ocpm_engine.analytics

Typed facade over the Rust model-construction and scoring kernels. Models are built and scored directly from sufficient-statistic rows; no event-level data is needed.

Input type

TransitionCount(source: str, target: str, edge_type: str,
                train_count: int, test_count: int)

One aggregate row per transition, carrying its training-window and test-window frequencies. All scoring functions below take an iterable of these rows.

Functions

dfg_conformance(rows: Iterable[TransitionCount], *,
                coverage: float = 0.95) -> ConformanceScore

Build a frequency-covered DFG in Rust and score held-out counts. Returns ConformanceScore(fitness, conforming, deviations).

variant_conformance(variants: Sequence[str],
                    train_counts: Sequence[int],
                    test_counts: Sequence[int], *,
                    coverage: float = 0.95)
    -> tuple[float, int, int, int, tuple[str, ...]]

Build a frequency-covered complete-variant model and score it. Returns (fitness, conforming, deviations, test_total, model_variants).

next_activity(rows: Iterable[TransitionCount]) -> PredictionScore

Fit deterministic next-activity choices on training counts and score held-out counts. Returns PredictionScore(accuracy, correct, test_total, predictions), where each prediction is a (source, predicted, actual) label triple.

bottleneck_order(frequencies: Sequence[int],
                 mean_durations: Sequence[float]) -> tuple[int, ...]

Stable transition indexes ranked by duration, frequency, then input order.

frequency_drift(labels: Sequence[str],
                baseline_counts: Sequence[int],
                current_counts: Sequence[int], *,
                top_n: int = 10) -> DriftScore

Score and explain change between two aligned frequency distributions. Returns DriftScore(divergence, baseline_total, current_total, contributors); each DriftContributor carries label, baseline_share, current_share, share_delta, and js_contribution.

Result types

Type	Fields
ConformanceScore	fitness, conforming, deviations
PredictionScore	accuracy, correct, test_total, predictions
DriftScore	divergence, baseline_total, current_total, contributors
DriftContributor	label, baseline_share, current_share, share_delta, js_contribution

Example

from ocpm_engine import TransitionCount, dfg_conformance, next_activity

rows = [
    TransitionCount("Create", "Approve", "directly_follows", 900, 95),
    TransitionCount("Create", "Reject", "directly_follows", 100, 5),
]

conformance = dfg_conformance(rows, coverage=0.95)
print(conformance.fitness, conformance.deviations)

prediction = next_activity(rows)
print(prediction.accuracy, prediction.predictions)