tensorflow-policy-gradient

Still under construction...

Dependencies

• Python 2.7
• TensorFlow >= 0.8.0
• NumPy >= 1.10.0
• openai gym
• matplotlib

Quick try

Run

python gym_experiment.py

to train a softmax policy (without bias) using vanilla policy gradient on CartPole task. You can see that the return is stochastically increasing until it reaches the maximum (200).