Intro
openLGD is a Python powered library for the statistical estimation of Credit Risk Loss (Also loss-given-default or LGD) models. openLGD can be used both as standalone library or in a federated learning context where data remain in distinct (separate) servers
Key Information
- Author: Open Risk
- License: Apache 2.0
- Development website: Github
- Mathematical Documentation around LGD Models: Open Risk Manual
- Discussion: Open Risk Commons
NB: This is a very early alpha release. openLGD is still in active development
Standalone Mode
The standalone mode is illustrated via the script standalone_run.py
Federated Mode
Getting started with the federated demo
- Clone the repo in a local linux environment
- Install the dependencies in a virtual environment
- Fire up a number of flask servers on different shells. Check the Spawn Cluster Script for how to export the environment. This will fire up several xterm's where server output is logged
- Run the Controller script to perform the demo calculation
Fabric based configuration
Going forward we'll use fabric and yaml to ease deployment. Check Fabfile for preliminary functionality
Dependencies
- The statistical model estimation is currently using scikit-learn / statstmodels components
- The model server is based on the python flask framework.
The complete dependency list in the requirements file
Startup of the model servers:
The demo Model Servers are python/flask based servers
- The model servers should startup on ports http://127.0.0.1:500X/ where X is the serial number
- You can check the server is live by pointing your browser to the port
- or by using curl from the console (curl -v http://127.0.0.1:500X/)
Model Server API endpoints:
The general structure of the simplified API is
- GET http://127.0.0.1:500X/ API Root, indicating the server is live
- GET http://127.0.0.1:500X/start URL to get initial locally estimated parameters (cold start)
- POST http://127.0.0.1:500X/update URL to post current averaged parameters (warm start)