Skip to content

Latest commit

 

History

History
95 lines (66 loc) · 3.02 KB

README.md

File metadata and controls

95 lines (66 loc) · 3.02 KB
Raw

gym_dockauv

Master Thesis Project Code, Cooperation of TU Berlin and NTNU

This package includes a custom OpenAI Gym environment with a very simple renderer with the Matplotlib library. In includes four custom environments, supports two underwater vehicles so far and is tested with PPO and SAC. For first run simply execute main.py. The configuration files are in gym_dockauv/config.

View the Sphinx generated doc at: https://erikx3.github.io/gym_dockauv/

Laser_demo.mp4

Install:

For installing all the required packages, use the following commands.

pip install -U gym
pip install stable-baselines3
pip install tensorflow
pip install -U scikit-image
pip install matplotlib

This should also install all the necessary other packages needed, otherwise compare with the packages at the bottom of the requirements.txt.

Additional: For Sphinx documentation generation when working locally

  1. Installation of Sphinx 4.4.0 over https://www.sphinx-doc.org/en/master/usage/installation.html
    1. Add tool for typehinting: https://github.com/tox-dev/sphinx-autodoc-typehints
    2. Add theme with pip install sphinx-rtd-theme

Sphinx usage:

Update rst and html files in separate folder. If no module is added, first command can be skipped.

sphinx-apidoc -f -o docsrc/source gym_dockauv EXCLUDE_PATTERN /*tests*
cd docsrc/
make clean
make html

If you want to update the html files and make them available on Github use:

make github

And push changes to repository.

Unittests

I provided some unittests, since I wanted to make sure in a structured manner, that my modules are working as expected. It is not a 100% test coverage, however, it tests the basic functionality of all packages, so after a change the test should be ran.

This is also good possibility to test, if all the packages and requirements are met in your environment. Here is the command to execute all tests. Make sure to be at the root of this repository.

python -m unittest discover -v -s gym_dockauv/tests
# To run a individual test do e.g.:
python -m unittest -v gym_dockauv.tests.test_integration

Other install

Chosen matplotlib backend on linux:

sudo apt-get install python3-tk

Tensorboard

For tensorboard analysis of the saved tensorboard logs, on linux run (https://pythonprogramming.net/saving-and-loading-reinforcement-learning-stable-baselines-3-tutorial/?completed=/introduction-reinforcement-learning-stable-baselines-3-tutorial/):

tensorboard --logdir /path/to/log/directory

Profiling

The computation time in this thesis has been optimized with Tuna, that is why for completeness it is mentioned here on how to do so. Other possibilities: https://stackoverflow.com/questions/582336/how-do-i-profile-a-python-script

pip install tuna

python3 -m cProfile -o program.prof gym_dockauv\train.py

# On linux
tuna program.prof

Cheers, Erik :)