https://github-investigation.firebaseapp.com/

This is a side project I did for fun over the weekend. I've been using Mapd at work for a genomics project and wanted to apply what I've learned to some different data.

The data is extracted from the github archive using BigQuery and data/extract_prs.sql.

The backend is powered by containerized MapD databases on GCP n1-highcpu-8 instances. I'm using CPU only instances because the data size is relatively small and I'm on a budget :) so performance wise this won't do MapD justice. If you're interested in the type of performance MapD is capable of, Mark Litwintschik blog is a great resource.

The web app is built with angular and hosted on firebase. Visualizations are created using mapd-charting, mapd-crossfilter and mapd-connector.

MapD can work with or without a GPU. I'm using docker to simplify the deployment process. There are two docker compose files, one for loading the data data/docker-compose.yml, and one for running the services docker-compose.yml. Both use a .env file to determine what MapD image to run. Currently I'm using my own personal images on docker hub, but you can build your own with https://github.com/shusson/docker-mapd or follow the official guide https://github.com/mapd/mapd-core/tree/master/docker.

I haven't automated getting the data from the github archive. The manual process is as follows:

• Use bigquery web console to create a new table from the results of data/extract_prs.sql.
• Export table into a GCP bucket (since the data is larger than 1GB you will have to use sharding)
• Download bucket contents and process with data/process_bq_data.sh to create a file data/prs.csv which will be loaded by mapd.

Load the data (only required the first time)

cd data
dc up

Run database and services

cd ..
dc up

Modify the environment file dashboard/src/environments/environment.ts to point to your MapD server.

cd dashboard
npm install
ng serve

Browse to http://localhost:4200