boxmot.engine.tuning.analysis

Generate descriptive, conditional-aware reports for saved tuning results.

LOGGER module-attribute

LOGGER = getLogger(__name__)

SAMPLE_COLOR module-attribute

SAMPLE_COLOR = '#3976a8'

NEAR_COLOR module-attribute

NEAR_COLOR = '#df9b19'

BASELINE_COLOR module-attribute

BASELINE_COLOR = '#16836b'

BEST_COLOR module-attribute

BEST_COLOR = '#c93b42'

IDF1_COLOR module-attribute

IDF1_COLOR = '#8656a8'

CATEGORY_COLORS module-attribute

CATEGORY_COLORS = (
    "#3976a8",
    "#e58a32",
    "#5b9b60",
    "#b66ba0",
    "#7e7e7e",
    "#53a5ac",
)

generate_tune_analysis

generate_tune_analysis(
    tune_dir: Path,
    tracker_name: str = "",
    n_trials: int | None = None,
) -> Path | None

Generate PNG reports from saved CSV, search metadata, and trial JSON.

The saved schema defines parameter activity and bounds. Missing metadata omits the corresponding diagnostics, while score-only reports remain useful. Returns the overview path, or None when no valid scored trials are available.

generate_selection_analysis

generate_selection_analysis(
    tune_dir: Path,
) -> tuple[Path, ...]

Rebuild selection images from reports, independently of completed trials.