Skip to content

Latest commit

 

History

History
24 lines (16 loc) · 1.42 KB

README.md

File metadata and controls

24 lines (16 loc) · 1.42 KB

DDPG-TensorFlow

This is an tensorflow implementation of the paper "Continuous control with deep reinforcement learning". The code is adapted from here with some improvement. The main difference between the MorvanZhou's version and this code is

  1. Using tf.train.ExponentialMovingAverage to soft update the target network instead of tf.assign;
  2. Changing the activation function tf.nn.relu to tf.nn.elu;
  3. Using tf.train.AdagradOptimizer to train the neural network;
  4. Batch Updating the parameters using accumulating averaged gradients instead of the mean of mini batch.

The speed is 4.0X faster than the MorvanZhou's version when running on the same desktop computer with GTX 960 GPU, and achieving more stable learning curve.

Typing python Train.py in the terminal to run, make sure you've already installed tensorflow and open ai gym.

Any comment or suggestion is welcome, you can open an issue or contact me via "[email protected]".

More deep RL algorithm is coming soon!

Reference

  1. Lillicrap, T. P., Hunt, J. J., Pritzel, A., Heess, N., Erez, T., Tassa, Y., ... & Wierstra, D. (2015). Continuous control with deep reinforcement learning. arXiv preprint arXiv:1509.02971.
  2. https://github.com/MorvanZhou/Reinforcement-learning-with-tensorflow
  3. https://github.com/songrotek/DDPG
  4. https://github.com/RuiShu/micro-projects/tree/master/tf-ema