Simple modular configuration refactoring

README.md

@@ -72,7 +72,7 @@ conda activate drqv2
 
 Train the agent:
 ```sh
-python train.py task=quadruped_walk
+python train.py env_cfg=quadruped_walk
 ```
 
 Monitor results:

cfgs/agent_cfg/drqv2.yaml

@@ -0,0 +1,16 @@
+experiment: exp
+
+agent:
+  _target_: drqv2.DrQV2Agent
+  obs_shape: ??? # to be specified later
+  action_shape: ??? # to be specified later
+  device: ${device}
+  lr: ${env_cfg.lr}
+  critic_target_tau: 0.01
+  update_every_steps: 2
+  use_tb: ${use_tb}
+  num_expl_steps: 2000
+  hidden_dim: 1024
+  feature_dim: ${env_cfg.feature_dim}
+  stddev_schedule: ${env_cfg.stddev_schedule}
+  stddev_clip: 0.3

cfgs/config.yaml

@@ -0,0 +1,29 @@
+defaults:
+  - agent_cfg: drqv2
+  - env_cfg: quadruped_walk
+  - override hydra/launcher: submitit_local
+
+# snapshot
+save_snapshot: false
+#misc
+seed: 1
+device: cuda
+save_video: true
+save_train_video: false
+use_tb: false
+
+
+hydra:
+  run:
+    dir: ./exp_local/${now:%Y.%m.%d}/${now:%H%M%S}_${hydra.job.override_dirname}
+  sweep:
+    dir: ./exp/${now:%Y.%m.%d}/${now:%H%M}_${agent_cfg.experiment}
+    subdir: ${hydra.job.num}
+  launcher:
+    timeout_min: 4300
+    cpus_per_task: 10
+    gpus_per_node: 1
+    tasks_per_node: 1
+    mem_gb: 160
+    nodes: 1
+    submitit_folder: ./exp/${now:%Y.%m.%d}/${now:%H%M%S}_${agent_cfg.experiment}/.slurm

cfgs/env_cfg/acrobot_swingup.yaml

@@ -0,0 +1,20 @@
+# task settings
+task: acrobot_swingup
+frame_stack: 3
+action_repeat: 2
+discount: 0.99
+# train settings
+num_train_frames: 3100000
+num_seed_frames: 4000
+# eval
+eval_every_frames: 10000
+num_eval_episodes: 10
+# replay buffer
+replay_buffer_size: 1000000
+replay_buffer_num_workers: 4
+nstep: 3
+batch_size: 256
+# agent
+lr: 1e-4
+feature_dim: 50
+stddev_schedule: 'linear(1.0,0.1,500000)'

cfgs/env_cfg/cartpole_swingup_sparse.yaml

@@ -0,0 +1,20 @@
+# task settings
+task: cartpole_swingup_sparse
+frame_stack: 3
+action_repeat: 2
+discount: 0.99
+# train settings
+num_train_frames: 3100000
+num_seed_frames: 4000
+# eval
+eval_every_frames: 10000
+num_eval_episodes: 10
+# replay buffer
+replay_buffer_size: 1000000
+replay_buffer_num_workers: 4
+nstep: 3
+batch_size: 256
+# agent
+lr: 1e-4
+feature_dim: 50
+stddev_schedule: 'linear(1.0,0.1,500000)'

cfgs/env_cfg/cheetah_run.yaml

@@ -0,0 +1,20 @@
+# task settings
+task: cheetah_run
+frame_stack: 3
+action_repeat: 2
+discount: 0.99
+# train settings
+num_train_frames: 3100000
+num_seed_frames: 4000
+# eval
+eval_every_frames: 10000
+num_eval_episodes: 10
+# replay buffer
+replay_buffer_size: 1000000
+replay_buffer_num_workers: 4
+nstep: 3
+batch_size: 256
+# agent
+lr: 1e-4
+feature_dim: 50
+stddev_schedule: 'linear(1.0,0.1,500000)'

cfgs/env_cfg/finger_turn_easy.yaml

@@ -0,0 +1,20 @@
+# task settings
+task: finger_turn_easy
+frame_stack: 3
+action_repeat: 2
+discount: 0.99
+# train settings
+num_train_frames: 3100000
+num_seed_frames: 4000
+# eval
+eval_every_frames: 10000
+num_eval_episodes: 10
+# replay buffer
+replay_buffer_size: 1000000
+replay_buffer_num_workers: 4
+nstep: 3
+batch_size: 256
+# agent
+lr: 1e-4
+feature_dim: 50
+stddev_schedule: 'linear(1.0,0.1,500000)'

cfgs/env_cfg/finger_turn_hard.yaml

@@ -0,0 +1,20 @@
+# task settings
+task: finger_turn_hard
+frame_stack: 3
+action_repeat: 2
+discount: 0.99
+# train settings
+num_train_frames: 3100000
+num_seed_frames: 4000
+# eval
+eval_every_frames: 10000
+num_eval_episodes: 10
+# replay buffer
+replay_buffer_size: 1000000
+replay_buffer_num_workers: 4
+nstep: 3
+batch_size: 256
+# agent
+lr: 1e-4
+feature_dim: 50
+stddev_schedule: 'linear(1.0,0.1,500000)'

cfgs/env_cfg/hopper_hop.yaml

@@ -0,0 +1,20 @@
+# task settings
+task: hopper_hop
+frame_stack: 3
+action_repeat: 2
+discount: 0.99
+# train settings
+num_train_frames: 3100000
+num_seed_frames: 4000
+# eval
+eval_every_frames: 10000
+num_eval_episodes: 10
+# replay buffer
+replay_buffer_size: 1000000
+replay_buffer_num_workers: 4
+nstep: 3
+batch_size: 256
+# agent
+lr: 1e-4
+feature_dim: 50
+stddev_schedule: 'linear(1.0,0.1,500000)'

cfgs/env_cfg/quadruped_run.yaml

@@ -0,0 +1,20 @@
+# task settings
+task: quadruped_run
+frame_stack: 3
+action_repeat: 2
+discount: 0.99
+# train settings
+num_train_frames: 3100000
+num_seed_frames: 4000
+# eval
+eval_every_frames: 10000
+num_eval_episodes: 10
+# replay buffer
+replay_buffer_size: 100000
+replay_buffer_num_workers: 4
+nstep: 3
+batch_size: 256
+# agent
+lr: 1e-4
+feature_dim: 50
+stddev_schedule: 'linear(1.0,0.1,500000)'

cfgs/env_cfg/quadruped_walk.yaml

@@ -0,0 +1,20 @@
+# task settings
+task: quadruped_walk
+frame_stack: 3
+action_repeat: 2
+discount: 0.99
+# train settings
+num_train_frames: 3100000
+num_seed_frames: 4000
+# eval
+eval_every_frames: 10000
+num_eval_episodes: 10
+# replay buffer
+replay_buffer_size: 1000000
+replay_buffer_num_workers: 4
+nstep: 3
+batch_size: 256
+# agent
+lr: 1e-4
+feature_dim: 50
+stddev_schedule: 'linear(1.0,0.1,500000)'

cfgs/env_cfg/reach_duplo.yaml

@@ -0,0 +1,20 @@
+# task settings
+task: reach_duplo
+frame_stack: 3
+action_repeat: 2
+discount: 0.99
+# train settings
+num_train_frames: 3100000
+num_seed_frames: 4000
+# eval
+eval_every_frames: 10000
+num_eval_episodes: 10
+# replay buffer
+replay_buffer_size: 1000000
+replay_buffer_num_workers: 4
+nstep: 3
+batch_size: 256
+# agent
+lr: 1e-4
+feature_dim: 50
+stddev_schedule: 'linear(1.0,0.1,500000)'

cfgs/env_cfg/reacher_easy.yaml

@@ -0,0 +1,20 @@
+# task settings
+task: reacher_easy
+frame_stack: 3
+action_repeat: 2
+discount: 0.99
+# train settings
+num_train_frames: 3100000
+num_seed_frames: 4000
+# eval
+eval_every_frames: 10000
+num_eval_episodes: 10
+# replay buffer
+replay_buffer_size: 1000000
+replay_buffer_num_workers: 4
+nstep: 3
+batch_size: 256
+# agent
+lr: 1e-4
+feature_dim: 50
+stddev_schedule: 'linear(1.0,0.1,500000)'

cfgs/env_cfg/reacher_hard.yaml

@@ -0,0 +1,20 @@
+# task settings
+task: reacher_hard
+frame_stack: 3
+action_repeat: 2
+discount: 0.99
+# train settings
+num_train_frames: 3100000
+num_seed_frames: 4000
+# eval
+eval_every_frames: 10000
+num_eval_episodes: 10
+# replay buffer
+replay_buffer_size: 1000000
+replay_buffer_num_workers: 4
+nstep: 3
+batch_size: 256
+# agent
+lr: 1e-4
+feature_dim: 50
+stddev_schedule: 'linear(1.0,0.1,500000)'

cfgs/env_cfg/walker_run.yaml

@@ -0,0 +1,20 @@
+# task settings
+task: walker_run
+frame_stack: 3
+action_repeat: 2
+discount: 0.99
+# train settings
+num_train_frames: 3100000
+num_seed_frames: 4000
+# eval
+eval_every_frames: 10000
+num_eval_episodes: 10
+# replay buffer
+replay_buffer_size: 1000000
+replay_buffer_num_workers: 4
+nstep: 1
+batch_size: 512
+# agent
+lr: 1e-4
+feature_dim: 50
+stddev_schedule: 'linear(1.0,0.1,500000)'