ast-toolbox 2020.9.1.7

A toolbox for worst-case validation of autonomous policies

Adaptive Stress Testing Toolbox

v2020.09.01.7.

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+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 [e-mail]: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

Changelog

2020.06.01.dev1 (2020-05-17)

• First release on PyPI.

2020.09.01.dev1 (2020-09-01)

• Update documentation.

• Added docstrings and full apidocs.

• Fix for Backward Algorithm.

• Separate the toy AV simulator from the AST wrapper.

• Change AST environments to save the cloned sim state from pre-action, not post-action.

• Add travis deployment to PyPI.

• Removal of unsupported files.

• Expanded codecov to 90+%.