NEWS
sqlm 1.0.0
sqlm 0.1.0 (2026-02-04)
Features
lm_sql() fits OLS linear regression models entirely in-database via a
single SQL aggregation query.- Automatic dummy encoding for character and factor predictors with
CASE WHEN SQL generation.
- Interaction terms (
*, :) fully supported, including interactions between
numeric and categorical variables.
- Dot expansion (
y ~ .) expands to all non-response columns.
- Formula transforms:
I(), log(), and sqrt() translated to SQL
equivalents (POWER, LN, SQRT).
- Date and datetime predictor support:
Date and POSIXct columns
automatically converted to numeric in SQL, matching base R's lm() behavior.
- Grouped regression via
dplyr::group_by() computes multiple models in a
single GROUP BY query.
- No-intercept models (
y ~ 0 + x) handled correctly.
- S7-based result objects with
broom::tidy() and broom::glance() methods.
orbital integration for in-database predictions.- Cholesky decomposition for matrix inversion with
MASS::ginv() fallback
for rank-deficient designs.