Encoders API

VSAX encoders convert structured data into hypervector representations.

Base Classes

vsax.encoders.AbstractEncoder

Bases: ABC

Abstract base class for encoding structured data into hypervectors.

All encoders must implement the encode() method. Encoders can optionally implement fit() for learned encodings and decode() for reconstruction.

Attributes:

Name	Type	Description
model		The VSAModel instance defining the VSA algebra.
memory		The VSAMemory instance for accessing basis hypervectors.

Example

class MyEncoder(AbstractEncoder): ... def encode(self, data): ... # Custom encoding logic ... return self.memory["basis"] encoder = MyEncoder(model, memory) hv = encoder.encode(some_data)

Source code in vsax/encoders/base.py

class AbstractEncoder(ABC):
    """Abstract base class for encoding structured data into hypervectors.

    All encoders must implement the `encode()` method. Encoders can optionally
    implement `fit()` for learned encodings and `decode()` for reconstruction.

    Attributes:
        model: The VSAModel instance defining the VSA algebra.
        memory: The VSAMemory instance for accessing basis hypervectors.

    Example:
        >>> class MyEncoder(AbstractEncoder):
        ...     def encode(self, data):
        ...         # Custom encoding logic
        ...         return self.memory["basis"]
        >>> encoder = MyEncoder(model, memory)
        >>> hv = encoder.encode(some_data)
    """

    def __init__(self, model: VSAModel, memory: VSAMemory) -> None:
        """Initialize the encoder.

        Args:
            model: The VSAModel instance.
            memory: The VSAMemory instance with basis symbols.
        """
        self.model = model
        self.memory = memory

    @abstractmethod
    def encode(self, *args: Any, **kwargs: Any) -> AbstractHypervector:
        """Encode data into a hypervector.

        Args:
            *args: Positional arguments (encoder-specific).
            **kwargs: Keyword arguments (encoder-specific).

        Returns:
            The encoded hypervector.

        Raises:
            NotImplementedError: This is an abstract method.

        Note:
            Different encoders may have different signatures. See specific
            encoder documentation for details.
        """
        pass

    def fit(self, data: Any) -> None:
        """Optionally fit encoder parameters to data.

        This method is optional and can be used for learned encodings.
        By default, it does nothing.

        Args:
            data: The training data.
        """
        pass

Functions

__init__(model, memory)

Initialize the encoder.

Parameters:

Name	Type	Description	Default
model	VSAModel	The VSAModel instance.	required
memory	VSAMemory	The VSAMemory instance with basis symbols.	required

Source code in vsax/encoders/base.py

def __init__(self, model: VSAModel, memory: VSAMemory) -> None:
    """Initialize the encoder.

    Args:
        model: The VSAModel instance.
        memory: The VSAMemory instance with basis symbols.
    """
    self.model = model
    self.memory = memory

encode(*args, **kwargs) abstractmethod

Encode data into a hypervector.

Parameters:

Name	Type	Description	Default
*args	Any	Positional arguments (encoder-specific).	()
**kwargs	Any	Keyword arguments (encoder-specific).	{}

Returns:

Type	Description
AbstractHypervector	The encoded hypervector.

Raises:

Type	Description
NotImplementedError	This is an abstract method.

Note

Different encoders may have different signatures. See specific encoder documentation for details.

Source code in vsax/encoders/base.py

@abstractmethod
def encode(self, *args: Any, **kwargs: Any) -> AbstractHypervector:
    """Encode data into a hypervector.

    Args:
        *args: Positional arguments (encoder-specific).
        **kwargs: Keyword arguments (encoder-specific).

    Returns:
        The encoded hypervector.

    Raises:
        NotImplementedError: This is an abstract method.

    Note:
        Different encoders may have different signatures. See specific
        encoder documentation for details.
    """
    pass

fit(data)

Optionally fit encoder parameters to data.

This method is optional and can be used for learned encodings. By default, it does nothing.

Parameters:

Name	Type	Description	Default
data	Any	The training data.	required

Source code in vsax/encoders/base.py

def fit(self, data: Any) -> None:
    """Optionally fit encoder parameters to data.

    This method is optional and can be used for learned encodings.
    By default, it does nothing.

    Args:
        data: The training data.
    """
    pass

Core Encoders

• ScalarEncoder - Numeric values
• SequenceEncoder - Ordered sequences
• SetEncoder - Unordered collections
• DictEncoder - Key-value pairs
• GraphEncoder - Graph structures