Skip to content
This repository has been archived by the owner on Jun 8, 2020. It is now read-only.

Implementation of Algorithms from the Policy Gradient Family. Currently includes: A2C, A3C, DDPG, TD3, SAC

Notifications You must be signed in to change notification settings

cyoon1729/Policy-Gradient-Methods

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

54 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Policy-Gradient-Methods

Author: Chris Yoon

Implementations of important policy gradient algorithms in deep reinforcement learning.

Implementations

More implementations will be added soon.

Known Dependencies

  • Python 3.6
  • PyTorch 1.0.2
  • gym 0.12.5

How to run:

Install package

git clone git@github.com:cyoon1729/Policy-Gradient-Methods.git
cd Policy-Gradient-Methods
pip install .

Example:

import gym

from policygradients.common.utils import mini_batch_train  # import training function
from policygradients.td3.td3 import TD3Agent  # import agent from algorithm of interest

# Create Gym environment
env = gym.make("Pendulum-v0")

# check agent class for initialization parameters and initialize agent
gamma = 0.99
tau = 1e-2
noise_std = 0.2
bound = 0.5
delay_step = 2
buffer_maxlen = 100000
critic_lr = 1e-3
actor_lr = 1e-3

agent = TD3Agent(env, gamma, tau, buffer_maxlen, delay_step, noise_std, bound, critic_lr, actor_lr)

# define training parameters
max_episodes = 100
max_steps = 500
batch_size = 32

# train agent with mini_batch_train function
episode_rewards = mini_batch_train(env, agent, max_episodes, max_steps, batch_size)

About

Implementation of Algorithms from the Policy Gradient Family. Currently includes: A2C, A3C, DDPG, TD3, SAC

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published