BenchMARL

python benchmarl/run.py algorithm=mappo task=vmas/balance

What is BenchMARL 🧐?

BenchMARL is a Multi-Agent Reinforcement Learning (MARL) training library created to enable reproducibility and benchmarking across different MARL algorithms and environments. Its mission is to present a standardized interface that allows easy integration of new algorithms and environments to provide a fair comparison with existing solutions. BenchMARL uses TorchRL as its backend, which grants it high performance and state-of-the-art implementations. It also uses hydra for flexible and modular configuration, and its data reporting is compatible with marl-eval for standardised and statistically strong evaluations.

BenchMARL core design tenets are:

Reproducibility through systematical grounding and standardization of configuration
Standardised and statistically-strong plotting and reporting
Experiments that are independent of the algorithm, environment, and model choices
Breadth over the MARL ecosystem
Easy implementation of new algorithms, environments, and models
Leveraging the know-how and infrastructure of TorchRL, without reinventing the wheel

Why would I BenchMARL 🤔?

Why would you BenchMARL, I see you ask. Well, you can BenchMARL to compare different algorithms, environments, models, to check how your new research compares to existing ones, or if you just want to approach the domain and want to easily take a picture of the landscape.

BenchMARL
- How to use
  - Notebooks
  - Install
  - Run
- Concept
- Fine-tuned public benchmarks
- Reporting and plotting
- Extending
- Configuring
  - Algorithm
  - Task
  - Model
- Features

How to use

Notebooks

Running BenchMARL experiments.

Install

Install TorchRL

You can install TorchRL from PyPi.

pip install torchrl

For more details, or for installing nightly versions, see the TorchRL installation guide.

Install BenchMARL

You can just install it from github

pip install git+https://github.com/facebookresearch/BenchMARL

Or also clone it locally to access the configs and scripts

git clone https://github.com/facebookresearch/BenchMARL.git
pip install -e BenchMARL

Install environments

All enviornment dependencies are optional in BenchMARL and can be installed separately.

VMAS

pip install vmas

PettingZoo

pip install "pettingzoo[all]"

SMACv2

Follow the instructions on the environment repository.

Here is how we install it on linux.

Run

Experiments are launched with a default configuration that can be overridden in many ways. To learn how to customize and override configurations please refer to the configuring section.

Command line

To launch an experiment from the command line you can do

python benchmarl/run.py algorithm=mappo task=vmas/balance

Thanks to hydra, you can run benchmarks as multi-runs like:

python benchmarl/run.py -m algorithm=mappo,qmix,masac task=vmas/balance,vmas/sampling seed=0,1

The default implementation for hydra multi-runs is sequential, but parallel and slurm launchers are also available.

Script

You can also load and launch your experiments from within a script

 experiment = Experiment(
    task=VmasTask.BALANCE.get_from_yaml(),
    algorithm_config=MappoConfig.get_from_yaml(),
    model_config=MlpConfig.get_from_yaml(),
    critic_model_config=MlpConfig.get_from_yaml(),
    seed=0,
    config=ExperimentConfig.get_from_yaml(),
)
experiment.run()

You can also run multiple experiments in a Benchmark.

benchmark = Benchmark(
    algorithm_configs=[
        MappoConfig.get_from_yaml(),
        QmixConfig.get_from_yaml(),
        MasacConfig.get_from_yaml(),
    ],
    tasks=[
        VmasTask.BALANCE.get_from_yaml(),
        VmasTask.SAMPLING.get_from_yaml(),
    ],
    seeds={0, 1},
    experiment_config=ExperimentConfig.get_from_yaml(),
    model_config=MlpConfig.get_from_yaml(),
    critic_model_config=MlpConfig.get_from_yaml(),
)
benchmark.run_sequential()

Concept

The goal of BenchMARL is to bring different MARL environments and algorithms under the same interfaces to enable fair and reproducible comparison and benchmarking. BenchMARL is a full-pipline unified training library with the goal of enabling users to run any comparison they want across our algorithms and tasks in just one line of code. To achieve this, BenchMARL interconnects components from TorchRL, which provides an efficient and reliable backend.

The library has a default configuration for each of its components. While parts of this configuration are supposed to be changed (for example experiment configurations), other parts (such as tasks) should not be changed to allow for reproducibility. To aid in this, each version of BenchMARL is paired to a default configuration.

Let's now introduce each component in the library.

Experiment. An experiment is a training run in which an algorithm, a task, and a model are fixed. Experiments are configured by passing these values alongside a seed and the experiment hyperparameters. The experiment hyperparameters cover both on-policy and off-policy algorithms, discrete and continuous actions, and probabilistic and deterministic policies (as they are agnostic of the algorithm or task used). An experiment can be launched from the command line or from a script. See the run section for more information.

Benchmark. In the library we call benchmark a collection of experiments that can vary in tasks, algorithm, or model. A benchmark shares the same experiment configuration across all of its experiments. Benchmarks allow to compare different MARL components in a standardized way. A benchmark can be launched from the command line or from a script. See the run section for more information.

Algorithms. Algorithms are an ensemble of components (e.g., losss, replay buffer) which determine the training strategy. Here is a table with the currently implemented algorithms in BenchMARL.

Name	On/Off policy	Actor-critic	Full-observability in critic	Action compatibility	Probabilistic actor
MAPPO	On	Yes	Yes	Continuous + Discrete	Yes
IPPO	On	Yes	No	Continuous + Discrete	Yes
MADDPG	Off	Yes	Yes	Continuous	No
IDDPG	Off	Yes	No	Continuous	No
MASAC	Off	Yes	Yes	Continuous + Discrete	Yes
ISAC	Off	Yes	No	Continuous + Discrete	Yes
QMIX	Off	No	NA	Discrete	No
VDN	Off	No	NA	Discrete	No
IQL	Off	No	NA	Discrete	No

Tasks. Tasks are scenarios from a specific environment which constitute the MARL challenge to solve. They differ based on many aspects, here is a table with the current environments in BenchMARL

Environment	Tasks	Cooperation	Global state	Reward function	Action space	Vectorized
VMAS	5	Cooperative + Competitive	No	Shared + Independent + Global	Continuous + Discrete	Yes
SMACv2	15	Cooperative	Yes	Global	Discrete	No
MPE	8	Cooperative + Competitive	Yes	Shared + Independent	Continuous + Discrete	No
SISL	2	Cooperative	No	Shared	Continuous	No

Note

BenchMARL uses the TorchRL MARL API for grouping agents. In competitive environments like MPE, for example, teams will be in different groups. Each group has its own loss, models, buffers, and so on. Parameter sharing options refer to sharing within the group. See the example on creating a custom algorithm for more info.

Models. Models are neural networks used to process data. They can be used as actors (policies) or, when requested, as critics. We provide a set of base models (layers) and a SequenceModel to concatenate different layers. All the models can be used with or without parameter sharing within an agent group. Here is a table of the models implemented in BenchMARL

Name	Decentralized	Centralized with local inputs	Centralized with global input
MLP	Yes	Yes	Yes

And the ones that are work in progress

Name	Decentralized	Centralized with local inputs	Centralized with global input
GNN	Yes	Yes	No
CNN	Yes	Yes	Yes

Fine-tuned public benchmarks

Warning

This section is under a work in progress. We are constantly working on fine-tuning our experiments to enable our users to have access to state-of-the-art benchmarks. If you would like to collaborate in this effort, please reach out to us.

In the fine_tuned folder we are collecting some tested hyperparameters for specific environments to enable users to bootstrap their benchmarking. You can just run the scripts in this folder to automatically use the proposed hyperparameters.

We will tune benchmarks for you and publish the config and benchmarking plots on Wandb publicly

Currently available ones are:

VMAS:

In the following, we report a table of the results:

Environment	Sample efficiency curves (all tasks)	Performance profile	Aggregate scores
VMAS

Reporting and plotting

Reporting and plotting is compatible with marl-eval. If experiment.create_json=True (this is the default in the experiment config) a file named {experiment_name}.json will be created in the experiment output folder with the format of marl-eval. You can load and merge these files using the utils in eval_results to create beautiful plots of your benchmarks. No more struggling with matplotlib and latex!

Extending

One of the core tenets of BenchMARL is allowing users to leverage the existing algorithm and tasks implementations to benchmark their newly proposed solution.

For this reason we expose standard interfaces with simple abstract methods for algorithms, tasks and models. To introduce your solution in the library, you just need to implement the abstract methods exposed by these base classes which use objects from the TorchRL library.

Here is an example on how you can create a custom algorithm .

Here is an example on how you can create a custom task .

Here is an example on how you can create a custom model .

Configuring

As highlighted in the run section, the project can be configured either in the script itself or via hydra. We suggest to read the hydra documentation to get familiar with all its functionalities.

The project can be configured either the script itself or via hydra. Each component in the project has a corresponding yaml configuration in the BenchMARL conf tree. Components' configurations are loaded from these files into python dataclasses that act as schemas for validation of parameter names and types. That way we keep the best of both words: separation of all configuration from code and strong typing for validation! You can also directly load and validate configuration yaml files without using hydra from a script by calling ComponentConfig.get_from_yaml().

Experiment

Experiment configurations are in benchmarl/conf/config.yaml, with the experiment hyperparameters in benchmarl/conf/experiment. Running custom experiments is extremely simplified by the Hydra configurations. The default configuration for the library is contained in the benchmarl/conf folder.

When running an experiment you can override its hyperparameters like so

python benchmarl/run.py task=vmas/balance algorithm=mappo experiment.lr=0.03 experiment.evaluation=true experiment.train_device="cpu"

Experiment hyperparameters are loaded from benchmarl/conf/experiment/base_experiment.yaml into a dataclass ExperimentConfig defining their domain. This makes it so that all and only the parameters expected are loaded with the right types. You can also directly load them from a script by calling ExperimentConfig.get_from_yaml().

Here is an example of overriding experiment hyperparameters from hydra or from a script .

Algorithm

You can override an algorithm configuration when launching BenchMARL.

python benchmarl/run.py task=vmas/balance algorithm=masac algorithm.num_qvalue_nets=3 algorithm.target_entropy=auto algorithm.share_param_critic=true

Available algorithms and their default configs can be found at benchmarl/conf/algorithm. They are loaded into a dataclass AlgorithmConfig, present for each algorithm, defining their domain. This makes it so that all and only the parameters expected are loaded with the right types. You can also directly load them from a script by calling YourAlgorithmConfig.get_from_yaml().

Here is an example of overriding algorithm hyperparameters from hydra or from a script .

Task

You can override a task configuration when launching BenchMARL. However this is not recommended for benchmarking as tasks should have fixed version and parameters for reproducibility.

python benchmarl/run.py task=vmas/balance algorithm=mappo task.n_agents=4

Available tasks and their default configs can be found at benchmarl/conf/task. They are loaded into a dataclass TaskConfig, defining their domain. Tasks are enumerations under the environment name. For example, VmasTask.NAVIGATION represents the navigation task in the VMAS simulator. This allows autocompletion and seeing all available tasks at once. You can also directly load them from a script by calling YourEnvTask.TASK_NAME.get_from_yaml().

Here is an example of overriding task hyperparameters from hydra or from a script .

Model

You can override the model configuration when launching BenchMARL. By default an MLP model will be loaded with the default config. You can change it like so:

python benchmarl/run.py task=vmas/balance algorithm=mappo model=layers/mlp model=layers/mlp model.layer_class="torch.nn.Linear" "model.num_cells=[32,32]" model.activation_class="torch.nn.ReLU"

Available models and their configs can be found at benchmarl/conf/model/layers. They are loaded into a dataclass ModelConfig, defining their domain. You can also directly load them from a script by calling YourModelConfig.get_from_yaml().

Here is an example of overriding model hyperparameters from hydra or from a script .

Sequence model

You can compose layers into a sequence model. Available layer names are in the benchmarl/conf/model/layers folder.

python benchmarl/run.py task=vmas/balance algorithm=mappo model=sequence "model.intermediate_sizes=[256]" "model/layers@model.layers.l1=mlp" "model/layers@model.layers.l2=mlp" "+model/layers@model.layers.l3=mlp" "model.layers.l3.num_cells=[3]"

Add a layer with "+model/layers@model.layers.l3=mlp".

Remove a layer with "~model.layers.l2".

Configure a layer with "model.layers.l1.num_cells=[3]".

Here is an example of creating a sequence model from hydra or from a script .

Features

BenchMARL has several features:

A test CI with integration and training test routines that are run for all simulators and algorithms
Integration in the official TorchRL ecosystem for dedicated support

Logging

BenchMARL is compatible with the TorchRL loggers. A list of logger names can be provided in the experiment config. Example of available options are: wandb, csv, mflow, tensorboard or any other option available in TorchRL. You can specify the loggers in the yaml config files or in the script arguments like so:

python benchmarl/run.py algorithm=mappo task=vmas/balance "experiment.loggers=[wandb]"

The wandb logger is fully compatible with experiment restoring and will automatically resume the run of the loaded experiment.

Checkpointing

Experiments can be checkpointed every experiment.checkpoint_interval collected frames. Experiments will use an output folder for logging and checkpointing which can be specified in experiment.save_folder. If this is left unspecified, the default will be the hydra output folder (if using hydra) or (otherwise) the current directory where the script is launched. The output folder will contain a folder for each experiment with the corresponding experiment name. Their checkpoints will be stored in a "checkpoints" folder within the experiment folder.

python benchmarl/run.py task=vmas/balance algorithm=mappo experiment.max_n_iters=3 experiment.on_policy_collected_frames_per_batch=100 experiment.checkpoint_interval=100

To load from a checkpoint, pass the absolute checkpoint file name to experiment.restore_file.

python benchmarl/run.py task=vmas/balance algorithm=mappo experiment.max_n_iters=6 experiment.on_policy_collected_frames_per_batch=100 experiment.restore_file="/hydra/experiment/folder/checkpoint/checkpoint_300.pt"

Callbacks

Experiments optionally take a list of Callback which have several methods that you can implement to see what's going on during training such as on_batch_collected, on_train_end, and on_evaluation_end.

License

BenchMARL is licensed under the MIT License. See LICENSE for details.

TakieddineSOUALHI/BenchMARL