Skip to content

Latest commit

 

History

History
91 lines (67 loc) · 3.12 KB

other.md

File metadata and controls

91 lines (67 loc) · 3.12 KB
Raw

Other Stuff

[deprecated] Jupyter Notebooks

Please see the jupyter_notebooks folder for an interactive version of the above examples.

packaging info

python -m pip install build twine

# update version number in pyproject.toml (e.g., 0.0.2a0 to test out a candidate 0.0.2 release)
cd nfl_veripy
python -m build
python -m twine upload --repository testpypi dist/nfl_veripy-0.0.2a0*

# in a separate terminal...
python -m virtualenv nfl_veripy_pypi_venv
source nfl_veripy_pypi_venv/bin/activate
python -m pip install "jax_verify @ git+https://gitlab.com/neu-autonomy/certifiable-learning/jax_verify.git" "crown_ibp @ git+https://gitlab.com/neu-autonomy/certifiable-learning/crown_ibp.git"
python -m pip install --extra-index-url https://test.pypi.org/simple/ nfl-veripy==0.0.2a0
python -m nfl_veripy.example --config example_configs/icra21/fig3_reach_lp.yaml

# if that works, build & release the pkg to the real pypi:
# - update version number in pyproject.toml (e.g., 0.0.2)
python -m build
python -m twine upload dist/nfl_veripy-0.0.2*

Aside: We acknowledge that the method for installing jax_verify and crown_ibp (dependencies of nfl_veripy) in the non-developer setup way is a little unconventional. It would be better to simply include these as dependencies of nfl_veripy and let pip find those packages, but (a) those packages are not available (or are too outdated) on PyPI, and (b) it is not allowed to include dependencies with direct URLs when releasing a package on PyPI. If there's a better way of doing this we would love to hear about it!

OVERT.jl

To use OVERT, first start a julia session, then start the HTTP server that hosts an OVERT query endpoint:

julia> include("<path_to_repo>/nfl_veripy/nfl_veripy/utils/run_overt.jl")

There won't be any output but you should now have a server at http://127.0.0.1:8000/ so you can send a POST request to call OVERT. Note that the reason for having an HTTP server is that compiling julia libraries can take a while and you don't want to wait for this every time you run the python script.

Now you can run OVERT as a propagator, just like CROWN or SDP:

python -m nfl_veripy.example \
    --partitioner None \
    --propagator OVERT \
    --system double_integrator \
    --state_feedback \
    --t_max 5 \
    --show_plot

Run tests locally

docker build -t nnrob -f docker/Dockerfile .
docker run -v $PWD:/home/nn_robustness_analysis nnrob:latest python -m nfl_veripy.tests.test

Run new closed-loop partitioner example

python -m nfl_veripy.example_backward \
    --partitioner None \
    --propagator CROWN \
    --system double_integrator \
    --state_feedback \
    --t_max 2 \
    --show_plot --overapprox --skip_save_plot

mypy debugging

Only the first time, get the type stubs for some installed pip pkgs using the typeshed (most that don't live in here are ignored in mypy.ini):

mypy --install-types

Run this to check types of entire nfl_veripy pkg (excluding a few files as defined in mypy.ini):

cd nfl_veripy
mypy -p nfl_veripy