langchain_community.vectorstores.annoy.Annoy¶

class langchain_community.vectorstores.annoy.Annoy(embedding_function: Callable, index: Any, metric: str, docstore: Docstore, index_to_docstore_id: Dict[int, str])[source]¶

Annoy vector store.

To use, you should have the annoy python package installed.

Example

from langchain_community.vectorstores import Annoy
db = Annoy(embedding_function, index, docstore, index_to_docstore_id)

Initialize with necessary components.

Attributes

embeddings

Access the query embedding object if available.

Methods

__init__(embedding_function, index, metric, ...)

Initialize with necessary components.

aadd_documents(documents, **kwargs)

Run more documents through the embeddings and add to the vectorstore.

aadd_texts(texts[, metadatas])

Run more texts through the embeddings and add to the vectorstore.

add_documents(documents, **kwargs)

Run more documents through the embeddings and add to the vectorstore.

add_texts(texts[, metadatas])

Run more texts through the embeddings and add to the vectorstore.

adelete([ids])

Delete by vector ID or other criteria.

afrom_documents(documents, embedding, **kwargs)

Return VectorStore initialized from documents and embeddings.

afrom_texts(texts, embedding[, metadatas])

Return VectorStore initialized from texts and embeddings.

amax_marginal_relevance_search(query[, k, ...])

Return docs selected using the maximal marginal relevance.

amax_marginal_relevance_search_by_vector(...)

Return docs selected using the maximal marginal relevance.

as_retriever(**kwargs)

Return VectorStoreRetriever initialized from this VectorStore.

asearch(query, search_type, **kwargs)

Return docs most similar to query using specified search type.

asimilarity_search(query[, k])

Return docs most similar to query.

asimilarity_search_by_vector(embedding[, k])

Return docs most similar to embedding vector.

asimilarity_search_with_relevance_scores(query)

Return docs and relevance scores in the range [0, 1], asynchronously.

asimilarity_search_with_score(*args, **kwargs)

Run similarity search with distance asynchronously.

delete([ids])

Delete by vector ID or other criteria.

from_documents(documents, embedding, **kwargs)

Return VectorStore initialized from documents and embeddings.

from_embeddings(text_embeddings, embedding)

Construct Annoy wrapper from embeddings.

from_texts(texts, embedding[, metadatas, ...])

Construct Annoy wrapper from raw documents.

load_local(folder_path, embeddings, *[, ...])

Load Annoy index, docstore, and index_to_docstore_id to disk.

max_marginal_relevance_search(query[, k, ...])

Return docs selected using the maximal marginal relevance.

max_marginal_relevance_search_by_vector(...)

Return docs selected using the maximal marginal relevance.

process_index_results(idxs, dists)

Turns annoy results into a list of documents and scores.

save_local(folder_path[, prefault])

Save Annoy index, docstore, and index_to_docstore_id to disk.

search(query, search_type, **kwargs)

Return docs most similar to query using specified search type.

similarity_search(query[, k, search_k])

Return docs most similar to query.

similarity_search_by_index(docstore_index[, ...])

Return docs most similar to docstore_index.

similarity_search_by_vector(embedding[, k, ...])

Return docs most similar to embedding vector.

similarity_search_with_relevance_scores(query)

Return docs and relevance scores in the range [0, 1].

similarity_search_with_score(query[, k, ...])

Return docs most similar to query.

similarity_search_with_score_by_index(...[, ...])

Return docs most similar to query.

similarity_search_with_score_by_vector(embedding)

Return docs most similar to query.

Parameters
  • embedding_function (Callable) –

  • index (Any) –

  • metric (str) –

  • docstore (Docstore) –

  • index_to_docstore_id (Dict[int, str]) –

__init__(embedding_function: Callable, index: Any, metric: str, docstore: Docstore, index_to_docstore_id: Dict[int, str])[source]¶

Initialize with necessary components.

Parameters
  • embedding_function (Callable) –

  • index (Any) –

  • metric (str) –

  • docstore (Docstore) –

  • index_to_docstore_id (Dict[int, str]) –

async aadd_documents(documents: List[Document], **kwargs: Any) List[str]¶

Run more documents through the embeddings and add to the vectorstore.

Parameters
  • (List[Document] (documents) – Documents to add to the vectorstore.

  • documents (List[Document]) –

  • kwargs (Any) –

Returns

List of IDs of the added texts.

Return type

List[str]

async aadd_texts(texts: Iterable[str], metadatas: Optional[List[dict]] = None, **kwargs: Any) List[str]¶

Run more texts through the embeddings and add to the vectorstore.

Parameters
  • texts (Iterable[str]) –

  • metadatas (Optional[List[dict]]) –

  • kwargs (Any) –

Return type

List[str]

add_documents(documents: List[Document], **kwargs: Any) List[str]¶

Run more documents through the embeddings and add to the vectorstore.

Parameters
  • (List[Document] (documents) – Documents to add to the vectorstore.

  • documents (List[Document]) –

  • kwargs (Any) –

Returns

List of IDs of the added texts.

Return type

List[str]

add_texts(texts: Iterable[str], metadatas: Optional[List[dict]] = None, **kwargs: Any) List[str][source]¶

Run more texts through the embeddings and add to the vectorstore.

Parameters
  • texts (Iterable[str]) – Iterable of strings to add to the vectorstore.

  • metadatas (Optional[List[dict]]) – Optional list of metadatas associated with the texts.

  • kwargs (Any) – vectorstore specific parameters

Returns

List of ids from adding the texts into the vectorstore.

Return type

List[str]

async adelete(ids: Optional[List[str]] = None, **kwargs: Any) Optional[bool]¶

Delete by vector ID or other criteria.

Parameters
  • ids (Optional[List[str]]) – List of ids to delete.

  • **kwargs (Any) – Other keyword arguments that subclasses might use.

Returns

True if deletion is successful, False otherwise, None if not implemented.

Return type

Optional[bool]

async classmethod afrom_documents(documents: List[Document], embedding: Embeddings, **kwargs: Any) VST¶

Return VectorStore initialized from documents and embeddings.

Parameters
  • documents (List[Document]) –

  • embedding (Embeddings) –

  • kwargs (Any) –

Return type

VST

async classmethod afrom_texts(texts: List[str], embedding: Embeddings, metadatas: Optional[List[dict]] = None, **kwargs: Any) VST¶

Return VectorStore initialized from texts and embeddings.

Parameters
  • texts (List[str]) –

  • embedding (Embeddings) –

  • metadatas (Optional[List[dict]]) –

  • kwargs (Any) –

Return type

VST

Return docs selected using the maximal marginal relevance.

Parameters
  • query (str) –

  • k (int) –

  • fetch_k (int) –

  • lambda_mult (float) –

  • kwargs (Any) –

Return type

List[Document]

async amax_marginal_relevance_search_by_vector(embedding: List[float], k: int = 4, fetch_k: int = 20, lambda_mult: float = 0.5, **kwargs: Any) List[Document]¶

Return docs selected using the maximal marginal relevance.

Parameters
  • embedding (List[float]) –

  • k (int) –

  • fetch_k (int) –

  • lambda_mult (float) –

  • kwargs (Any) –

Return type

List[Document]

as_retriever(**kwargs: Any) VectorStoreRetriever¶

Return VectorStoreRetriever initialized from this VectorStore.

Parameters
  • search_type (Optional[str]) – Defines the type of search that the Retriever should perform. Can be “similarity” (default), “mmr”, or “similarity_score_threshold”.

  • search_kwargs (Optional[Dict]) –

    Keyword arguments to pass to the search function. Can include things like:

    k: Amount of documents to return (Default: 4) score_threshold: Minimum relevance threshold

    for similarity_score_threshold

    fetch_k: Amount of documents to pass to MMR algorithm (Default: 20) lambda_mult: Diversity of results returned by MMR;

    1 for minimum diversity and 0 for maximum. (Default: 0.5)

    filter: Filter by document metadata

  • kwargs (Any) –

Returns

Retriever class for VectorStore.

Return type

VectorStoreRetriever

Examples:

# Retrieve more documents with higher diversity
# Useful if your dataset has many similar documents
docsearch.as_retriever(
    search_type="mmr",
    search_kwargs={'k': 6, 'lambda_mult': 0.25}
)

# Fetch more documents for the MMR algorithm to consider
# But only return the top 5
docsearch.as_retriever(
    search_type="mmr",
    search_kwargs={'k': 5, 'fetch_k': 50}
)

# Only retrieve documents that have a relevance score
# Above a certain threshold
docsearch.as_retriever(
    search_type="similarity_score_threshold",
    search_kwargs={'score_threshold': 0.8}
)

# Only get the single most similar document from the dataset
docsearch.as_retriever(search_kwargs={'k': 1})

# Use a filter to only retrieve documents from a specific paper
docsearch.as_retriever(
    search_kwargs={'filter': {'paper_title':'GPT-4 Technical Report'}}
)
async asearch(query: str, search_type: str, **kwargs: Any) List[Document]¶

Return docs most similar to query using specified search type.

Parameters
  • query (str) –

  • search_type (str) –

  • kwargs (Any) –

Return type

List[Document]

Return docs most similar to query.

Parameters
  • query (str) –

  • k (int) –

  • kwargs (Any) –

Return type

List[Document]

async asimilarity_search_by_vector(embedding: List[float], k: int = 4, **kwargs: Any) List[Document]¶

Return docs most similar to embedding vector.

Parameters
  • embedding (List[float]) –

  • k (int) –

  • kwargs (Any) –

Return type

List[Document]

async asimilarity_search_with_relevance_scores(query: str, k: int = 4, **kwargs: Any) List[Tuple[Document, float]]¶

Return docs and relevance scores in the range [0, 1], asynchronously.

0 is dissimilar, 1 is most similar.

Parameters
  • query (str) – input text

  • k (int) – Number of Documents to return. Defaults to 4.

  • **kwargs (Any) –

    kwargs to be passed to similarity search. Should include: score_threshold: Optional, a floating point value between 0 to 1 to

    filter the resulting set of retrieved docs

Returns

List of Tuples of (doc, similarity_score)

Return type

List[Tuple[Document, float]]

async asimilarity_search_with_score(*args: Any, **kwargs: Any) List[Tuple[Document, float]]¶

Run similarity search with distance asynchronously.

Parameters
  • args (Any) –

  • kwargs (Any) –

Return type

List[Tuple[Document, float]]

delete(ids: Optional[List[str]] = None, **kwargs: Any) Optional[bool]¶

Delete by vector ID or other criteria.

Parameters
  • ids (Optional[List[str]]) – List of ids to delete.

  • **kwargs (Any) – Other keyword arguments that subclasses might use.

Returns

True if deletion is successful, False otherwise, None if not implemented.

Return type

Optional[bool]

classmethod from_documents(documents: List[Document], embedding: Embeddings, **kwargs: Any) VST¶

Return VectorStore initialized from documents and embeddings.

Parameters
  • documents (List[Document]) –

  • embedding (Embeddings) –

  • kwargs (Any) –

Return type

VST

classmethod from_embeddings(text_embeddings: List[Tuple[str, List[float]]], embedding: Embeddings, metadatas: Optional[List[dict]] = None, metric: str = 'angular', trees: int = 100, n_jobs: int = -1, **kwargs: Any) Annoy[source]¶

Construct Annoy wrapper from embeddings.

Parameters
  • text_embeddings (List[Tuple[str, List[float]]]) – List of tuples of (text, embedding)

  • embedding (Embeddings) – Embedding function to use.

  • metadatas (Optional[List[dict]]) – List of metadata dictionaries to associate with documents.

  • metric (str) – Metric to use for indexing. Defaults to “angular”.

  • trees (int) – Number of trees to use for indexing. Defaults to 100.

  • n_jobs (int) – Number of jobs to use for indexing. Defaults to -1

  • kwargs (Any) –

Return type

Annoy

This is a user friendly interface that:
  1. Creates an in memory docstore with provided embeddings

  2. Initializes the Annoy database

This is intended to be a quick way to get started.

Example

from langchain_community.vectorstores import Annoy
from langchain_community.embeddings import OpenAIEmbeddings
embeddings = OpenAIEmbeddings()
text_embeddings = embeddings.embed_documents(texts)
text_embedding_pairs = list(zip(texts, text_embeddings))
db = Annoy.from_embeddings(text_embedding_pairs, embeddings)
classmethod from_texts(texts: List[str], embedding: Embeddings, metadatas: Optional[List[dict]] = None, metric: str = 'angular', trees: int = 100, n_jobs: int = -1, **kwargs: Any) Annoy[source]¶

Construct Annoy wrapper from raw documents.

Parameters
  • texts (List[str]) – List of documents to index.

  • embedding (Embeddings) – Embedding function to use.

  • metadatas (Optional[List[dict]]) – List of metadata dictionaries to associate with documents.

  • metric (str) – Metric to use for indexing. Defaults to “angular”.

  • trees (int) – Number of trees to use for indexing. Defaults to 100.

  • n_jobs (int) – Number of jobs to use for indexing. Defaults to -1.

  • kwargs (Any) –

Return type

Annoy

This is a user friendly interface that:
  1. Embeds documents.

  2. Creates an in memory docstore

  3. Initializes the Annoy database

This is intended to be a quick way to get started.

Example

from langchain_community.vectorstores import Annoy
from langchain_community.embeddings import OpenAIEmbeddings
embeddings = OpenAIEmbeddings()
index = Annoy.from_texts(texts, embeddings)
classmethod load_local(folder_path: str, embeddings: Embeddings, *, allow_dangerous_deserialization: bool = False) Annoy[source]¶

Load Annoy index, docstore, and index_to_docstore_id to disk.

Parameters
  • folder_path (str) – folder path to load index, docstore, and index_to_docstore_id from.

  • embeddings (Embeddings) – Embeddings to use when generating queries.

  • allow_dangerous_deserialization (bool) – whether to allow deserialization of the data which involves loading a pickle file. Pickle files can be modified by malicious actors to deliver a malicious payload that results in execution of arbitrary code on your machine.

Return type

Annoy

Return docs selected using the maximal marginal relevance.

Maximal marginal relevance optimizes for similarity to query AND diversity among selected documents.

Parameters
  • query (str) – Text to look up documents similar to.

  • k (int) – Number of Documents to return. Defaults to 4.

  • fetch_k (int) – Number of Documents to fetch to pass to MMR algorithm.

  • lambda_mult (float) – Number between 0 and 1 that determines the degree of diversity among the results with 0 corresponding to maximum diversity and 1 to minimum diversity. Defaults to 0.5.

  • kwargs (Any) –

Returns

List of Documents selected by maximal marginal relevance.

Return type

List[Document]

max_marginal_relevance_search_by_vector(embedding: List[float], k: int = 4, fetch_k: int = 20, lambda_mult: float = 0.5, **kwargs: Any) List[Document][source]¶

Return docs selected using the maximal marginal relevance.

Maximal marginal relevance optimizes for similarity to query AND diversity among selected documents.

Parameters
  • embedding (List[float]) – Embedding to look up documents similar to.

  • fetch_k (int) – Number of Documents to fetch to pass to MMR algorithm.

  • k (int) – Number of Documents to return. Defaults to 4.

  • lambda_mult (float) – Number between 0 and 1 that determines the degree of diversity among the results with 0 corresponding to maximum diversity and 1 to minimum diversity. Defaults to 0.5.

  • kwargs (Any) –

Returns

List of Documents selected by maximal marginal relevance.

Return type

List[Document]

process_index_results(idxs: List[int], dists: List[float]) List[Tuple[Document, float]][source]¶

Turns annoy results into a list of documents and scores.

Parameters
  • idxs (List[int]) – List of indices of the documents in the index.

  • dists (List[float]) – List of distances of the documents in the index.

Returns

List of Documents and scores.

Return type

List[Tuple[Document, float]]

save_local(folder_path: str, prefault: bool = False) None[source]¶

Save Annoy index, docstore, and index_to_docstore_id to disk.

Parameters
  • folder_path (str) – folder path to save index, docstore, and index_to_docstore_id to.

  • prefault (bool) – Whether to pre-load the index into memory.

Return type

None

search(query: str, search_type: str, **kwargs: Any) List[Document]¶

Return docs most similar to query using specified search type.

Parameters
  • query (str) –

  • search_type (str) –

  • kwargs (Any) –

Return type

List[Document]

Return docs most similar to query.

Parameters
  • query (str) – Text to look up documents similar to.

  • k (int) – Number of Documents to return. Defaults to 4.

  • search_k (int) – inspect up to search_k nodes which defaults to n_trees * n if not provided

  • kwargs (Any) –

Returns

List of Documents most similar to the query.

Return type

List[Document]

similarity_search_by_index(docstore_index: int, k: int = 4, search_k: int = -1, **kwargs: Any) List[Document][source]¶

Return docs most similar to docstore_index.

Parameters
  • docstore_index (int) – Index of document in docstore

  • k (int) – Number of Documents to return. Defaults to 4.

  • search_k (int) – inspect up to search_k nodes which defaults to n_trees * n if not provided

  • kwargs (Any) –

Returns

List of Documents most similar to the embedding.

Return type

List[Document]

similarity_search_by_vector(embedding: List[float], k: int = 4, search_k: int = -1, **kwargs: Any) List[Document][source]¶

Return docs most similar to embedding vector.

Parameters
  • embedding (List[float]) – Embedding to look up documents similar to.

  • k (int) – Number of Documents to return. Defaults to 4.

  • search_k (int) – inspect up to search_k nodes which defaults to n_trees * n if not provided

  • kwargs (Any) –

Returns

List of Documents most similar to the embedding.

Return type

List[Document]

similarity_search_with_relevance_scores(query: str, k: int = 4, **kwargs: Any) List[Tuple[Document, float]]¶

Return docs and relevance scores in the range [0, 1].

0 is dissimilar, 1 is most similar.

Parameters
  • query (str) – input text

  • k (int) – Number of Documents to return. Defaults to 4.

  • **kwargs (Any) –

    kwargs to be passed to similarity search. Should include: score_threshold: Optional, a floating point value between 0 to 1 to

    filter the resulting set of retrieved docs

Returns

List of Tuples of (doc, similarity_score)

Return type

List[Tuple[Document, float]]

similarity_search_with_score(query: str, k: int = 4, search_k: int = -1) List[Tuple[Document, float]][source]¶

Return docs most similar to query.

Parameters
  • query (str) – Text to look up documents similar to.

  • k (int) – Number of Documents to return. Defaults to 4.

  • search_k (int) – inspect up to search_k nodes which defaults to n_trees * n if not provided

Returns

List of Documents most similar to the query and score for each

Return type

List[Tuple[Document, float]]

similarity_search_with_score_by_index(docstore_index: int, k: int = 4, search_k: int = -1) List[Tuple[Document, float]][source]¶

Return docs most similar to query.

Parameters
  • query – Text to look up documents similar to.

  • k (int) – Number of Documents to return. Defaults to 4.

  • search_k (int) – inspect up to search_k nodes which defaults to n_trees * n if not provided

  • docstore_index (int) –

Returns

List of Documents most similar to the query and score for each

Return type

List[Tuple[Document, float]]

similarity_search_with_score_by_vector(embedding: List[float], k: int = 4, search_k: int = -1) List[Tuple[Document, float]][source]¶

Return docs most similar to query.

Parameters
  • query – Text to look up documents similar to.

  • k (int) – Number of Documents to return. Defaults to 4.

  • search_k (int) – inspect up to search_k nodes which defaults to n_trees * n if not provided

  • embedding (List[float]) –

Returns

List of Documents most similar to the query and score for each

Return type

List[Tuple[Document, float]]

Examples using Annoy¶