[Meltingpot] Fine tuning PPO

benchmarl/environments/meltingpot/common.py

@@ -81,6 +81,7 @@ def get_env_fun(
         return lambda: MeltingpotEnv(
             substrate=self.name.lower(),
             categorical_actions=True,
+            device=device,
             **self.config,
         )
 

fine_tuned/meltingpot/conf/config.yaml

@@ -0,0 +1,72 @@
+defaults:
+  - experiment: base_experiment
+  - algorithm: ippo
+  - task: meltingpot/commons_harvest__open
+  - model: layers/cnn
+  - model@critic_model: layers/cnn
+  - _self_
+
+hydra:
+  searchpath:
+   # Tells hydra to add the default benchmarl configuration to its path
+    - pkg://benchmarl/conf
+
+seed: 0
+
+task:
+  max_steps: 200
+
+model:
+  mlp_num_cells: [ 256, 256 ]
+
+  cnn_num_cells: [ 16, 32, 256 ]
+  cnn_kernel_sizes: [ 8, 4, 11 ]
+  cnn_strides: [4, 2, 1]
+  cnn_paddings: [2, 1, 5]
+  cnn_activation_class: torch.nn.ReLU
+
+critic_model:
+  mlp_num_cells: [ 256, 256 ]
+
+  cnn_num_cells: [ 16, 32, 256 ]
+  cnn_kernel_sizes: [ 8, 4, 11 ]
+  cnn_strides: [ 4, 2, 1 ]
+  cnn_paddings: [ 2, 1, 5 ]
+  cnn_activation_class: torch.nn.ReLU
+
+algorithm:
+  entropy_coef: 0.001
+  use_tanh_normal: True
+
+experiment:
+  sampling_device: "cpu"
+  train_device: "cuda"
+
+  share_policy_params: True
+  gamma: 0.99
+
+  adam_eps: 0.000001
+  lr: 0.00025
+  clip_grad_norm: True
+  clip_grad_val: 5
+
+  max_n_iters: null
+  max_n_frames: 10_000_000
+
+  on_policy_collected_frames_per_batch: 1000
+  on_policy_n_envs_per_worker: 1
+  on_policy_n_minibatch_iters: 45
+  on_policy_minibatch_size: 100
+
+  evaluation: True
+  render: True
+  evaluation_interval: 1000
+  evaluation_episodes: 1
+  evaluation_deterministic_actions: False
+
+  loggers: [wandb]
+  create_json: False
+
+  save_folder: null
+  restore_file: null
+  checkpoint_interval: 0

fine_tuned/meltingpot/meltingpot_run.py

@@ -0,0 +1,31 @@
+#  Copyright (c) Meta Platforms, Inc. and affiliates.
+#
+#  This source code is licensed under the license found in the
+#  LICENSE file in the root directory of this source tree.
+#
+
+import hydra
+
+from benchmarl.experiment import Experiment
+
+from benchmarl.hydra_config import load_experiment_from_hydra
+from hydra.core.hydra_config import HydraConfig
+from omegaconf import DictConfig, OmegaConf
+
+
+@hydra.main(version_base=None, config_path="conf", config_name="config")
+def hydra_experiment(cfg: DictConfig) -> None:
+    hydra_choices = HydraConfig.get().runtime.choices
+    task_name = hydra_choices.task
+    algorithm_name = hydra_choices.algorithm
+
+    print(f"\nAlgorithm: {algorithm_name}, Task: {task_name}")
+    print("\nLoaded config:\n")
+    print(OmegaConf.to_yaml(cfg))
+
+    experiment: Experiment = load_experiment_from_hydra(cfg, task_name=task_name)
+    experiment.run()
+
+
+if __name__ == "__main__":
+    hydra_experiment()