firs commit for our new MARL algorithms library, contains working implementations of IAC, SNAC and SEAC

2022-01-21 15:31:07 +01:00
parent 3e19970a60
commit ffc47752a7
24 changed files with 762 additions and 847 deletions
--- a/algorithms/marl/example_config.yaml
+++ b/algorithms/marl/example_config.yaml
@ -0,0 +1,24 @@
+agent:
+  classname:           algorithms.marl.networks.RecurrentAC
+  n_agents:            2
+  obs_emb_size:        96
+  action_emb_size:     16
+  hidden_size_actor:   64
+  hidden_size_critic:  64
+  use_agent_embedding: False
+env:
+  classname:          environments.factory.make
+  env_name:           "DirtyFactory-v0"
+  n_agents:           2
+  max_steps:          250
+  pomdp_r:            2
+  stack_n_frames:     0
+  individual_rewards: True
+method:               algorithms.marl.LoopSEAC
+algorithm:
+  gamma:              0.99
+  entropy_coef:       0.01
+  vf_coef:            0.5
+  n_steps:            5
+  max_steps:          1000000
+