Implementation of Reinforcement Learning algorithms in Python3 using he OpenAI framework and TensorFlow as neural network libraries. Results are shown with TensorBoard.
#Gym-A3C Implementation of the A3C algorithm for OpenAI Gym's Atari games. The implementation uses processes instead of threads to achieve real concurrency.
##How to run the A3C ? To launch the A3C with the default parameters, just use the following command. It is possible to see the available hyper parameters with the command -h.
python3.5 main.py Pong-v0
To see different plots like the rewards, the losses... it is necessary to launch TensorBoard with the command
tensorboard --logdir=tf_logs/
##Results
Pong-v0, 4 actors, 5 local steps and updating the network with an Adam Optimizer (learning rate: 1e-4).
#Useful papers A3C