Machine Learning project for personal experimentation and learning.
Based on the Stable Baselines Lunar Lander example: https://stable-baselines.readthedocs.io/en/master/guide/examples.html
The relevant files are mainly train_climber.py, run_climber, and custom_env/lunar_climber_env.py. The rest can be ignored relatively safely, including folders.
Dockerfile.gpu is my attempt at creating a docker image that enables the use of GPU. I'm not sure if this actually works, as I can't test on my machine (no GPU) and I don't know what drivers exist on the machines running Sagemaker Notebook. I'm currently attempting to test this.
I'm also not sure whether Tensorflow actually makes use of the GPU, given that I configure the Docker image correctly. According to Tensorflow's GPU guide, any version after tensorflow==1.15 combines CPU and GPU compilations into the same package. I am using exactly 1.15.0, however, so I suppose I need pip install tensorflow-gpu. I haven't been able to make this work yet.
This Github issue comment seems to indicate that simply installing tensorflow-gpu instead of tensorflow will make the algorithm use GPU instead, but I'm not sure whether that's reliable.
”
Multi-processing is enabled simply by replacing
env = LunarLander()
with
env = SubprocVecEnv([lambda: LunarLander() for i in range(n_envs)])
The PPO2 algorithm I use automatically coordinates the interaction between the different processes. I assume this also works for GPU when I get that working.