Checks if current metrics represent a regression from the last run.
Usage
check_benchmark_regression(
conn,
step_name,
format,
current_metrics,
lower_is_better = c("mlogloss", "logloss", "log_loss", "brier_score", "poisson_loss",
"rmse", "mae"),
regression_threshold = 2
)Arguments
- conn
DBI connection
- step_name
Character. Pipeline step name
- format
Character. Cricket format
- current_metrics
Named list of current metric values
- lower_is_better
Character vector of metric names where lower = better (default: common loss metrics)
- regression_threshold
Numeric. Percentage threshold to flag regression (default 2 = 2% worse triggers warning)