NEORL

NEORL (NeuroEvolution Optimization with Reinforcement Learning) is a set of implementations of hybrid algorithms combining neural networks and evolutionary computation based on a wide range of machine learning and evolutionary intelligence architectures. NEORL aims to solve large-scale optimization problems relevant to operation & optimization research, engineering, business, and other disciplines.

NEORL can be used for multidisciplinary applications for research, industrial, academic, and/or teaching purposes. NEORL can be used as a standalone platform or an additional benchmarking tool to supplement or validate other optimization packages. Our objective when we built NEORL is to give the user a simple and easy-to-use framework with an access to a wide range of algorithms, covering both standalone and hybrid algorithms in evolutionary, swarm, supervised learning, deep learning, and reinforcement learning. We hope NEORL will allow beginners to enjoy more advanced optimization and algorithms, without being involved in too many theoretical/implementation details, and give experts an opportunity to solve large-scale optimization problems.

Documentation

Documentation is available online: https://neorl.readthedocs.io/en/latest/index.html

The framework paper is available online: https://arxiv.org/abs/2112.07057

Copyright

This repository and its content are copyright of Exelon Corporation © in collaboration with MIT Nuclear Science and Engineering 2021. All rights reserved.

You can read the first successful and the baseline application of NEORL for nuclear fuel optimization in this News Article.

Basic Features

| Features | NEORL
| -----------------------------------------| ----------------------------------- | Reinforcement Learning (standalone) | :heavy_check_mark: | | Evolutionary Computation (standalone) | :heavy_check_mark: | | Hybrid Neuroevolution | :heavy_check_mark: | | Supervised Learning | :heavy_check_mark: | | Parallel processing | :heavy_check_mark: | | Combinatorial/Discrete Optimization | :heavy_check_mark: | | Continuous Optimization | :heavy_check_mark: | | Mixed Discrete/Continuous Optimization | :heavy_check_mark: | | Hyperparameter Tuning | :heavy_check_mark: | | Ipython / Notebook friendly | :heavy_check_mark: | | Detailed Documentation | :heavy_check_mark: | | Advanced logging | :heavy_check_mark: | | Optimization Benchmarks | :heavy_check_mark: |

Knowledge Prerequisites

Note: despite the simplicity of NEORL usage, most algorithms, especially the neuro-based, need some basic knowledge about the optimization research and neural networks in supervised and reinforcement learning. Using NEORL without sufficient knowledge may lead to undesirable results due to the poor selection of algorithm hyperparameters. You should not utilize this package without basic knowledge in machine learning and optimization.

Safe Installation (Strongly Recommended)

Safe installation will setup NEORL in a separate virtual environment with its own dependencies. This eliminates any conflict with your existing package versions (e.g. numpy, Tensorflow).

To install on Linux, here are the steps:

https://neorl.readthedocs.io/en/latest/guide/detinstall.html

For Windows, the steps can be found here:

https://neorl.readthedocs.io/en/latest/guide/detinstall.html#windows-10

Quick Installation

For both Ubuntu and Windows, you can install NEORL via pip

pip install neorl

However, we strongly recommend following safe installation steps to avoid any conflict between NEORL dependencies (e.g. TensorFlow) and your current Python packages.

Testing NEORL Installation

Upon successful installation, NEORL offers a robust unit test package to test all algorithms, you can run the tests via terminal using

neorl --test

All unit tests in NEORL can be executed using pytest runner. If pytest is not installed, please use pip install pytest pytest-cov before running the tests.

Example

Here is a quick example of how to use NEORL to minimize a 5-D sphere function: ```python

---------------------------------

Import packages

---------------------------------

import numpy as np import matplotlib.pyplot as plt from neorl import DE, XNES

---------------------------------

Fitness

---------------------------------

Define the fitness function

def FIT(individual): """Sphere test objective function. F(x) = sum_{i=1}^d xi^2 d=1,2,3,... Range: [-100,100] Minima: 0 """

return sum(x**2 for x in individual)

---------------------------------

Parameter Space

---------------------------------

Setup the parameter space (d=5)

nx=5 BOUNDS={} for i in range(1,nx+1): BOUNDS['x'+str(i)]=['float', -100, 100]

---------------------------------

DE

---------------------------------

de=DE(mode='min', bounds=BOUNDS, fit=FIT, npop=50, CR=0.5, F=0.7, ncores=1, seed=1) x_best, y_best, de_hist=de.evolute(ngen=120, verbose=0) print('---DE Results---', ) print('x:', x_best) print('y:', y_best)

---------------------------------

NES

---------------------------------

x0=[-50]*len(BOUNDS) amat = np.eye(nx) xnes=XNES(mode='min', bounds=BOUNDS, fit=FIT, npop=50, eta_mu=0.9, eta_sigma=0.5, adapt_sampling=True, seed=1) x_best, y_best, nes_hist=xnes.evolute(120, x0=x0, verbose=0) print('---XNES Results---', ) print('x:', x_best) print('y:', y_best)

---------------------------------

Plot

---------------------------------

Plot fitness for both methods

plt.figure() plt.plot(np.array(de_hist), label='DE') plt.plot(np.array(nes_hist['fitness']), label='NES') plt.xlabel('Generation') plt.ylabel('Fitness') plt.legend() plt.show() ```

Implemented Algorithms

NEORL offers a wide range of algorithms, where some algorithms could be used with a specific parameter space.

| Algorithm | Discrete Space | Continuous Space| Mixed Space | Multiprocessing|
| ------------------- | ------------------ | ------------------ | ------------------ | ------------------ | | ACER | :heavy_check_mark: | :x: | :x: | :heavy_check_mark: | | ACKTR | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | A2C | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | PPO | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | DQN | :heavy_check_mark: | :x: | :x: | :x: | | ES | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | PSO | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | DE | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | XNES | :x: | :heavy_check_mark: | :x: | :heavy_check_mark: | | GWO | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | PESA | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | PESA2 | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | RNEAT | :x: | :heavy_check_mark: | :x: | :heavy_check_mark: | | FNEAT | :x: | :heavy_check_mark: | :x: | :heavy_check_mark: | | SA | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | SSA | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | WOA | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | JAYA | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | MFO | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | HHO | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | BAT | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | PPO-ES | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | ACKTR-DE | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | ACO | :x: | :heavy_check_mark: | :x: | :heavy_check_mark: | | NGA | :x: | :heavy_check_mark: | :x: | :x: | | NHHO | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | CS | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | | TS | :heavy_check_mark: | :heavy_check_mark: | :x: | :x: |

Major Founding Papers of NEORL

1- Radaideh, M. I., Wolverton, I., Joseph, J., Tusar, J. J., Otgonbaatar, U., Roy, N., Forget, B., Shirvan, K. (2021). Physics-informed reinforcement learning optimization of nuclear assembly design. Nuclear Engineering and Design, 372, p. 110966.

2- Radaideh, M. I., Shirvan, K. (2021). Rule-based reinforcement learning methodology to inform evolutionary algorithms for constrained optimization of engineering applications. Knowledge-Based Systems, 217, p. 106836.

3- Radaideh, M. I., Forget, B., & Shirvan, K. (2021). Large-scale design optimisation of boiling water reactor bundles with neuroevolution. Annals of Nuclear Energy, 160, p. 108355.

Citing the Project

To cite this repository in publications:

@article{radaideh2021neorl, title={NEORL: NeuroEvolution Optimization with Reinforcement Learning}, author={Radaideh, Majdi I and Du, Katelin and Seurin, Paul and Seyler, Devin and Gu, Xubo and Wang, Haijia and Shirvan, Koroush}, journal={arXiv preprint arXiv:2112.07057}, year={2021} }

Maintainers

See our team here Contributors. We are welcoming new contributors to the project.

Important Note: We do not do technical support and we do not answer personal questions via email.

Acknowledgments

NEORL was established in MIT back to 2020 with feedback, validation, and usage of different colleagues: Issac Wolverton (MIT Quest for Intelligence), Joshua Joseph (MIT Quest for Intelligence), Benoit Forget (MIT Nuclear Science and Engineering), Ugi Otgonbaatar (Exelon Corporation), and James Tusar (Exelon Corporation). We also thank our fellows at Stable Baselines, DEAP, and EvoloPy for sharing their implementation, which inspired us to leverage our optimization classes.

Issues

-bash: fork: retry: Resource temporarily unavailable...NEORL starting many sleeping processes

opened on 2021-12-21 13:48:13 by deanrp2

This post is for users who may encounter: -bash: fork: retry: Resource temporarily unavailable or Segmentation fault (core dumped) nohup python de_expl.py errors.

On linux computers, users often have a maximum number of processes they are allowed to have running on a computer. This can be checked with the ulimit -a command under the max user processes row. For example: ```

ulimit -a core file size (blocks, -c) 0 data seg size (kbytes, -d) unlimited scheduling priority (-e) 0 file size (blocks, -f) unlimited pending signals (-i) 1028858 max locked memory (kbytes, -l) 64 max memory size (kbytes, -m) unlimited open files (-n) 1024 pipe size (512 bytes, -p) 8 POSIX message queues (bytes, -q) 819200 real-time priority (-r) 0 stack size (kbytes, -s) 8192 cpu time (seconds, -t) unlimited max user processes (-u) 4096 virtual memory (kbytes, -v) unlimited file locks (-x) unlimited ```

A user can check how many processes (in total) they have running with the command: ps --no-headers auxwwwm | awk '$2 == "-" { print $1 }' | sort | uniq -c | sort -n. If a user wants to see the specific listing: ps --no-headers auxwwwm.

A typical Python program may start <50 processes. I am not exactly sure why this is the case but I checked a few different random scripts I had lying around and this is the conclusion I came to.

For some reason, when running NEORL in serial, around 300 processes are started. I think this has something to do with parallelization. Most of the processes are sleeping, for whatever reason.

This becomes a problem if a user wants to run multiple independent python programs which use NEORL. Regardless of computer size, the process limit is quickly reached. A fix to this is to simply raise the max number of user processes: ulimit -u ####. But this cannot be done without sudo access.

I do not think this is an urgent problem for NEORL as it only comes up in a specific use case but I wanted to post this to provide information to users who encounter the same problem.

Possibly relevant links: https://stackoverflow.com/questions/20614309/find-reason-for-sleeping-python-process https://stackoverflow.com/questions/31193449/python-multiprocessing-big-data-turn-process-into-sleep https://stackoverflow.com/questions/1032813/dump-stacktraces-of-all-active-threads/7317379#7317379

Objective function memoization

opened on 2021-10-11 19:22:39 by deanrp2

Many algorithms may require multiple identical evaluations of the objective functions due to members of a population remaining the same between generations. Consider looking into automatic cacheing of objective function calls. See here for an example.

Releases

NEORL-1.5.2b 2021-08-11 00:22:24