Time to Merge (TTM) as a Service

TTM is a model that classifies a GitHub PR into 1 of 10 time categories ranging from under a minute to over 4 days.

• Model training: https://github.com/aicoe-aiops/ocp-ci-analysis/tree/master/notebooks/time-to-merge-prediction/thoth-station
• Model usage: https://github.com/aicoe-aiops/ocp-ci-analysis/tree/master/models/thoth-support-github-ttm

This repo takes the TTM model and turns it into a service. This repo hosts a bot that uses GitHub webhooks to respond to pull requests. Currently, any user that installs the bot is assigned to use the Thoth trained bot. In the future, there will be a flow that auto-trains on an org or set of repos.

Output classes

• 0-1 mins
• 1-2 mins
• 2-5 mins
• 5-10 mins
• 10-30 mins
• < 2 hours
• 2-8 hours
• 8-17 hours
• < 4 days
• > 4 days

Local Testing

The application is developed using flask. Use the following step to setup local env for Testing.

1. Install and Set python env. pipenv install pipenv shell
2. Run the app. python wsgi.py

Deployment

This project is managed with kustomize. Use the following command to deploy on OpenShift cluster. kustomize build manifests/overlays/test | oc apply -f -