gwrs 0.4.0.9015
- Adds selected-model local coefficient inference to
gwr_select()by default, with raw two-sided reference p-values and BH adjustment separately across locations for each coefficient. Setlocal_inference = FALSEto skip it. Search candidates retain standard diagnostics; only the selected GWR final refit uses full diagnostics. Model ranking is unchanged. -
gwr_local_inference()also accepts a selection result, retains selection context, and explicitly labels reference calibration as unestablished. BH does not account for response-based model/tuning selection. - Invalid reference degrees of freedom now return NA p-values with a warning instead of silently switching to normal references. An explicit
df = "normal"remains available. Missing tests have NA significance flags and finite-or-NA summaries, with available/unavailable counts. - Clarifies that local coefficient-zero tests differ from F3 spatial constancy, and documents per-coefficient versus global adjustment families.
gwrs 0.4.0.9014
- Adds
gwr_select()for joint predictor-subset, kernel and bandwidth selection in ordinary Gaussian GWR. AICc is the default; AIC, BIC, GCV and conditional leave-one-out error are also reported on the same observations. - Supports exhaustive or deterministic stepwise search, common user-supplied Moran weights, design inspection, bounded execution and verified resume.
- Improves candidate-grid construction and contiguous C++ F-moment access. Numerical moment definitions are preserved. Small benchmarks measured about 2.1–2.4 times faster F diagnostics at n=320; this is not a whole-search claim.
- Labels all F references exploratory and uncalibrated. F1 and F2 transform the same RSS ratio and do not provide independent confirmations.
- Adds an executable joint-selection guide and corresponding website reference pages, retaining the ACS period-name alias and historical reproduction engines.
gwrs 0.4.0.9013
Adds
acs2020_2024_countiesas the primary name for the unchanged 2020–2024 ACS county table, with period-estimate documentation;acs2024_countiesremains available for compatibility.Focused documentation and examples for GWR-Ridge, GWR-Lasso, GWR-ENET, GWR-SCAD and GWR-MCP.
Retains the validated coordinate cache, regularization paths and spatial prediction implementation.
Documentation packaging does not change numerical functions, defaults, tuning or stopping rules.