Spaces:
Sleeping
Sleeping
| from easydict import EasyDict | |
| nstep = 1 | |
| lunarlander_trex_dqn_config = dict( | |
| exp_name='lunarlander_trex_dqn_seed0', | |
| env=dict( | |
| # Whether to use shared memory. Only effective if "env_manager_type" is 'subprocess' | |
| # Env number respectively for collector and evaluator. | |
| collector_env_num=8, | |
| evaluator_env_num=8, | |
| env_id='LunarLander-v2', | |
| n_evaluator_episode=8, | |
| stop_value=200, | |
| ), | |
| reward_model=dict( | |
| type='trex', | |
| min_snippet_length=30, | |
| max_snippet_length=100, | |
| checkpoint_min=1000, | |
| checkpoint_max=9000, | |
| checkpoint_step=1000, | |
| num_snippets=60000, | |
| learning_rate=1e-5, | |
| update_per_collect=1, | |
| # Users should add their own model path here. Model path should lead to a model. | |
| # Absolute path is recommended. | |
| # In DI-engine, it is ``exp_name/ckpt/ckpt_best.pth.tar``. | |
| expert_model_path='model_path_placeholder', | |
| # Path where to store the reward model | |
| reward_model_path='data_path_placeholder + /lunarlander.params', | |
| # Users should add their own data path here. Data path should lead to a file to store data or load the stored data. | |
| # Absolute path is recommended. | |
| # In DI-engine, it is usually located in ``exp_name`` directory | |
| # e.g. 'exp_name/expert_data.pkl' | |
| data_path='data_path_placeholder', | |
| ), | |
| policy=dict( | |
| # Whether to use cuda for network. | |
| cuda=False, | |
| model=dict( | |
| obs_shape=8, | |
| action_shape=4, | |
| encoder_hidden_size_list=[512, 64], | |
| # Whether to use dueling head. | |
| dueling=True, | |
| ), | |
| # Reward's future discount factor, aka. gamma. | |
| discount_factor=0.99, | |
| # How many steps in td error. | |
| nstep=nstep, | |
| # learn_mode config | |
| learn=dict( | |
| update_per_collect=10, | |
| batch_size=64, | |
| learning_rate=0.001, | |
| # Frequency of target network update. | |
| target_update_freq=100, | |
| ), | |
| # collect_mode config | |
| collect=dict( | |
| # You can use either "n_sample" or "n_episode" in collector.collect. | |
| # Get "n_sample" samples per collect. | |
| n_sample=64, | |
| # Cut trajectories into pieces with length "unroll_len". | |
| unroll_len=1, | |
| ), | |
| # command_mode config | |
| other=dict( | |
| # Epsilon greedy with decay. | |
| eps=dict( | |
| # Decay type. Support ['exp', 'linear']. | |
| type='exp', | |
| start=0.95, | |
| end=0.1, | |
| decay=50000, | |
| ), | |
| replay_buffer=dict(replay_buffer_size=100000, ) | |
| ), | |
| ), | |
| ) | |
| lunarlander_trex_dqn_config = EasyDict(lunarlander_trex_dqn_config) | |
| main_config = lunarlander_trex_dqn_config | |
| lunarlander_trex_dqn_create_config = dict( | |
| env=dict( | |
| type='lunarlander', | |
| import_names=['dizoo.box2d.lunarlander.envs.lunarlander_env'], | |
| ), | |
| env_manager=dict(type='subprocess'), | |
| policy=dict(type='dqn'), | |
| ) | |
| lunarlander_trex_dqn_create_config = EasyDict(lunarlander_trex_dqn_create_config) | |
| create_config = lunarlander_trex_dqn_create_config | |
| if __name__ == '__main__': | |
| # Users should first run ``lunarlander_dqn_config.py`` to save models (or checkpoints). | |
| # Note: Users should check that the checkpoints generated should include iteration_'checkpoint_min'.pth.tar, iteration_'checkpoint_max'.pth.tar with the interval checkpoint_step | |
| # where checkpoint_max, checkpoint_min, checkpoint_step are specified above. | |
| import argparse | |
| import torch | |
| from ding.entry import trex_collecting_data | |
| from ding.entry import serial_pipeline_trex | |
| parser = argparse.ArgumentParser() | |
| parser.add_argument('--cfg', type=str, default='please enter abs path for this file') | |
| parser.add_argument('--seed', type=int, default=0) | |
| parser.add_argument('--device', type=str, default='cuda' if torch.cuda.is_available() else 'cpu') | |
| args = parser.parse_args() | |
| # The function ``trex_collecting_data`` below is to collect episodic data for training the reward model in trex. | |
| trex_collecting_data(args) | |
| serial_pipeline_trex([main_config, create_config]) | |