coonfit#
coonfit — Parallel multiple linear regression for spatial raster data.
This package implements a block-parallel multiple linear regression workflow where both predictors and the response variable are provided as raster bands (GeoTIFF files). Regression weights are estimated from valid pixels across the full spatial extent and can be used to produce spatially continuous model prediction maps.
Submodules#
- inference
Core functions for constructing the predictor matrix
X, computingX.T @ X, solving the normal equations, and extracting response vectors from raster data. Supports optional intercept terms, spatial windowing, and boolean pixel selectors.- parallel
High-level parallelized workflows for computing regression weights (
compute_weights()), generating model prediction rasters (compute_model()), and evaluating model quality via RMSE and R².- parallel_helpers
Internal worker functions used by
parallelfor block-wise computation of partial matrix products, beta coefficients, residual sums, and predictor consistency checks.- helper
Utility functions for detecting rank-deficient predictor matrices and counting usable pixels within a boolean selector mask.
- exceptions
Custom exceptions raised during predictor validation and inference (e.g.
InvalidPredictorError,InferenceError).