Policy Relevance, Model Upgrades & Deployment

In this release

• Onboarded our first non US city: Brisbane!
• Worked towards more streamlined deployment of our public facing visualization and exploring Docker for the same.
• Enhancements to the visualization.
  • Address search
  • Map filters
  • List of highest risk segments
  • Segment-specific panel with details for the segment selected
• Incorporating adjacency information into segments
• Moved our model to be segment based.
• Documenting our model features in a data dictionary.
• Our first attempt to understand how segments in a city can be grouped together based on their risk level, and what factors contribute to the formation of those groups.

Tech Upgrade

Upgrade to python 3
Continue data standardization work - predictions, volumes, point-based features
Upgrade pipeline process to work for cities without OSM polygons
Better Docker integration with automated image builds
Restructure predictions output to support visualization upgrade in v2.0
Implement improved code testing strategies
Move to using GeoJSON format

Pipeline

This release demonstrates running the full pipeline process for any of the demo cities (Boston MA, Cambridge MA or Washington D.C) using the data referenced in the readme. Alternatively it can be run against any other city that has compatible crashes (and optionally concerns) data available, see the readme for additional instructions.