fable-model 0.2.2

Data models utilized in the FABLE ecosystem's HTTP-based services.

FABLE Model

This package contains model classes that are used in the FABLE (Federated Anonymized Bloom filter Linkage Engine) ecosystem's services for validation purposes. They were developed with the intention of creating HTTP-based services to perform Bloom filter-based record linkage in a federated setting. It includes models for data transformation, masking and bit vector matching routines that are used by the FABLE PPRL service as well as models used by the FABLE Broker service. Validation, serialization and deserialization are done using Pydantic. It is rare to use this package directly. Rather, it powers the functionalities of other packages.

Installation

pip install fable-model

PPRL service

Models for entity pre-processing, masking and bit vector matching are exposed through this package. The following examples are taken from the test suites of the PPRL service package and show additional validation steps in addition to the ones native to Pydantic.

Entity transformation

from fable_model import (
    EntityTransformRequest,
    TransformConfig,
    EmptyValueHandling,
    AttributeValueEntity,
    AttributeTransformerConfig,
    NumberTransformer,
    GlobalTransformerConfig,
    NormalizationTransformer,
    CharacterFilterTransformer,
)

# This is a valid config.
_ = EntityTransformRequest(
    config=TransformConfig(empty_value=EmptyValueHandling.ignore),
    entities=[
        AttributeValueEntity(
            id="001",
            attributes={
                "bar1": "  12.345  ",
                "bar2": "  12.345  "
            }
        )
    ],
    attribute_transformers=[
        AttributeTransformerConfig(
            attribute_name="bar1",
            transformers=[
                NumberTransformer(decimal_places=2)
            ]
        )
    ],
    global_transformers=GlobalTransformerConfig(
        before=[
            NormalizationTransformer()
        ],
        after=[
            CharacterFilterTransformer(characters=".")
        ]
    )
)

from uuid import uuid4

# Validation will fail since no transformers have been defined.
_ = EntityTransformRequest(
    config=TransformConfig(empty_value=EmptyValueHandling.ignore),
    entities=[
        AttributeValueEntity(
            id=str(uuid4()),
            attributes={
                "foo": "bar"
            }
        )
    ],
    attribute_transformers=[]
)
# => ValidationError: attribute and global transformers are empty: must contain at least one

Entity masking

from fable_model import (
    EntityMaskRequest,
    MaskConfig,
    HashConfig,
    HashFunction,
    HashAlgorithm,
    DoubleHash,
    CLKFilter,
    AttributeValueEntity,
    StaticAttributeConfig,
    AttributeSalt,
    CLKRBFFilter,
)

# This is a valid config.
_ = EntityMaskRequest(
    config=MaskConfig(
        token_size=2,
        hash=HashConfig(
            function=HashFunction(algorithms=[HashAlgorithm.sha1]),
            strategy=DoubleHash()
        ),
        filter=CLKFilter(filter_size=1024, hash_values=5),
        padding="_"
    ),
    entities=[
        AttributeValueEntity(
            id="001",
            attributes={
                "first_name": "John",
                "last_name": "Doe",
                "date_of_birth": "1987-06-05",
                "gender": "m"
            }
        )
    ]
)

# This is an invalid config since salting an attribute can only be done through a fixed value
# or another attribute on an entity, not both at the same time.
_ = EntityMaskRequest(
    config=MaskConfig(
        token_size=2,
        hash=HashConfig(
            function=HashFunction(algorithms=[HashAlgorithm.sha1]),
            strategy=DoubleHash()
        ),
        filter=CLKFilter(filter_size=1024, hash_values=5),
        padding="_"
    ),
    entities=[
        AttributeValueEntity(
            id="001",
            attributes={
                "first_name": "foobar",
                "salt": "0123456789"
            }
        )
    ],
    attributes=[
        StaticAttributeConfig(
            attribute_name="first_name",
            salt=AttributeSalt(
                value="my_salt",
                attribute="salt"
            )
        )
    ]
)
# => ValidationError: value and attribute cannot be set at the same time

# This also fails if neither a static value nor an attribute are set for salting.
_ = EntityMaskRequest(
    config=MaskConfig(
        token_size=2,
        hash=HashConfig(
            function=HashFunction(algorithms=[HashAlgorithm.sha1]),
            strategy=DoubleHash()
        ),
        filter=CLKFilter(filter_size=1024, hash_values=5),
        padding="_"
    ),
    entities=[
        AttributeValueEntity(
            id="001",
            attributes={
                "first_name": "foobar",
                "salt": "0123456789"
            }
        )
    ],
    attributes=[
        StaticAttributeConfig(
            attribute_name="first_name",
            salt=AttributeSalt()
        )
    ]
)
# => ValidationError: neither value nor attribute is set

# When using a weighted filter (RBF, CLKRBF), an error will be thrown if any attribute configuration
# provided is static, not weighted. The same applies vice versa, meaning if CLK is specified as a filter and
# weighted attribute configurations are provided.
_ = EntityMaskRequest(
    config=MaskConfig(
        token_size=2,
        hash=HashConfig(
            function=HashFunction(algorithms=[HashAlgorithm.sha1]),
            strategy=DoubleHash()
        ),
        filter=CLKRBFFilter(hash_values=5),
        padding="_"
    ),
    entities=[
        AttributeValueEntity(
            id="001",
            attributes={
                "first_name": "foobar",
                "salt": "0123456789"
            }
        )
    ],
    attributes=[
        StaticAttributeConfig(
            attribute_name="first_name",
            salt=AttributeSalt(value="my_salt")
        )
    ]
)
# => ValidationError: `clkrbf` filters require weighted attribute configurations, but static ones were found

# Weighted filters (RBF, CLKRBF) always require weighted attribute configurations. If none
# are provided, validation fails.
_ = EntityMaskRequest(
    config=MaskConfig(
        token_size=2,
        hash=HashConfig(
            function=HashFunction(algorithms=[HashAlgorithm.sha1]),
            strategy=DoubleHash()
        ),
        filter=CLKRBFFilter(hash_values=5),
        padding="_"
    ),
    entities=[
        AttributeValueEntity(
            id="001",
            attributes={
                "first_name": "foobar",
                "salt": "0123456789"
            }
        )
    ]
)
# => ValidationError: `clkrbf` filters require weighted attribute configurations, but none were found

# If a configuration is provided for an attribute that doesn't exist on some entities, validation fails.
_ = EntityMaskRequest(
    config=MaskConfig(
        token_size=2,
        hash=HashConfig(
            function=HashFunction(algorithms=[HashAlgorithm.sha1]),
            strategy=DoubleHash()
        ),
        filter=CLKFilter(filter_size=1024, hash_values=5),
        padding="_"
    ),
    entities=[
        AttributeValueEntity(
            id="001",
            attributes={
                "first_name": "foobar"
            }
        )
    ],
    attributes=[
        StaticAttributeConfig(
            attribute_name="last_name",
            salt=AttributeSalt(value="my_salt")
        )
    ]
)
# => ValidationError: some configured attributes are not present on entities: `last_name` on entities with ID `001`

Bit vector matching

from fable_model import (
    VectorMatchRequest,
    MatchConfig,
    SimilarityMeasure,
    BitVectorEntity,
    SimilarityAggregator,
)

_ = VectorMatchRequest(
    config=MatchConfig(
        measures=[SimilarityMeasure.jaccard, SimilarityMeasure.cosine],
        thresholds=0.8,
        aggregator=SimilarityAggregator.avg,
    ),
    domain=[
        BitVectorEntity(
            id="D001",
            value="kY7yXn+rmp8L0nyGw5NlMw=="
        )
    ],
    range=[
        BitVectorEntity(
            id="R001",
            value="qig0C1i8YttqhPwo4VqLlg=="
        )
    ]
)

Broker service

The broker module contains models used by the Broker service to validate requests and responses to create, manage and delete matching sessions. Follow the link to the repository of that service for further information.

License

MIT.