e-bh-cc

Documentation soon!

Setting up the virtual environments

The simulation scripts require certain Python libraries to run. Furthermore, specific instantiations of e-BH-CC require more specific machinery. We will describe how to create virtual environments for each of the three settings with the needed imports.

CC for z-testing and t-testing

Create the Python3 virtual environment (venv) by running the following terminal command:

python3 -m venv venv_ebhcc

To activate the venv and install the required dependencies for the z-test and t-test scripts, run in the terminal

source venv_ebhcc/bin/activate
pip install -r requirements.txt

In this virtual environment, you can now run the z-testing and t-testing experiments (e.g. ztesting_CC.py and ttesting_CC.py).

CC for knockoffs

Create the Python3 virtual environment (venv) by running the following terminal command:

python3 -m venv venv_kn_mvr

To activate the venv and install the required dependencies for the knockoffs scripts, run in the terminal

source venv_kn_mvr/bin/activate
pip install -r req_knockoffs.txt
# need the choldate package for MVR knockoffs
pip install git+https://github.com/jcrudy/choldate.git@d37246f4fc1775f11b84d42b5ceba08e6392d285  

In this virtual environment, you can now run the model-X knockoffs experiments, which use the mxknockoffs_CC.py file.

(Note: we will add instructions on how to use SDP knockoffs at a later point. These require a different set of Python dependencies.)

CC for conformalized outlier detection

Create the Python3 virtual environment (venv) by running the following terminal command:

python3 -m venv venv_numba

This is named due to its usage of the numba JIT compiler, which makes the numpy operations required in our implementation of conformal selection extremely fast.

To activate the venv and install the required dependencies for conformal outlier detection scripts, run in the terminal

source venv_numba/bin/activate
pip install -r req_numba.txt

In this virtual environment, you can now run outlier detection experiments, which use the outlier_detection_CC.py file.