Skip to content

Errors

Validation of bundles, models, schemas and prediction inputs fails with a bounded MLBundleError. Messages never echo record values or native third-party exception text. They can name a member, a claim, an escaped and truncated feature key, an extra or a writer/version hint. Filesystem failures and API misuse remain builtin exceptions, listed at the end.

from mlbundle import MLBundleError, load
try:
bundle = load(b"not a bundle")
except MLBundleError as error:
print(type(error).__name__) # InvalidContainerError
MLBundleError
├── InvalidContainerError
├── LimitExceededError
├── UnsupportedFormatVersionError
├── InvalidManifestError
├── InvalidJSONError
├── IntegrityMismatchError
├── ClaimMismatchError
├── UnsupportedModelError
├── InvalidModelError
│ ├── EnvironmentMismatchError
│ └── PredictionInputError (also a ValueError)
├── InvalidSchemaError
├── InvalidSamplesError
├── MissingOptionalDependencyError
└── AdapterNotImplementedError
Error Raised when Attributes
InvalidContainerError The ZIP violates the container rules: unknown, duplicate or missing members, unsafe names, bad records or CRCs, a missing schema in an old bundle.
LimitExceededError An input exceeds an archive or member limit.
UnsupportedFormatVersionError The manifest declares a format version other than 1.
InvalidManifestError The manifest fails strict validation; may hint at a different writer version.
InvalidJSONError A JSON member is malformed or uses an unsupported value.
IntegrityMismatchError A member’s SHA-256 or size disagrees with the manifest. member
ClaimMismatchError A recomputed model claim or the structure differs from the bundle. claim
UnsupportedModelError Export of a model outside the adapter boundary. untrusted_types
InvalidModelError Native model content is malformed or violates the adapter contract.
EnvironmentMismatchError A skops load failed and the bundle declares a different scikit-learn version.
PredictionInputError Invalid prediction input or record.
InvalidSchemaError Training inputs or schema.json violate the schema contract.
InvalidSamplesError Invalid sample_rows, cell conversion or sample keys.
MissingOptionalDependencyError A needed extra is not installed. extra
AdapterNotImplementedError A library has no implemented adapter. library

Identifiers in member, claim, extra and library are restricted to short ASCII names; anything else is reported as unrecognized.

Not every failure is an MLBundleError: an existing destination without overwrite=True raises FileExistsError, and calling predict on a classifier (or predict_proba on a regressor) raises ValueError.

The command line maps MLBundleError to exit code 1 and I/O failures to exit code 2.