Overview¶
A toolbox for worst-case validation of autonomous policies.
Adaptive Stress Testing is a worst-case validation method for autonomous policies. This toolbox is being actively developed by the Stanford Intelligent Systems Lab.
See https://ast-toolbox.readthedocs.io/en/latest/ for documentation.
Maintained by the Stanford Intelligent Systems Lab (SISL)
- Free software: MIT license
Installation¶
Pip Installation Method¶
You can install the latest stable release from pypi:
pip install ast-toolbox
You can also install the latest version with:
pip install git+ssh://git@https://github.com/sisl/AdaptiveStressTestingToolbox.git@master
Using the Go-Explore work requires having a Berkely DB installation findable on your system. If you are on Linux:
sudo apt-get update
sudo apt install libdb-dev python3-bsddb3
If you are on OSX:
brew install berkeley-db
export BERKELEYDB_DIR=$(brew --cellar)/berkeley-db/5.3
export YES_I_HAVE_THE_RIGHT_TO_USE_THIS_BERKELEY_DB_VERSION=1
Once you have the Berkeley DB system dependency met, you can install the toolbox:
pip install ast-toolbox[ge]
Git Installation Method¶
If you are interested in development, you should clone the repo. You can use https:
git clone https://github.com/sisl/AdaptiveStressTestingToolbox.git
You can also use ssh:
git clone git@github.com:sisl/AdaptiveStressTestingToolbox.git
If you are on Linux, use the following commands to setup the Toolbox:
cd AdaptiveStressTestingToolbox
git submodule update --init --recursive
sudo chmod a+x scripts/install_all.sh
sudo scripts/install_all.sh
source scripts/setup.sh
Documentation¶
You can find our documentation on readthedocs.
Development¶
Please see our Contributions Guide.
Acknowledgements¶
Built using the cookiecutter-pylibrary by Ionel Cristian Mărieș