Skip to content

CLAUDE.md

Latest commit

 

History

History
46 lines (35 loc) · 1.59 KB

CLAUDE.md

File metadata and controls

46 lines (35 loc) · 1.59 KB
Raw

Guidance for Claude Code in INLASPDERasters.jl

Extends /CLAUDE.md. Scoped to raster-SPDE glue.

Scope

This package owns:

  • extract_at_mesh(raster, mesh) — sample a Raster at mesh vertex coordinates; handle CRS transformation via CoordRefSystems.
  • predict_raster(fit, mesh, template_raster) — project a fitted SPDE posterior back onto a raster grid matching the template's extent/resolution.
  • Type piracy is forbidden — do not add methods on Raster or INLASPDE types that could live in one of those packages.

Out of scope:

  • Prediction uncertainty aggregation (separate question; handled by core LGM's marginal machinery).
  • GDAL-specific IO — Rasters.jl abstracts that.

Dependencies allowed

Core:

  • INLASPDE — the host.
  • Rasters — the raster abstraction.

That's it. No others without an ADR.

Testing

  • test/regression/ — synthetic raster + synthetic mesh, check that extract_at_mesh recovers the analytic raster value at each mesh vertex to machine precision (no interpolation ambiguity).
  • test/regression/ — CRS mismatch: mesh in UTM, raster in WGS84; extraction respects CRS.
  • test/oracle/ — Meuse zinc fit, compare predicted raster to R-INLA's predict.inla on the same grid.

Style

  • CRS handling is the subtle part. Every CRS-crossing operation must assert equality (or explicit conversion) at the API boundary with a helpful error if the CRSs do not match.
  • Prediction rasters have the same CRS / extent / resolution as the template; callers expecting a different grid must provide a different template.