[rllib] MAML Agent (#8862)

* Halfway done with transferring MAML to new Ray * MAML Beta Out * Debugging MAML atm * Distributed Execution * Pendulum Mass Working * All experiments complete * Cleaned up codebase * Travis CI * Travis CI * Tests * Merged conflicts * Fixed variance bug conflict * Comment resolved * Apply suggestions from code review fixed test_maml * Update rllib/agents/maml/tests/test_maml.py * asdf * Fix testing Co-authored-by: Sven Mika <sven@anyscale.io>
2026-09-09 11:32:43 +08:00 · 2020-06-23 09:48:23 -07:00
parent b449ece2ea
commit cf0894d396
13 changed files with 974 additions and 0 deletions
@@ -0,0 +1,26 @@
+ant-rand-goal-maml:
+    env: ray.rllib.examples.env.ant_rand_goal.AntRandGoalEnv
+    run: MAML
+    stop:
+        training_iteration: 1000
+    config:
+        horizon: 200
+        rollout_fragment_length: 200
+        num_envs_per_worker: 20
+        inner_adaptation_steps: 2
+        maml_optimizer_steps: 5
+        gamma: 0.99
+        lambda: 1.0
+        lr: 0.001
+        vf_loss_coeff: 0.5
+        clip_param: 0.3
+        kl_target: 0.01
+        kl_coeff: 0.0005
+        num_workers: 32
+        num_gpus: 1
+        inner_lr: 0.03
+        explore: True
+        clip_actions: False
+        model:
+            fcnet_hiddens: [64, 64]
+            free_log_std: True
@@ -0,0 +1,25 @@
+halfcheetah-rand-direc-maml:
+    env: ray.rllib.examples.env.halfcheetah_rand_direc.HalfCheetahRandDirecEnv
+    run: MAML
+    stop:
+        training_iteration: 1000    
+    config:
+        horizon: 100
+        rollout_fragment_length: 100
+        num_envs_per_worker: 20
+        inner_adaptation_steps: 1
+        maml_optimizer_steps: 5
+        gamma: 0.99
+        lambda: 1.0
+        lr: 0.001
+        vf_loss_coeff: 0.5
+        clip_param: 0.3
+        kl_target: 0.01
+        kl_coeff: 0.0005
+        num_workers: 31
+        num_gpus: 1
+        inner_lr: 0.1
+        clip_actions: False
+        model:
+            fcnet_hiddens: [64, 64]
+            free_log_std: True
@@ -0,0 +1,26 @@
+pendulum-mass-maml:
+    env: ray.rllib.examples.env.pendulum_mass.PendulumMassEnv
+    run: MAML
+    stop:
+        training_iteration: 500   
+    config:
+        horizon: 200
+        rollout_fragment_length: 200
+        num_envs_per_worker: 10
+        inner_adaptation_steps: 1
+        maml_optimizer_steps: 5
+        gamma: 0.99
+        lambda: 1.0
+        lr: 0.001
+        vf_loss_coeff: 0.5
+        clip_param: 0.3
+        kl_target: 0.01
+        kl_coeff: 0.001
+        num_workers: 20
+        num_gpus: 1
+        inner_lr: 0.03
+        explore: True
+        clip_actions: False
+        model:
+            fcnet_hiddens: [64, 64]
+            free_log_std: True