implement and encapsulate the reusable code and abstract modules for evolutionary optimization algorithms, neural network parameter optimization, reinforcement learning. You can refer to the examples dir to modify/implement a parallel EA or RL quickly.

Features:

  • large-scale parallel communication (backend: openMPI)
  • evolution strategy framework
  • parallel RL rollout (gym)

Features for special algorithm:

  • negative correlation search utils
  • decision variables decomposition
  • decision variables random embedding

Usage

Dependency

  • mpi4py (need mpirun environment)
  • tensorflow=1.15
  • gym=0.9.1
  • click
  • numpy
  • opencv-python

NCSCC

mpirun -np cpus python main.py [-v][-r][-e][-g][-f][-d][--epoch][--sigma0][--rvalue]
--version, -v     
--run_name, -r  
--env_type, -e   
--game,-g         
--function_id, -f 
--dimension, -d   
--k, -k       
--epoch,     
--sigma0,  
--rvalue 

NCSRE

mpirun -np cpus python NCSRE.py --[hyperparameter]

NCNES

mpirun -np cpus python NCNES.py --[hyperparameter]

NCS-C

mpirun -n cpus python NCS.py [-e][-g][-c][-r]
-n the num of cpus  
-e the num of individuals on 1 cpu
-g the name of benchmark (gamename in atari benchmark)
-r the name of log file
-c the configuration files default =./configurations/sample_configuration.json

Files Tree

src/ base.py main class of EA algorithm for NN parameters optimization models.py Definition of Neural Network Models of policy policy.py Definition of RL policy (with rollout) decomposer.py utils for decision varibles decomposition (CC) env_wrappers.py utils for env preprocess in gym
ops.py utils for RL policy building testfunc.py utils for CEC benchmarks (test EA algorithms) logger.py utils for logging replay.py utils for replay buffer data/
test/
scripts/
examples/

Citation

If the repo is useful for you, please cite the paper as

@incollection{yang2021,
    author = {Yang, Qi and Yang, Peng and Tang, Ke},
	title = {Parallel Random Embedding with Negatively Correlated Search},
	volume = {12690},
    year = {2021},
	doi = {10.1007/978-3-030-78811-7_33}
	isbn = {978-3-030-78810-0 978-3-030-78811-7},
	pages = {339--351},
	booktitle = {Advances in Swarm Intelligence},
	publisher = {Springer International Publishing},
	editor = {Tan, Ying and Shi, Yuhui},
}

About

The Implementation of Reinforcement Learning Algorithms with Negatively correlated search framework. It contains four algorithms: NCS-C, NCSCC, NCSRE, NCNES. It based on OpenAI CES API and details will be showed on the readme.md.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages