Installation

RL4CO is now available for installation on pip!

pip install rl4co

Local install and development

If you want to develop RL4CO or access the latest builds, you may install it locally after downloading the repo:

git clone https://github.com/ai4co/rl4co && cd rl4co

The simplest way is via pip in editable mode with

pip install -e .

To install optional dependencies, you may specify them as follows pip install -e ".[dev,graph,routing,docs]".

We recommend installing in virtual environments with a package manager such as the blazing-fast uv, poetry, or conda, with quickstart commands below:

Install with `uv`

You first need to install `uv`, i.e., with `pip`:

pip install uv

Then, you can create a virtual environment locally and activate it:

uv sync
source .venv/bin/activate

Note that `uv` directly generates the `.venv` folder in the current directory. To install (all) extras, you may use `uv sync --all-extras` or specify them individually with `uv sync --extra dev --extra graph --extra routing --extra docs`.

Install with `poetry`

Make sure that you have `poetry` installed from the [official website](https://python-poetry.org/docs/). Then, you can create a virtual environment locally:

poetry install
poetry env activate # poetry shell removed in poetry 2.0.0

Note: you need to upgrade `poetry` to the latest version with `poetry self update` to versions >=2.0.0 (see [blog post](https://python-poetry.org/blog/announcing-poetry-2.0.0/)). This is also the reason why we don't need a special `pyproject.toml` anymore.

Install with `conda`

After [installing `conda`](https://docs.conda.io/projects/conda/en/latest/user-guide/install/index.html), you can create a virtual environment locally with:

conda create -n rl4co python=3.12
conda activate rl4co

Minimalistic Example

Here is a minimalistic example training the Attention Model with greedy rollout baseline on TSP in less than 30 lines of code:

from rl4co.envs.routing import TSPEnv, TSPGenerator
from rl4co.models import AttentionModelPolicy, POMO
from rl4co.utils import RL4COTrainer

# Instantiate generator and environment
generator = TSPGenerator(num_loc=50, loc_distribution="uniform")
env = TSPEnv(generator)

# Create policy and RL model
policy = AttentionModelPolicy(env_name=env.name, num_encoder_layers=6)
model = POMO(env, policy, batch_size=64, optimizer_kwargs={"lr": 1e-4})

# Instantiate Trainer and fit
trainer = RL4COTrainer(max_epochs=10, accelerator="gpu", precision="16-mixed")
trainer.fit(model)

Tip

We recommend checking out our quickstart notebook!