Repository navigation
GeoAI-ready benchmark and tutorial dataset, benchmark-v1.0.1 (manuscript snapshot, archived on Zenodo)" #520
jshoughtaling
announced in
Announcements
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
GeoAI-ready benchmark and tutorial dataset, v1.0.1 (manuscript snapshot, archived on Zenodo)
Tag:
benchmark-v1.0.1· Commit: [TBD] · Zenodo DOI: [TDB] (all versions: [TBD])This release is the code and input snapshot behind the manuscript An OMOP-Compatible Exposure Infrastructure for GeoAI-Ready Environmental Health Research: The OHDSI GIS Toolchain (OHDSI GIS Working Group). It also provides frozen materials for the OHDSI GIS tutorial on October 20, 2026. All data are synthetic; the exposure and boundary sources are public.
It supersedes
benchmark-v1.0.0(commit479b371). Unlike that release, it is archived on Zenodo and includes citation metadata (CITATION.cff,.zenodo.json). The database (GaiaDB), R package (GaiaCore), metadata catalog (GaiaCatalog) and deployment (GaiaDocker) live in separate repositories and are referenced under Reproducibility.What changed since benchmark-v1.0.0
The datasets, generators and answer keys are unchanged. The changes are analyses, scripts and documentation:
syntheticDataGIS/benchmark/tract_pm25_seam_check.R): aggregated to counties, the tract series (two CDC releases, 2014–2015 and 2016–2019) agrees with the single-release county series to within 0.003 µg/m³ in every month, with no step at January 2016. The test cannot detect shifts that cancel within a county.syntheticDataGIS/tract/timing_bundle.sh, run withbenchmark/baseline_join.sh): restores the tract tables into a freshgaia-dbcontainer on another machine. On one native amd64 machine the Gaia join took a median of 28 s against 17 s for the equivalent hand-written PostGIS SQL (about 1.7 times, three runs, identical rows).geocoding_error_propagation.R --variant exposure|ses --fast): displacing only the exposure leaves the effect estimates unchanged; displacing only SES reproduces the upward drift.Contents
geocode-benchmark-2023(CC BY 4.0)..github/benchmark/run_checks.sh) and a beta R tutorial track.Reproducibility
c62b5a0(2026-10-07), tract benchmark built at15efabc(2026-10-08). Both are ancestors of this tag. No generator code changed betweenbenchmark-v1.0.0and this release, and the built data files are not rebuilt;syntheticDataGIS/data/build_info.csv(tutorial) andBUILD_INFO.txtof the tract build record the build commits.830b75a(tutorial build),11cbafc(tract build),a5e077f(adds the stress-test entries).ohdsi/gaia-db@sha256:bd044e2931b11d98a99e730582ab11788fc5beac1442b5c0ada1871c6398957a(all builds and the timing run). R image:ohdsi/gaia-core:main, digestsha256:3216156e902eb942a9e3de11112f8528cad6681ee4353dc2d5e7a4e4d82cd7b2at the time of the release checks; a pinned tag will be created after the tutorial.570a40b, GaiaCore7f26c7d, GaiaDocker0975e65(the deployment used in the tutorial).Known limitations
Not included
Manuscript results, figures and tables, which will be deposited with the publication.
Licence and third-party content
Apache 2.0 for code. CC BY 4.0 for the geocoding error distributions. U.S. Census TIGER/Line boundary files are not redistributed (only county identifiers and names). The repository does redistribute derived CDC monthly PM2.5 series (
syntheticDataGIS/data/county_pm25_monthly.csv.gzand the exposure rows derived from it; U.S. federal data) and a small mini vocabulary for the tutorial, which includes OMOP Exposome, SDOH and GIS concepts and a few concepts from SNOMED CT (25), RxNorm (15), LOINC (15) and UCUM (12); these retain the terms of their sources (seesyntheticDataGIS/data/VOCABULARY_NOTICE.mdfor the attribution notices).How to cite
Cite this record using the Zenodo DOI above or the "Cite this repository" button (from
CITATION.cff). The article citation will be added after publication.This discussion was created from the release GeoAI-ready benchmark and tutorial dataset, benchmark-v1.0.1 (manuscript snapshot, archived on Zenodo)".
All reactions