Metrics¶
The metrics module scores model predictions against available targets and produces aggregated evaluation results.
Providing ground-truth labels¶
Without labels the pipeline falls back to using model predictions as their
own targets, which trivially scores 100% and is not informative. Configure
data.labels to point at the ground-truth labels alongside your samples; see
Data configuration for the available
labels.source, labels.column, labels.id_column, and labels.encoding
options.
When labels are missing or fail to align with predictions, raitap emits a warning so the resulting metrics are clearly flagged as fallback values.
Metrics module documentation