# you need to install Anaconda first
conda create -n joyrl python=3.7
conda activate joyrl
pip install -U joyrl
Torch:
# CPU
conda install pytorch==1.10.0 torchvision==0.11.0 torchaudio==0.10.0 cpuonly -c pytorch
# GPU
conda install pytorch==1.10.0 torchvision==0.11.0 torchaudio==0.10.0 cudatoolkit=11.3 -c pytorch -c conda-forge
# GPU with mirrors
pip install torch==1.10.0+cu113 torchvision==0.11.0+cu113 torchaudio==0.10.0 --extra-index-url https://download.pytorch.org/whl/cu113
the following presents a demo to use joyrl, you donot need to care about complicated details of code. All your need is just to set hyper parameters including GeneralConfig()
and AlgoConfig()
, which is also shown in examples folder, and well trained results are shown in the benchmarks folder as well.
import joyrl
class GeneralConfig():
def __init__(self) -> None:
self.env_name = "CartPole-v1" # name of environment
self.algo_name = "DQN" # name of algorithm
self.mode = "train" # train or test
self.seed = 0 # random seed
self.device = "cpu" # device to use
self.train_eps = 100 # number of episodes for training
self.test_eps = 20 # number of episodes for testing
self.eval_eps = 10 # number of episodes for evaluation
self.eval_per_episode = 5 # evaluation per episode
self.max_steps = 200 # max steps for each episode
self.load_checkpoint = False
self.load_path = "tasks" # path to load model
self.show_fig = False # show figure or not
self.save_fig = True # save figure or not
class AlgoConfig():
def __init__(self) -> None:
# set epsilon_start=epsilon_end can obtain fixed epsilon=epsilon_end
self.epsilon_start = 0.95 # epsilon start value
self.epsilon_end = 0.01 # epsilon end value
self.epsilon_decay = 500 # epsilon decay rate
self.gamma = 0.95 # discount factor
self.lr = 0.0001 # learning rate
self.buffer_size = 100000 # size of replay buffer
self.batch_size = 64 # batch size
self.target_update = 4 # target network update frequency
self.value_layers = [
{'layer_type': 'linear', 'layer_dim': ['n_states', 256],
'activation': 'relu'},
{'layer_type': 'linear', 'layer_dim': [256, 256],
'activation': 'relu'},
{'layer_type': 'linear', 'layer_dim': [256, 'n_actions'],
'activation': 'none'}]
if __name__ == "__main__":
general_cfg = GeneralConfig()
algo_cfg = AlgoConfig()
joyrl.run(general_cfg,algo_cfg)
More tutorials and API documentation are hosted on https://datawhalechina.github.io/joyrl/
Name | Reference | Author | Notes |
---|---|---|---|
DQN | DQN Paper | johnjim0816 |