Stack Overflow Title Picker

Language/Framework

E2E Pipeline

Project Period

2 Weeks, Oct - Nov 2018

Stack Overflow Title Picker is an end-to-end Natural Language Processing (NLP) project built in Keras/Tensorflow, trained on a GKE cluster and served with Seldon.

  • Platform: GCP
  • Language: Python
  • Library: Kubeflow
  • Workflow Tooling:

    • Argo
    • Ksonnet
    • Kubernetes
    • Docker
    • Seldon

Description

The project uses over 2M Stack Overflow questions queried using Big Query to train an NLP model on a GKE cluster to predict a good Stack Overflow title given a question.

The primary purpose of this project is to demonstrate an end-to-end, Kubernetes-native ML workflow that is composable, portable and scalable. This project uses the Kubeflow library, which is intended to be a multi-framework ML toolkit for Kubernetes. The final model is served using Seldon.

Sample Prediction

Sample Prediction

Links