configs.yaml

defaults:

  logdir: logs
  id: default
  traindir: ''
  evaldir: ''
  offline_traindir: ''
  offline_evaldir: ''
  offline_data_path: ''
  save_eps: True
  seed: 0
  steps: 5e5
  eval_every: 1e4
  log_every: 1e4
  reset_every: 0
  gpu_growth: True
  precision: 16
  debug: False
  expl_gifs: False
  use_wandb: True
  wandb_project: 'dreamer'
  only_eval: False
  only_eval_length: 500
  n_eval_eps: 1
  mode: 'train'
  probe_features_type: 'post'

  # Environment
  task: 'dmc_walker_walk'
  size: [64, 64]
  envs: 1
  action_repeat: 2
  time_limit: 1000
  grayscale: False
  prefill: 2500
  eval_noise: 0.0
  clip_rewards: 'identity'

  # Model
  model_type: 'rssm'
  dyn_cell: 'gru'
  dyn_hidden: 200
  dyn_deter: 200
  dyn_stoch: 32
  dyn_discrete: 0
  dyn_input_layers: 1
  dyn_output_layers: 1
  dyn_rec_depth: 1
  dyn_shared: False
  dyn_mean_act: 'none'
  dyn_std_act: 'sigmoid2'
  dyn_min_std: 0.1
  dyn_temp_post: True
  dyn_kl_mask: False
  dyn_gate_prior: 0.4
  dyn_gate_free: 0.1
  dyn_gate_scale: 0.0
  grad_heads: ['image', 'reward']
  units: 400
  reward_layers: 2
  discount_layers: 3
  value_layers: 3
  actor_layers: 4
  act: 'elu'
  cnn_depth: 32
  encoder_kernels: [4, 4, 4, 4]
  decoder_kernels: [5, 5, 6, 6]
  decoder_thin: True
  value_head: 'normal'
  kl_scale: '1.0'
  kl_balance: '0.8'
  kl_free: '1.0'
  kl_forward: False
  kl_post: False
  pred_discount: False
  discount_scale: 1.0
  reward_scale: 1.0
  weight_decay: 0.0

  # Training
  batch_size: 50
  batch_length: 50
  train_every: 5
  train_steps: 1
  pretrain: 100
  model_lr: 3e-4
  value_lr: 8e-5
  actor_lr: 8e-5
  opt_eps: 1e-5
  grad_clip: 100
  value_grad_clip: 100
  actor_grad_clip: 100
  dataset_size: 0
  oversample_ends: False
  slow_value_target: True
  slow_actor_target: True
  slow_target_update: 100
  slow_target_fraction: 1
  opt: 'adam'

  # Behavior.
  discount: 0.99
  discount_lambda: 0.95
  imag_horizon: 15
  imag_gradient: 'dynamics'
  imag_gradient_mix: '0.1'
  imag_sample: True
  actor_dist: 'trunc_normal'
  actor_entropy: '1e-4'
  actor_state_entropy: 0.0
  actor_init_std: 1.0
  actor_min_std: 0.1
  actor_disc: 5
  actor_temp: 0.1
  actor_outscale: 0.0
  expl_amount: 0.0
  eval_state_mean: False
  collect_dyn_sample: True
  behavior_stop_grad: True
  value_decay: 0.0
  future_entropy: False

  # Exploration
  expl_behavior: 'greedy'
  expl_until: 0
  expl_extr_scale: 0.0
  expl_intr_scale: 1.0
  disag_target: 'stoch'
  disag_log: True
  disag_models: 10
  disag_offset: 1
  disag_layers: 4
  disag_units: 400
  disag_action_cond: False

bringbackshapes:
  steps: 1e6
  eval_every: 5e4
  log_every: 5e4
  wandb_project: 'bringbackshapes'
  n_eval_eps: 10

  # Environment
  task: 'bringbackshapes2D_sparse'
  action_repeat: 4
  time_limit: 3000
  prefill: 10000
  max_distractors: 2
  max_objects: 5
  variable_num_objects: False
  variable_num_distractors: False
  variable_goal_position: False
  agent_view_size: 125
  arena_scale: 1.0

  # Model
  dyn_cell: 'gru_layer_norm'
  dyn_hidden: 1024
  dyn_deter: 1024
  dyn_stoch: 32
  dyn_discrete: 32
  grad_heads: ['image', 'reward', 'discount']
  reward_layers: 4
  discount_layers: 4
  value_layers: 4
  actor_layers: 4
  cnn_depth: 48
  kl_free: '0.0'
  pred_discount: True
  weight_decay: 1e-6

  # Training
  batch_size: 16
  batch_length: 50
  pretrain: 1
  model_lr: 1e-4
  value_lr: 2e-4
  actor_lr: 8e-5
  oversample_ends: True

  # Behavior
  actor_entropy: '2e-3'

bringbackshapes_gaussian:
  dyn_discrete: 0
  kl_free: 1.0

bringbackshapes_svsg:
  dyn_cell: 'sgru'
  dyn_stoch: 1024
  dyn_discrete: 0
  dyn_kl_mask: True
  model_type: 'srssm'
  kl_free: 1.0

bringbackshapes3D:
  task: 'bringbackshapes3D_sparse'

dmc:
  # Environment
  task: 'dmc_walker_walk'
  time_limit: 1000
  action_repeat: 2

  # Model
  dyn_cell: 'gru_layer_norm'
  cnn_depth: 32
  dyn_deter: 200
  dyn_stoch: 32
  dyn_discrete: 0
  reward_layers: 2
  value_layers: 3
  actor_layers: 4
  reward_scale: 2

  # Training
  dataset_size: 0
  pretrain: 100
  prefill: 2500

  # Behavior
  actor_dist: 'trunc_normal'
  actor_entropy: '1e-4'
  imag_gradient_mix: 1.0 # Do we want to keep it 1.0?

  # Training
  model_lr: 3e-4
  value_lr: 8e-5
  actor_lr: 8e-5

debug:

  debug: True
  pretrain: 1
  prefill: 1
  train_steps: 1
  batch_size: 10
  batch_length: 20