DDPG implimentaion in Tensorflow-2.0
- Implimenting DDPG Algorithm in Tensorflow-2.0
- Tested on Open-AI Pendulum-v0 and Continous mountain car gym environments.
- DDPG - algorthim : https://arxiv.org/abs/1509.02971
- pip install DDPG-TF
import gym from ddpg import DDPG env = gym.make('Pendulum-v0') ddpg = DDPG( env , # Gym environment with continous action space actor(None), # Tensorflow/keras model critic (None), # Tensorflow/keras model buffer (None), # pre-recorded buffer action_bound_range=1, max_buffer_size =10000, # maximum transitions to be stored in buffer batch_size =64, # batch size for training actor and critic networks max_time_steps = 1000 ,# no of time steps per epoch tow = 0.001, # for soft target update discount_factor = 0.99, explore_time = 1000, # time steps for random actions for exploration actor_learning_rate = 0.0001, critic_learning_rate = 0.001 dtype = 'float32', n_episodes = 1000 ,# no of episodes to run reward_plot = True ,# (bool) to plot reward progress per episode model_save = 1) # epochs to save models and buffer ddpg.train()
- On pendulum problem explored for 5 episodes
- On Continous mountain car problem explored for 100 episodes
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
|Filename, size||File type||Python version||Upload date||Hashes|
|Filename, size DDPG_TF-2.0.3-py3-none-any.whl (5.9 kB)||File type Wheel||Python version py3||Upload date||Hashes View hashes|
|Filename, size DDPG-TF-2.0.3.tar.gz (4.8 kB)||File type Source||Python version None||Upload date||Hashes View hashes|