Geocoding at Scale on RCC Clusters

Geocoding at Scale on RCC Clusters

Abstract:

Geocoding is the process of converting written addresses into latitude and longitude coordinates, a routine first step in geospatial analysis across the social sciences, public health, economics, and urban research. Commercial and community services such as EsriSRI and Nominatim expose this capability through web forms and APIs, and for small address lists, they are sufficient. Two constraints appear as soon as the work scales. First, free and low-cost tiers cap both the total number of addresses and the request rate, so a study with hundreds of thousands or millions of records becomes impractical or expensive. The second is that API-based geocoding requires internet access, which most compute nodes on shared clusters, including those at the UChicago Research Computing Center, do not have for security reasons. Many address datasets are also sensitive under IRB or data use agreements that prohibit transmitting records to third-party services at all.

This hands-on workshop presents the geocoding options RCC provides to the UChicago community and shows how to match each option to the size and sensitivity of a given dataset. We begin with the EsriSRI geocoding platform available under the institutional license RCC maintains. We then turn to offline geocoding with Pelias, which RCC has configured on Midway3, and cover how to structure a Pelias job so that millions of addresses are processed in parallel rather than serially. The session closes with hybrid approaches that combine offline and licensed services, and with the errors that most often compromise large geocoding runs, including address normalization failures, silent low-confidence matches, and misinterpreted coordinate output.

Objectives:

Participants will learn to:

• Choose an appropriate geocoding approach based on dataset size, sensitivity, and network requirements.
• Geocode addresses using the Esri platform available under RCC's institutional license.
• Run offline geocoding with the Pelias service configured on Midway3.
• Parallelize a large geocoding job across cluster resources to process millions of addresses.
• Identify and correct common failure modes in large-scale geocoding, including normalization errors and unreviewed low-confidence matches.

Presenter: Hamid Dashti

Level: Introductory

Duration: 1:30 hours

Prerequisites: No prior geocoding or geospatial experience is required. Participants need an RCC account or a valid CNetID, which allows temporary access to be granted for the session. Familiarity with basic Linux command- line navigation is helpful but not required.