[Algorithm] Td3 (#684)

Author: BY571

examples/td3/config.yaml

@@ -0,0 +1,58 @@
+# Environment
+env_name: HalfCheetah-v4
+env_task: ""
+exp_name: "debugging"
+env_library: gym
+record_video: 0
+normalize_rewards_online: 0
+normalize_rewards_online_scale: 5
+normalize_rewards_online_decay: 0.99
+total_frames: 1000000
+frames_per_batch: 1000
+max_frames_per_traj: 1000
+frame_skip: 1
+from_pixels: 0
+seed: 0
+
+# Collection
+init_random_frames: 25000
+init_env_steps: 10000
+record_interval: 10
+record_frames: 10000
+async_collection: 1
+#collector_devices: [cuda:1,cuda:1,cuda:1,cuda:1]
+collector_devices: [cpu] # ,cpu,cpu,cpu]
+env_per_collector: 1
+num_workers: 1
+
+# Replay Buffer
+buffer_size: 1000000
+
+# Optimization
+utd_ratio: 1.0
+gamma: 0.99
+loss: double
+loss_function: smooth_l1
+lr: 3e-4
+weight_decay: 0.0
+lr_scheduler: ""
+optim_steps_per_batch: 128
+batch_size: 256
+target_update_polyak: 0.995
+
+# Algorithm
+prb: 0 # use prioritized experience replay
+policy_update_delay: 2
+multi_step: 0
+n_steps_return: 1
+activation: relu
+gSDE: 0
+
+# Logging
+logger: wandb
+
+# Extra
+batch_transform: 1
+buffer_prefetch: 64
+norm_stats: 1
+device: "cpu"