skill_chaining.yaml

defaults:
  - hier_agent@agent: hier_agent
  - sc_agent: sc_sac_sil
  - sl_agent: sl_dex
  - _self_

# File path
cwd: ${hydra:runtime.output_dir}

# Training params
task: BiPegTransfer-v0
init_subtask: grasp
subtask: ${init_subtask}
postfix: null
skill_chaining: True
dont_save: False
checkpoint_dir: ./exp/skill_learning/${task}/${sl_agent.name}/d${num_demo}/s${model_seed}

num_demo: 200
seed: 1
model_seed: 1
device: cuda:0
update_epoch: 20
replay_buffer_capacity: 100_000
batch_size: 128

n_train_steps: 1_000_001
n_eval: 1600
n_save: 800
n_log: 9600
n_eval_episodes: 10
eval_frequency: 2_000
n_seed_steps: 200
use_demo_buffer: True
ckpt_episode: latest

use_wb: True
project_name: viskill
entity_name: thuang22

mpi: {rank: null, is_chef: null, num_workers: null}
# Working space
hydra:
  run:
    dir: ./exp/viskill/${task}/${sc_agent.name}_${sl_agent.name}/d${num_demo}/rs${sc_agent.reward_scale}/s${seed}
  sweep:
    dir: ./exp/viskill/${task}/${sc_agent.name}_${sl_agent.name}/d${num_demo}/rs${sc_agent.reward_scale}
    subdir: s${seed}
  sweeper:
    params:
      num_demo: 200
      seed: 1,2,3,4,5