---

🌟Distributed GPU-accelerated Framework for Scalable Evolutionary Computation🌟

---

Table of Contents

1. Overview
2. Key Features
3. Main Contents
4. Installation Guide
5. Quick Start
6. Sister Projects
7. Community & Support

Overview

EvoX is a distributed GPU-accelerated evolutionary computation framework compatible with PyTorch. With a user-friendly programming model, it offers a comprehensive suite of 50+ Evolutionary Algorithms (EAs) and a wide range of 100+ Benchmark Problems/Environments. For more details, please refer to our Paper and Documentation / 文档.

Note

Users of the previous JAX-based version can access it on the v0.9.0 branch.

Key Features

💻 High-Performance Computing

🚀 Ultra Performance

• Supports acceleration on heterogeneous hardware, including both CPUs and GPUs, achieving over 100x speedups.
• Integrates distributed workflows that scale seamlessly across multiple nodes or devices.

🌐 All-in-One Solution

• Includes 50+ algorithms for a wide range of use cases, fully supporting single- and multi-objective optimization.
• Provides a hierarchical architecture for complex tasks such as meta learning, hyperparameter optimization, and neuroevolution.

🛠️ Easy-to-Use Design

• Fully compatible with PyTorch and its ecosystem, simplifying algorithmic development with a tailored programming model.
• Ensures effortless setup with one-click installation for Windows users.

📊 Versatile Benchmarking

📚 Extensive Benchmark Suites

• Features 100+ benchmark problems spanning single-objective optimization, multi-objective optimization, and real-world engineering challenges.

🎮 Support for Physics Engines

• Integrates seamlessly with physics engines like Brax and other popular frameworks for reinforcement learning.

⚙️ Customizable Problems

• Provides an encapsulated module for defining and evaluating custom problems tailored to user needs, with seamless integration into real-world applications and datasets.

📈 Flexible Visualization

🔍 Ready-to-Use Tools

• Offers a comprehensive set of visualization tools for analyzing evolutionary processes across various tasks.

🛠️ Customizable Modules

• Enables users to integrate their own visualization code, allowing for tailored and flexible visualizations.

📂 Real-Time Data Streaming

• Leverages the tailored .exv format to simplify and accelerate real-time data streaming.

Main Contents

Category	Subcategory	Notable Algorithms / Benchmark Problems
Single-objective Optimization	Differential Evolution	CoDE, JaDE, SaDE, SHADE, IMODE, ...
	Evolution Strategy	CMA-ES, PGPE, OpenES, CR-FM-NES, xNES, ...
	Particle Swarm Optimization	FIPS, CSO, CPSO, CLPSO, SL-PSO, ...
Multi-objective Optimization	Dominance-based	NSGA-II, NSGA-III, SPEA2, BiGE, KnEA, ...
	Decomposition-based	MOEA/D, RVEA, t-DEA, MOEAD-M2M, EAG-MOEAD, ...
	Indicator-based	IBEA, HypE, SRA, MaOEA-IGD, AR-MOEA, ...
Benchmark Problems / Environments	Numerical	DTLZ, LSMOP, MaF, ZDT, CEC'22, ...
	Neuroevolution / RL	Brax, TorchVision Dataset, ...

For a comprehensive list and detailed descriptions of all algorithms, please check the Algorithms API, and for benchmark problems/environments, refer to the Problems API.

Installation Guide

Install evox with default feature sets via pip:

pip install "evox[default]"

Install the latest version from the source code for testing or development:

git clone https://github.com/EMI-Group/evox.git
cd evox
pip install -e .

Tip

Windows users can use the win-install.bat script for installation.

Quick Start

Here are some examples to get you started with EvoX:

Single-objective Optimization

Solve the Ackley problem using the PSO algorithm:

import torch
from evox.algorithms import PSO
from evox.problems.numerical import Ackley
from evox.workflows import StdWorkflow, EvalMonitor

# torch.set_default_device("cuda") # Uncomment this line if you want to use GPU by default

algorithm = PSO(pop_size=100, lb=-32 * torch.ones(10), ub=32 * torch.ones(10))
problem = Ackley()
monitor = EvalMonitor()
workflow = StdWorkflow(algorithm, problem, monitor)
workflow.init_step()
for i in range(100):
    workflow.step()

monitor.plot() # or monitor.plot().show() if you are using headless mode

Example Output

Multi-objective Optimization

Solve the DTLZ2 problem using the RVEA algorithm:

import torch
from evox.algorithms import RVEA
from evox.problems.numerical import DTLZ2
from evox.workflows import StdWorkflow, EvalMonitor

# torch.set_default_device("cuda") # Uncomment this line if you want to use GPU by default

prob = DTLZ2(m=2)
pf = prob.pf()
algo = RVEA(
    pop_size=100,
    n_objs=2,
    lb=-torch.zeros(12),
    ub=torch.ones(12)
)
monitor = EvalMonitor()
workflow = StdWorkflow(algo, prob, monitor)
workflow.init_step()
for i in range(100):
    workflow.step()

monitor.plot() # or monitor.plot().show() if you are using headless mode

Example Output

Neuroevolution

Evolving a simple MLP model to solve the Brax HalfCheetah environment:

import torch
import torch.nn as nn
from evox.algorithms import PSO
from evox.problems.neuroevolution.brax import BraxProblem
from evox.utils import ParamsAndVector
from evox.workflows import EvalMonitor, StdWorkflow

# torch.set_default_device("cuda") # Uncomment this line if you want to use GPU by default

class SimpleMLP(nn.Module):
    def __init__(self):
        super().__init__()
        # observation space is 17-dim, action space is 6-dim.
        self.features = nn.Sequential(nn.Linear(17, 8), nn.Tanh(), nn.Linear(8, 6))

    def forward(self, x):
        return torch.tanh(self.features(x))

# Initialize the MLP model
model = SimpleMLP()
adapter = ParamsAndVector(dummy_model=model)
# Set the population size
POP_SIZE = 1024
# Get the bound of the PSO algorithm
model_params = dict(model.named_parameters())
pop_center = adapter.to_vector(model_params)
lb = torch.full_like(pop_center, -5)
ub = torch.full_like(pop_center, 5)
# Initialize the PSO, and you can also use any other algorithms
algorithm = PSO(pop_size=POP_SIZE, lb=lb, ub=ub)
# Initialize the Brax problem
problem = BraxProblem(
    policy=model,
    env_name="halfcheetah",
    max_episode_length=1000,
    num_episodes=3,
    pop_size=POP_SIZE,
)
# set an monitor, and it can record the top 3 best fitnesses
monitor = EvalMonitor(topk=3)
# Initiate an workflow
workflow = StdWorkflow(
    algorithm=algorithm,
    problem=problem,
    monitor=monitor,
    opt_direction="max",
    solution_transform=adapter,
)
workflow.init_step()
for i in range(50):
    workflow.step()

monitor.plot() # or monitor.plot().show() if you are using headless mode

Example Output

Note

For comprehensive guidance, please visit our Documentation, where you'll find detailed installation steps, tutorials, practical examples, and complete API references.

Sister Projects

• EvoRL: GPU-accelerated framework for Evolutionary Reinforcement Learning. Check out here.
• EvoGP: GPU-accelerated framework for Genetic Programming. Check out here.
• EvoMO: GPU-accelerated library for Evolutionary Multiobjective Optimization (EMO).Check out here.
• TensorNEAT: Tensorized NeuroEvolution of Augmenting Topologies (NEAT) for GPU Acceleration. Check out here.
• TensorACO: Tensorized Ant Colony Optimization (ACO) for GPU Acceleration. Check out here.
• EvoXBench: A real-world benchmark platform for solving various optimization problems, such as Neural Architecture Search (NAS). It operates without the need for GPUs/PyTorch/TensorFlow and supports multiple programming environments. Check out here.

Stay tuned - more exciting developments are on the way! ✨

Community & Support

• Join discussions on the GitHub Discussion Board.
• Connect via Discord or QQ group (ID: 297969717).
• Visit Offical Website.

Citing EvoX

If EvoX contributes to your research, please cite it:

@article{evox,
  title = {{EvoX}: {A} {Distributed} {GPU}-accelerated {Framework} for {Scalable} {Evolutionary} {Computation}},
  author = {Huang, Beichen and Cheng, Ran and Li, Zhuozhao and Jin, Yaochu and Tan, Kay Chen},
  journal = {IEEE Transactions on Evolutionary Computation},
  year = 2024,
  doi = {10.1109/TEVC.2024.3388550}
}

License Notice

EvoX is licensed under the GNU General Public License v3.0 (GPL-3.0). For full terms and conditions, please refer to the LICENSE file.