jacobastern / ppo

A simple implementation of the ppo algorithm

ppo

Usage

The easiest way to run this code is in a docker container. The only requirements are Docker and nvidia-docker.

Option 1 - Basic Usage

Clone the repo:

git clone https://github.com/jastern33/ppo.git
cd ppo

Build the docker image (takes awhile):

cd docker
docker build -t deep-learning-pytorch -f Dockerfile . --rm

Run a docker container based on that image:

cd ..
nvidia-docker run -it --name ppo --rm -v $(pwd):/code deep-learning-pytorch

Then run the following command in the docker container for a basic demo:

python3 ppo

To see additional options run:

python3 ppo --help

(hint: look here ^ to see how to try a new dataset!)

Option 2 - To run the Jupyter Notebook demo:

If you want to run the jupyter notebook, it requires a couple of modifications.

Clone the repo:

git clone https://github.com/jastern33/ppo
cd ppo

Build the docker image (takes awhile):

cd docker
docker build -t deep-learning-pytorch -f Dockerfile . --rm

Run a docker container based on that image (notice the addition of the -p flag for port forwarding):

cd ..
nvidia-docker run -it --name ppo --rm -p 8890:8890 -v $(pwd):/code deep-learning-pytorch

In the docker container, run:

jupyter notebook --ip 0.0.0.0 --port 8890 --no-browser --allow-root

Then copy the url that it gives you, and paste it into a browser:

http://127.0.0.1:8890/?token=<really_long_token>

Now you can access and run the Jupyter Notebook (which is running inside the container) as if it were running on your own machine.