bayesian-dpddm 1.0.1

Implements Bayesian D-PDDM for Post-Deployment Deterioration Monitoring of ML models.

Bayesian D-PDDM

Bayesian implementation of the D-PDDM algorithm for post-deployment deterioration monitoring. Bayesian D-PDDM is a Bayesian approximation to the D-PDDM algorithm which provably monitors model deterioration at deployment time. Bayesian D-PDDM:

• Flags deteriorating shifts in the unsupervised deployment data distribution
• Resists flagging non-deteriorating shifts, unlike classical OOD detection leveraging distances and/or metrics between data distributions.

Install

The easiest way to install bayesian_dpddm is with pip:

pip install bayesian_dpddm

You can also install by cloning the GitHub repo:

# Clone the repo
git clone https://github.com/opent03/bayesian_dpddm.git

# Navigate into repo directory 
cd bayesian_dpddm

# Install the required dependencies
pip install .

Usage and Tutorials

Coming soon.

Citation

Coming soon.