Python API

The package root exports the supported API. Everything else, including adapters and container internals, is private and can change without notice.

from mlbundle import (
    FORMAT_NAME,  # "mlbundle"
    FORMAT_VERSION,  # "1"
    ArchiveLimits,
    LoadedBundle,
    Manifest,
    MetadataView,
    OutputPrediction,
    export,
    inspect,
    load,
)

Model records are strict, frozen Pydantic models: they reject unknown fields and type coercion.

Functions

export

def export(
    model: object,
    destination: str | PathLike[str],
    *,
    training_data: object | None = None,
    feature_names: Sequence[str] | None = None,
    class_labels: Sequence[Sequence[Label]] | None = None,
    target_names: Sequence[str] | None = None,
    sample_rows: int | None = None,
    onehot_separator: str = "__",
    metadata: Mapping[str, JsonValue] | None = None,
    overwrite: bool = False,
    limits: ArchiveLimits | None = None,
) -> MetadataView

Publishes a bundle only after its claims survive a native serialization round trip and the reader validates it. Requires the schema extra. Label is a JSON scalar class label. Every argument is described in Export a bundle.

inspect

def inspect(source: BundleSource, *, limits: ArchiveLimits | None = None) -> MetadataView

Validates the container, manifest, JSON bounds, hashes, schema and samples without importing a model library. It cannot attest model-derived claims.

load

def load(source: BundleSource, *, limits: ArchiveLimits | None = None) -> LoadedBundle

Everything inspect does, then deserializes through the closed adapter and verifies recomputed claims and structure. BundleSource is str | PathLike[str] | bytes | BinaryIO.

Results

MetadataView

Frozen dataclass returned by export and inspect.

Field	Type
manifest	Manifest
members	tuple[MemberName, ...], in canonical order
schema	dict[str, JsonValue] | None
samples	list[dict[str, JsonValue]] | None

LoadedBundle

Frozen dataclass returned by load, constructed only after verification.

Member	Type / behaviour
manifest, members, schema, samples	As in MetadataView.
structure	dict[str, JsonValue] | None
native_model	The native model object.
predictor	The adapter’s uniform predictor.
kind	ModelKind
library	Library
class_name	str, the estimator class
feature_names	tuple[str, ...]
feature_origin	FeatureOrigin
class_labels	list[list[Label]] | None, output-major
predict_proba(data)	list, classifiers only; otherwise ValueError
predict(data)	list[float] or rows of floats, regressors only; otherwise ValueError
row_from_record(record)	list[float] in feature order
predict_record(record)	tuple[OutputPrediction, ...], one per output

OutputPrediction

Frozen dataclass with output, kind, labels, probabilities, value and predicted_label. See Records and predictions.

Manifest records

Class	Fields
Manifest	format_name, format_version, bundle_id, created_at, writer_version, model, features, target, environment, integrity, references, user_metadata
ModelInfo	estimator_class, kind, hyperparameters, n_features, n_classes, n_outputs, library, serialization, pipeline and state fields
FeatureInfo	names, dtypes (training), model_dtypes (verified), origin
TargetInfo	class_labels, names
TrainingEnvironment	python_version, python_implementation, os, os_version, architecture, packages, onnx
OnnxInfo, Opset	ONNX opsets, IR version and producer
IntegrityRecord	sha256, size
MemberReferences	schema, structure, samples
ModelClaims	The model-derived claims the reader recomputes

Field types and verification rules are normative in Format 1.

Enums

All are StrEnums whose values appear in the manifest.

Enum	Values
Library	sklearn, xgboost, catboost, onnx
ModelKind	classifier, regressor
SerializationFormat	skops, ubjson, cbm, onnx, pipeline
FeatureOrigin	model, dataframe, declared, generated
MemberName	manifest.json, model.skops, model.ubj, model.cbm, model.onnx, schema.json, structure.json, samples.json

Limits and errors

ArchiveLimits lowers the safety ceilings. Every exception mlbundle raises for a bundle or model derives from MLBundleError; see Errors.

mlbundle.__version__ is the installed package version.