Preprocess clinical trials data (sequential snapshots) into ML-able outlier detection datasets and evaluate outlier detection approaches.
With access to the data, you should be able to reproduce all results and visualizations with:
$ OPENBLAS_NUM_THREADS=1 METAOD_SERVICE_URL=http://hostname:port poetry run dvc repro
where http://hostname:port is the base URL of the MetaOD service launched under Python 3.7.
We assume that python3 points to the Python 3.7 executable with the relevant MetaOD dependencies
available (see below).
MetaOD depends on outdated software versions. To integrate it with our evaluation, we stand it up as an HTTP service. To launch:
$ python3 -m uvicorn traq.services.metaod_api:app --host 0.0.0.0
The requirements are in metaod_requirements_py3.7.txt and run fine under Python 3.7.17 (a secure,
but non-bugfixed version of Python as of the time of this writing).
Poetry is used to manage dependencies (see Usage). It is recommended
that you set up Git hooks using pre-commit to auto-lint your code prior
to committing, which will ensure the CI passes on the server side:
$ poetry run pre-commit install
This will auto-install a set of Git hooks into the local copy of your repo which will automatically
run each time to you run git commit. These hooks include Black
for auto-formatting, flake8
for auto-linting, and isort for automatic import sorting.