A/B Testing for ML applications

Deploy A/B testing infrastructure in a containerized microservice architecture for Machine Learning applications.

Requirements

This repository uses Kubernetes, helm, ambassador, seldon-core, and seldon-core-analytics.

Getting started

Install prerequisites

1. Install helm

   brew install helm
   

2. Create a Kubernetes cluster: AKS, EKS, GKE or local cluster. For local clusters, one can use minikube, kind, or k3s. For instance:

   1. GKE

    gcloud container clusters create demo-cluster-ab-test \
      --zone=europe-west3-a \
      --disk-size=30GB \
      --cluster-version=1.24.12-gke.1000 \
      --machine-type=e2-highcpu-4

Train model

Prepare the model artifacts:

make train

Build container images

One can build the images locally, or use Cloud Submit:

Locally

docker build -t ab-test:a -f Dockerfile.a .
docker build -t ab-test:b -f Dockerfile.b .
docker build -t streamlit-app:v1.0.0 -f Dockerfile.streamlit .

Cloud Submit

sh build-modela.sh
sh build-modelb.sh
sh build-streamlit.sh

Deploy required components

• emissary-ingress

make emissary

• Prometheus

make prometheus

• Grafana

make grafana

• seldon-core

make seldon-core

• podmonitor

make podmonitor

Deployment

make abtest

Start Streamlit App

make streamlit

Port-forward Grafana

make port-grafana

Contact

Sadik Bakiu (sadik [at] data-max.io)

Developed with ❤ at Data Max

References

• Model Training: https://github.com/SeldonIO/seldon-core/tree/master/examples/models/sklearn_spacy_text
• Seldon-core installation:
  • Install Emissary Ingress: https://www.getambassador.io/docs/emissary/latest/topics/install
  • Install seldon-core operator: https://docs.seldon.io/projects/seldon-core/en/latest/workflow/install.html
• PodMonitor with Prometheus installation: https://docs.seldon.io/projects/seldon-core/en/latest/analytics/analytics.html
• Custom metrics: https://docs.seldon.io/projects/seldon-core/en/latest/examples/runtime_metrics_tags.html
• Streamlit: https://docs.streamlit.io/library/cheatsheet