Scripts for setting up a long-running Clipper cluster for stress testing and other purposes. Currently trains a model on Criteo's 2014 display advertising dataset and deploys it to Clipper. The specific model is LIBFFM, the first place winner of the Kaggle competition with this dataset. The testbed is currently running on a GCP instance on Kubernetes with Redis configured to run externally in fault-tolerant mode.
If you want to set this up on a new computer, run the following:
$ python start-clipper.py clipper-server-IP redis-service-IP # starts a Clipper instance
$ python train-dataset.py path-to-dataset clipper-server-IP redis-service-IP # run periodically to train model and deploy it to Clipper
$ ./start-client.sh # start client to query Clipper at a Poisson rate