Vectors#

Vector, SparseVector, Distance and VectorIndex are importable from agensgraph; the rest from agensgraph.vector. See Embedding vectors.

Values#

class agensgraph.Vector(values: Iterable[float] | None = None, *, half: bool = False)#

A dense vector, single precision, kept as the bytes it arrived in until a number is read. Sent as a vector. half=True reads text at half precision.

property values: array[float]#: The numbers, an array('f').

tolist() → list[float]#: The numbers, a list.

to_bytes() → bytes#: pgvector’s binary form.

index(value: float) → int#

count(value: float) → int#

classmethod from_wire(raw: bytes, *, half: bool = False) → Vector#: From pgvector’s binary form.

classmethod from_wire_text(text: bytes, *, half: bool = False) → Vector#: From pgvector’s text form.

class agensgraph.SparseVector( entries: Mapping[int, float] | Iterable[tuple[int, float]], dimensions: int, )#

A sparse vector of dimensions, from a mapping or pairs of index and value. Indices count from zero. Sent as sparsevec text.

property dimensions: int#

property indices: tuple[int, ...]#

property values: tuple[float, ...]#

to_dict() → dict[int, float]#

to_dense() → list[float]#

to_text() → str#: pgvector’s text form, counting from one.

classmethod from_dense(values: Sequence[float]) → SparseVector#

classmethod from_text(text: str | bytes) → SparseVector#

class agensgraph.Distance#

A distance, whose value is its operator.

L2#: <->

INNER_PRODUCT#: <#>, negated

COSINE#: <=>

L1#: <+>

HAMMING#: <~>, for bits

JACCARD#: <%>, for bits

property operator_class: str#: The vector operator class that serves it.

property is_for_bits: bool#

class agensgraph.VectorIndex( label: str, name: str, property: str, method: str, operator_class: str | None, type: str | None, dimensions: int | None, options: dict[str, str], )#

A vector index, in parts.

label: str#

name: str#

property: str#

method: str#

operator_class: str | None#: None for the method’s default

type: str | None#

dimensions: int | None#

options: dict[str, str]#: the settings it was built with, as strings

Statements#

agensgraph.vector.generated_column(name: str, dimensions: int, *, type: str = 'vector') → str#: A column definition, name vector(n) generated, for CREATE VLABEL.

agensgraph.vector.vector_index( label: str, property: str, *, dimensions: int | None = None, type: str = 'vector', method: str = 'hnsw', operator_class: str = 'vector_cosine_ops', name: str | None = None, options: Mapping[str, object] | None = None, ) → str#: A property index for a vector property. Pass dimensions for a property in the map, and leave it out for a column of its own.

agensgraph.vector.nearest( label: str, property: str, *, dimensions: int | None = None, type: str = 'vector', limit: int = 10, operator: str = '<=>', variable: str = 'n', ) → str#: The nearest-neighbour search matching vector_index(), with one %s for the query vector.

agensgraph.vector.search_option_statements( options: Mapping[str, object], *, local: bool = True, ) → list[str]#: The SET statements vector_search_options() sends.

agensgraph.vector.SEARCH_OPTIONS = {'hnsw.ef_search': <class 'int'>, 'hnsw.iterative_scan': <class 'str'>, 'hnsw.max_scan_tuples': <class 'int'>, 'hnsw.scan_mem_multiplier': <class 'float'>, 'ivfflat.probes': <class 'int'>, 'ivfflat.iterative_scan': <class 'str'>, 'ivfflat.max_probes': <class 'int'>}#: The search settings, with the type each takes.

agensgraph.vector.TYPES = ('vector', 'halfvec', 'sparsevec')#: pgvector’s types the driver reads.