steffen/marl-factory-grid
1
0
Fork 0
mirror of https://github.com/illiumst/marl-factory-grid.git synced 2025-07-05 09:01:36 +02:00
Code Issues Packages Projects Releases Wiki Activity

Add various RL adapted configs

Browse Source
This commit is contained in:
Julian Schönberger
2024-05-02 11:00:35 +02:00
parent 48d708bbcd
commit c7c2c4e5a3
9 changed files with 334 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
marl_factory_grid/algorithms/marl/configs/dirt_quadrant_config.yaml Normal file
View File

@ -0,0 +1,28 @@
agent:
classname: marl_factory_grid.algorithms.marl.networks.RecurrentAC
n_agents: 1
obs_emb_size: 96
action_emb_size: 16
hidden_size_actor: 64
hidden_size_critic: 64
use_agent_embedding: False
env:
classname: marl_factory_grid.configs.custom
env_name: "custom/dirt_quadrant_random_pos"
n_agents: 1
max_steps: 250
pomdp_r: 2
stack_n_frames: 0
individual_rewards: True
train_render: False
eval_render: True
method: marl_factory_grid.algorithms.marl.LoopSEAC
algorithm:
gamma: 0.99
entropy_coef: 0.01
vf_coef: 0.05
n_steps: 0 # How much experience should be sampled at most (n-TD) until the next value and policy update is performed. Default 0: MC
max_steps: 80000
advantage: "Advantage-AC" # Options: "Advantage-AC", "TD-Advantage-AC", "Reinforce"
pile-order: "fixed" # Options: "fixed", "random", "none", "agents"

3
marl_factory_grid/algorithms/marl/configs/environment_changes Normal file
View File

@ -0,0 +1,3 @@
marl_factory_grid>environment>rules.py#SpawnEntity.on_reset()
marl_factory_grid>environment>rewards.py
marl_factory_grid>modules>clean_up>groups.py#DirtPiles.trigger_spawn()

14
marl_factory_grid/algorithms/marl/example_config.yaml → marl_factory_grid/algorithms/marl/configs/two_rooms_one_door_modified_config.yaml
View File

@ -7,20 +7,22 @@ agent:
hidden_size_critic: 64
use_agent_embedding: False
env:
classname: marl_factory_grid.configs
env_name: "simple_crossing"
classname: marl_factory_grid.configs.custom
env_name: "custom/two_rooms_one_door_modified_random_pos"
n_agents: 2
max_steps: 250
pomdp_r: 2
stack_n_frames: 0
individual_rewards: True
train_render: True
train_render: False
eval_render: True
method: marl_factory_grid.algorithms.marl.LoopSEAC
algorithm:
gamma: 0.99
entropy_coef: 0.01
vf_coef: 0.5
n_steps: 5
max_steps: 10000
vf_coef: 0.05
n_steps: 0 # How much experience should be sampled at most (n-TD) until the next value and policy update is performed. Default 0: MC
max_steps: 100000
advantage: "TD-Advantage-AC" # Options: "Advantage-AC", "TD-Advantage-AC", "Reinforce"
pile-order: "agents" # Options: "fixed", "random", "none", "agents"