Brief (~150 lines of code), quick and dirty SAC implementation using Tensorflow 2.x
. The entropy coeficient alpha = 0.2
is fixed and not learned. A solution for the Pendulum-v0
should be found after about 30 episodes / 6000 steps.
gym
numpy
tensorflow (2.x)
tensorflow_probability
The episode returns are also logged into a log
directory inside the working directory for usage with tensorboard.
tensorboard --logdir logs