github/langchain
1
0
Fork 0
mirror of https://github.com/hwchase17/langchain.git synced 2025-09-13 21:47:12 +00:00
Code Issues Packages Projects Releases Wiki Activity

community[patch]: add hybrid search to singlestoredb vectorstore (#20793)

Browse Source 
Implemented the ability to enable full-text search within the
SingleStore vector store, offering users a versatile range of search
strategies. This enhancement allows users to seamlessly combine
full-text search with vector search, enabling the following search
strategies:

* Search solely by vector similarity.
* Conduct searches exclusively based on text similarity, utilizing
Lucene internally.
* Filter search results by text similarity score, with the option to
specify a threshold, followed by a search based on vector similarity.
* Filter results by vector similarity score before conducting a search
based on text similarity.
* Perform searches using a weighted sum of vector and text similarity
scores.

Additionally, integration tests have been added to comprehensively cover
all scenarios.
Updated notebook with examples.

CC: @baskaryan, @hwchase17

---------

Co-authored-by: Volodymyr Tkachuk <vtkachuk-ua@singlestore.com>
Co-authored-by: Bagatur <baskaryan@gmail.com>
This commit is contained in:
volodymyr-memsql
2024-04-25 00:34:50 +03:00
committed by GitHub
parent 9efab3ed66
commit 493afe4d8d
3 changed files with 879 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files

531
libs/community/langchain_community/vectorstores/singlestoredb.py
View File

@@ -2,6 +2,7 @@ from __future__ import annotations
import json
import re
from enum import Enum
from typing import (
Any,
Callable,
@@ -38,6 +39,15 @@ class SingleStoreDB(VectorStore):
optionally, the names of the table and the fields to use.
"""
class SearchStrategy(str, Enum):
"""Enumerator of the Search strategies for searching in the vectorstore."""
VECTOR_ONLY = "VECTOR_ONLY"
TEXT_ONLY = "TEXT_ONLY"
FILTER_BY_TEXT = "FILTER_BY_TEXT"
FILTER_BY_VECTOR = "FILTER_BY_VECTOR"
WEIGHTED_SUM = "WEIGHTED_SUM"
def _get_connection(self: SingleStoreDB) -> Any:
try:
import singlestoredb as s2
@@ -57,10 +67,12 @@ class SingleStoreDB(VectorStore):
content_field: str = "content",
metadata_field: str = "metadata",
vector_field: str = "vector",
id_field: str = "id",
use_vector_index: bool = False,
vector_index_name: str = "",
vector_index_options: Optional[dict] = None,
vector_size: int = 1536,
use_full_text_search: bool = False,
pool_size: int = 5,
max_overflow: int = 10,
timeout: float = 30,
@@ -81,7 +93,8 @@ class SingleStoreDB(VectorStore):
- EUCLIDEAN_DISTANCE: Computes the Euclidean distance between
two vectors. This metric considers the geometric distance in
the vector space, and might be more suitable for embeddings
that rely on spatial relationships.
that rely on spatial relationships. This metric is not
compatible with the WEIGHTED_SUM search strategy.
table_name (str, optional): Specifies the name of the table in use.
Defaults to "embeddings".
@@ -91,6 +104,8 @@ class SingleStoreDB(VectorStore):
Defaults to "metadata".
vector_field (str, optional): Specifies the field to store the vector.
Defaults to "vector".
id_field (str, optional): Specifies the field to store the id.
Defaults to "id".
use_vector_index (bool, optional): Toggles the use of a vector index.
Works only with SingleStoreDB 8.5 or later. Defaults to False.
@@ -113,6 +128,14 @@ class SingleStoreDB(VectorStore):
Should be set to the same value as the size of the vectors
stored in the vector_field.
use_full_text_search (bool, optional): Toggles the use a full-text index
on the document content. Defaults to False. If set to True, the table
will be created with a full-text index on the content field,
and the simularity_search method will all using TEXT_ONLY,
FILTER_BY_TEXT, FILTER_BY_VECTOR, and WIGHTED_SUM search strategies.
If set to False, the simularity_search method will only allow
VECTOR_ONLY search strategy.
Following arguments pertain to the connection pool:
pool_size (int, optional): Determines the number of active connections in
@@ -165,7 +188,7 @@ class SingleStoreDB(VectorStore):
.. code-block:: python
from langchain_community.embeddings import OpenAIEmbeddings
from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import SingleStoreDB
vectorstore = SingleStoreDB(
@@ -177,7 +200,7 @@ class SingleStoreDB(VectorStore):
.. code-block:: python
from langchain_community.embeddings import OpenAIEmbeddings
from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import SingleStoreDB
vectorstore = SingleStoreDB(
@@ -197,7 +220,7 @@ class SingleStoreDB(VectorStore):
.. code-block:: python
from langchain_community.embeddings import OpenAIEmbeddings
from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import SingleStoreDB
os.environ['SINGLESTOREDB_URL'] = 'me:p455w0rd@s2-host.com/my_db'
@@ -207,7 +230,7 @@ class SingleStoreDB(VectorStore):
.. code-block:: python
from langchain_community.embeddings import OpenAIEmbeddings
from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import SingleStoreDB
os.environ['SINGLESTOREDB_URL'] = 'me:p455w0rd@s2-host.com/my_db'
@@ -215,6 +238,18 @@ class SingleStoreDB(VectorStore):
OpenAIEmbeddings(),
use_vector_index=True,
)
Using full-text index:
.. code-block:: python
from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import SingleStoreDB
os.environ['SINGLESTOREDB_URL'] = 'me:p455w0rd@s2-host.com/my_db'
vectorstore = SingleStoreDB(
OpenAIEmbeddings(),
use_full_text_search=True,
)
"""
self.embedding = embedding
@@ -223,6 +258,7 @@ class SingleStoreDB(VectorStore):
self.content_field = self._sanitize_input(content_field)
self.metadata_field = self._sanitize_input(metadata_field)
self.vector_field = self._sanitize_input(vector_field)
self.id_field = self._sanitize_input(id_field)
self.use_vector_index = bool(use_vector_index)
self.vector_index_name = self._sanitize_input(vector_index_name)
@@ -230,6 +266,8 @@ class SingleStoreDB(VectorStore):
self.vector_index_options["metric_type"] = self.distance_strategy
self.vector_size = int(vector_size)
self.use_full_text_search = bool(use_full_text_search)
# Pass the rest of the kwargs to the connection.
self.connection_kwargs = kwargs
@@ -238,7 +276,7 @@ class SingleStoreDB(VectorStore):
self.connection_kwargs["conn_attrs"] = dict()
self.connection_kwargs["conn_attrs"]["_connector_name"] = "langchain python sdk"
self.connection_kwargs["conn_attrs"]["_connector_version"] = "1.0.2"
self.connection_kwargs["conn_attrs"]["_connector_version"] = "2.0.0"
# Create connection pool.
self.connection_pool = QueuePool(
@@ -266,6 +304,9 @@ class SingleStoreDB(VectorStore):
try:
cur = conn.cursor()
try:
full_text_index = ""
if self.use_full_text_search:
full_text_index = ", FULLTEXT({})".format(self.content_field)
if self.use_vector_index:
index_options = ""
if self.vector_index_options and len(self.vector_index_options) > 0:
@@ -274,10 +315,11 @@ class SingleStoreDB(VectorStore):
)
cur.execute(
"""CREATE TABLE IF NOT EXISTS {}
({} TEXT CHARACTER SET utf8mb4 COLLATE utf8mb4_general_ci,
{} VECTOR({}, F32) NOT NULL, {} JSON,
VECTOR INDEX {} ({}) {});""".format(
({} BIGINT AUTO_INCREMENT PRIMARY KEY, {} LONGTEXT CHARACTER
SET utf8mb4 COLLATE utf8mb4_general_ci, {} VECTOR({}, F32)
NOT NULL, {} JSON, VECTOR INDEX {} ({}) {}{});""".format(
self.table_name,
self.id_field,
self.content_field,
self.vector_field,
self.vector_size,
@@ -285,17 +327,21 @@ class SingleStoreDB(VectorStore):
self.vector_index_name,
self.vector_field,
index_options,
full_text_index,
),
)
else:
cur.execute(
"""CREATE TABLE IF NOT EXISTS {}
({} TEXT CHARACTER SET utf8mb4 COLLATE utf8mb4_general_ci,
{} BLOB, {} JSON);""".format(
({} BIGINT AUTO_INCREMENT PRIMARY KEY, {} LONGTEXT CHARACTER
SET utf8mb4 COLLATE utf8mb4_general_ci, {} BLOB, {} JSON{});
""".format(
self.table_name,
self.id_field,
self.content_field,
self.vector_field,
self.metadata_field,
full_text_index,
),
)
finally:
@@ -365,8 +411,12 @@ class SingleStoreDB(VectorStore):
else self.embedding.embed_documents([text])[0]
)
cur.execute(
"INSERT INTO {} VALUES (%s, JSON_ARRAY_PACK(%s), %s)".format(
self.table_name
"""INSERT INTO {}({}, {}, {})
VALUES (%s, JSON_ARRAY_PACK(%s), %s)""".format(
self.table_name,
self.content_field,
self.vector_field,
self.metadata_field,
),
(
text,
@@ -374,7 +424,7 @@ class SingleStoreDB(VectorStore):
json.dumps(metadata),
),
)
if self.use_vector_index:
if self.use_vector_index or self.use_full_text_search:
cur.execute("OPTIMIZE TABLE {} FLUSH;".format(self.table_name))
finally:
cur.close()
@@ -383,7 +433,16 @@ class SingleStoreDB(VectorStore):
return []
def similarity_search(
self, query: str, k: int = 4, filter: Optional[dict] = None, **kwargs: Any
self,
query: str,
k: int = 4,
filter: Optional[dict] = None,
search_strategy: SearchStrategy = SearchStrategy.VECTOR_ONLY,
filter_threshold: float = 0,
text_weight: float = 0.5,
vector_weight: float = 0.5,
vector_select_count_multiplier: int = 10,
**kwargs: Any,
) -> List[Document]:
"""Returns the most similar indexed documents to the query text.
@@ -393,29 +452,119 @@ class SingleStoreDB(VectorStore):
query (str): The query text for which to find similar documents.
k (int): The number of documents to return. Default is 4.
filter (dict): A dictionary of metadata fields and values to filter by.
Default is None.
search_strategy (SearchStrategy): The search strategy to use.
Default is SearchStrategy.VECTOR_ONLY.
Available options are:
- SearchStrategy.VECTOR_ONLY: Searches only by vector similarity.
- SearchStrategy.TEXT_ONLY: Searches only by text similarity. This
option is only available if use_full_text_search is True.
- SearchStrategy.FILTER_BY_TEXT: Filters by text similarity and
searches by vector similarity. This option is only available if
use_full_text_search is True.
- SearchStrategy.FILTER_BY_VECTOR: Filters by vector similarity and
searches by text similarity. This option is only available if
use_full_text_search is True.
- SearchStrategy.WEIGHTED_SUM: Searches by a weighted sum of text and
vector similarity. This option is only available if
use_full_text_search is True and distance_strategy is DOT_PRODUCT.
filter_threshold (float): The threshold for filtering by text or vector
similarity. Default is 0. This option has effect only if search_strategy
is SearchStrategy.FILTER_BY_TEXT or SearchStrategy.FILTER_BY_VECTOR.
text_weight (float): The weight of text similarity in the weighted sum
search strategy. Default is 0.5. This option has effect only if
search_strategy is SearchStrategy.WEIGHTED_SUM.
vector_weight (float): The weight of vector similarity in the weighted sum
search strategy. Default is 0.5. This option has effect only if
search_strategy is SearchStrategy.WEIGHTED_SUM.
vector_select_count_multiplier (int): The multiplier for the number of
vectors to select when using the vector index. Default is 10.
This parameter has effect only if use_vector_index is True and
search_strategy is SearchStrategy.WEIGHTED_SUM or
SearchStrategy.FILTER_BY_TEXT.
The number of vectors selected will
be k * vector_select_count_multiplier.
This is needed due to the limitations of the vector index.
Returns:
List[Document]: A list of documents that are most similar to the query text.
Examples:
Basic Usage:
.. code-block:: python
from langchain_community.vectorstores import SingleStoreDB
from langchain_community.embeddings import OpenAIEmbeddings
from langchain_openai import OpenAIEmbeddings
s2 = SingleStoreDB.from_documents(
docs,
OpenAIEmbeddings(),
host="username:password@localhost:3306/database"
)
s2.similarity_search("query text", 1,
{"metadata_field": "metadata_value"})
results = s2.similarity_search("query text", 1,
{"metadata_field": "metadata_value"})
Different Search Strategies:
.. code-block:: python
from langchain_community.vectorstores import SingleStoreDB
from langchain_openai import OpenAIEmbeddings
s2 = SingleStoreDB.from_documents(
docs,
OpenAIEmbeddings(),
host="username:password@localhost:3306/database",
use_full_text_search=True,
use_vector_index=True,
)
results = s2.similarity_search("query text", 1,
search_strategy=SingleStoreDB.SearchStrategy.FILTER_BY_TEXT,
filter_threshold=0.5)
Weighted Sum Search Strategy:
.. code-block:: python
from langchain_community.vectorstores import SingleStoreDB
from langchain_openai import OpenAIEmbeddings
s2 = SingleStoreDB.from_documents(
docs,
OpenAIEmbeddings(),
host="username:password@localhost:3306/database",
use_full_text_search=True,
use_vector_index=True,
)
results = s2.similarity_search("query text", 1,
search_strategy=SingleStoreDB.SearchStrategy.WEIGHTED_SUM,
text_weight=0.3,
vector_weight=0.7)
"""
docs_and_scores = self.similarity_search_with_score(
query=query, k=k, filter=filter
query=query,
k=k,
filter=filter,
search_strategy=search_strategy,
filter_threshold=filter_threshold,
text_weight=text_weight,
vector_weight=vector_weight,
vector_select_count_multiplier=vector_select_count_multiplier,
**kwargs,
)
return [doc for doc, _ in docs_and_scores]
def similarity_search_with_score(
self, query: str, k: int = 4, filter: Optional[dict] = None
self,
query: str,
k: int = 4,
filter: Optional[dict] = None,
search_strategy: SearchStrategy = SearchStrategy.VECTOR_ONLY,
filter_threshold: float = 1,
text_weight: float = 0.5,
vector_weight: float = 0.5,
vector_select_count_multiplier: int = 10,
**kwargs: Any,
) -> List[Tuple[Document, float]]:
"""Return docs most similar to query. Uses cosine similarity.
@@ -424,20 +573,156 @@ class SingleStoreDB(VectorStore):
k: Number of Documents to return. Defaults to 4.
filter: A dictionary of metadata fields and values to filter by.
Defaults to None.
search_strategy (SearchStrategy): The search strategy to use.
Default is SearchStrategy.VECTOR_ONLY.
Available options are:
- SearchStrategy.VECTOR_ONLY: Searches only by vector similarity.
- SearchStrategy.TEXT_ONLY: Searches only by text similarity. This
option is only available if use_full_text_search is True.
- SearchStrategy.FILTER_BY_TEXT: Filters by text similarity and
searches by vector similarity. This option is only available if
use_full_text_search is True.
- SearchStrategy.FILTER_BY_VECTOR: Filters by vector similarity and
searches by text similarity. This option is only available if
use_full_text_search is True.
- SearchStrategy.WEIGHTED_SUM: Searches by a weighted sum of text and
vector similarity. This option is only available if
use_full_text_search is True and distance_strategy is DOT_PRODUCT.
filter_threshold (float): The threshold for filtering by text or vector
similarity. Default is 0. This option has effect only if search_strategy
is SearchStrategy.FILTER_BY_TEXT or SearchStrategy.FILTER_BY_VECTOR.
text_weight (float): The weight of text similarity in the weighted sum
search strategy. Default is 0.5. This option has effect only if
search_strategy is SearchStrategy.WEIGHTED_SUM.
vector_weight (float): The weight of vector similarity in the weighted sum
search strategy. Default is 0.5. This option has effect only if
search_strategy is SearchStrategy.WEIGHTED_SUM.
vector_select_count_multiplier (int): The multiplier for the number of
vectors to select when using the vector index. Default is 10.
This parameter has effect only if use_vector_index is True and
search_strategy is SearchStrategy.WEIGHTED_SUM or
SearchStrategy.FILTER_BY_TEXT.
The number of vectors selected will
be k * vector_select_count_multiplier.
This is needed due to the limitations of the vector index.
Returns:
List of Documents most similar to the query and score for each
document.
Raises:
ValueError: If the search strategy is not supported with the
distance strategy.
Examples:
Basic Usage:
.. code-block:: python
from langchain_community.vectorstores import SingleStoreDB
from langchain_openai import OpenAIEmbeddings
s2 = SingleStoreDB.from_documents(
docs,
OpenAIEmbeddings(),
host="username:password@localhost:3306/database"
)
results = s2.similarity_search_with_score("query text", 1,
{"metadata_field": "metadata_value"})
Different Search Strategies:
.. code-block:: python
from langchain_community.vectorstores import SingleStoreDB
from langchain_openai import OpenAIEmbeddings
s2 = SingleStoreDB.from_documents(
docs,
OpenAIEmbeddings(),
host="username:password@localhost:3306/database",
use_full_text_search=True,
use_vector_index=True,
)
results = s2.similarity_search_with_score("query text", 1,
search_strategy=SingleStoreDB.SearchStrategy.FILTER_BY_VECTOR,
filter_threshold=0.5)
Weighted Sum Search Strategy:
.. code-block:: python
from langchain_community.vectorstores import SingleStoreDB
from langchain_openai import OpenAIEmbeddings
s2 = SingleStoreDB.from_documents(
docs,
OpenAIEmbeddings(),
host="username:password@localhost:3306/database",
use_full_text_search=True,
use_vector_index=True,
)
results = s2.similarity_search_with_score("query text", 1,
search_strategy=SingleStoreDB.SearchStrategy.WEIGHTED_SUM,
text_weight=0.3,
vector_weight=0.7)
"""
if (
search_strategy != SingleStoreDB.SearchStrategy.VECTOR_ONLY
and not self.use_full_text_search
):
raise ValueError(
"""Search strategy {} is not supported
when use_full_text_search is False""".format(search_strategy)
)
if (
search_strategy == SingleStoreDB.SearchStrategy.WEIGHTED_SUM
and self.distance_strategy != DistanceStrategy.DOT_PRODUCT
):
raise ValueError(
"Search strategy {} is not supported with distance strategy {}".format(
search_strategy, self.distance_strategy
)
)
# Creates embedding vector from user query
embedding = self.embedding.embed_query(query)
embedding = []
if search_strategy != SingleStoreDB.SearchStrategy.TEXT_ONLY:
embedding = self.embedding.embed_query(query)
self.embedding.embed_query(query)
conn = self.connection_pool.connect()
result = []
where_clause: str = ""
where_clause_values: List[Any] = []
if filter:
if filter or search_strategy in [
SingleStoreDB.SearchStrategy.FILTER_BY_TEXT,
SingleStoreDB.SearchStrategy.FILTER_BY_VECTOR,
]:
where_clause = "WHERE "
arguments = []
if search_strategy == SingleStoreDB.SearchStrategy.FILTER_BY_TEXT:
arguments.append(
"MATCH ({}) AGAINST (%s) > %s".format(self.content_field)
)
where_clause_values.append(query)
where_clause_values.append(float(filter_threshold))
if search_strategy == SingleStoreDB.SearchStrategy.FILTER_BY_VECTOR:
condition = "{}({}, JSON_ARRAY_PACK(%s)) ".format(
self.distance_strategy.name
if isinstance(self.distance_strategy, DistanceStrategy)
else self.distance_strategy,
self.vector_field,
)
if self.distance_strategy == DistanceStrategy.EUCLIDEAN_DISTANCE:
condition += "< %s"
else:
condition += "> %s"
arguments.append(condition)
where_clause_values.append("[{}]".format(",".join(map(str, embedding))))
where_clause_values.append(float(filter_threshold))
def build_where_clause(
where_clause_values: List[Any],
sub_filter: dict,
@@ -459,29 +744,98 @@ class SingleStoreDB(VectorStore):
where_clause_values += prefix_args + [key]
where_clause_values.append(json.dumps(sub_filter[key]))
build_where_clause(where_clause_values, filter)
if filter:
build_where_clause(where_clause_values, filter)
where_clause += " AND ".join(arguments)
try:
cur = conn.cursor()
try:
cur.execute(
"""SELECT {}, {}, {}({}, JSON_ARRAY_PACK(%s)) as __score
FROM {} {} ORDER BY __score {} LIMIT %s""".format(
self.content_field,
self.metadata_field,
self.distance_strategy.name
if isinstance(self.distance_strategy, DistanceStrategy)
else self.distance_strategy,
self.vector_field,
self.table_name,
where_clause,
ORDERING_DIRECTIVE[self.distance_strategy],
),
("[{}]".format(",".join(map(str, embedding))),)
+ tuple(where_clause_values)
+ (k,),
)
if (
search_strategy == SingleStoreDB.SearchStrategy.VECTOR_ONLY
or search_strategy == SingleStoreDB.SearchStrategy.FILTER_BY_TEXT
):
search_options = ""
if (
self.use_vector_index
and search_strategy
== SingleStoreDB.SearchStrategy.FILTER_BY_TEXT
):
search_options = "SEARCH_OPTIONS '{\"k\":%d}'" % (
k * vector_select_count_multiplier
)
cur.execute(
"""SELECT {}, {}, {}({}, JSON_ARRAY_PACK(%s)) as __score
FROM {} {} ORDER BY __score {}{} LIMIT %s""".format(
self.content_field,
self.metadata_field,
self.distance_strategy.name
if isinstance(self.distance_strategy, DistanceStrategy)
else self.distance_strategy,
self.vector_field,
self.table_name,
where_clause,
search_options,
ORDERING_DIRECTIVE[self.distance_strategy],
),
("[{}]".format(",".join(map(str, embedding))),)
+ tuple(where_clause_values)
+ (k,),
)
elif (
search_strategy == SingleStoreDB.SearchStrategy.FILTER_BY_VECTOR
or search_strategy == SingleStoreDB.SearchStrategy.TEXT_ONLY
):
cur.execute(
"""SELECT {}, {}, MATCH ({}) AGAINST (%s) as __score
FROM {} {} ORDER BY __score DESC LIMIT %s""".format(
self.content_field,
self.metadata_field,
self.content_field,
self.table_name,
where_clause,
),
(query,) + tuple(where_clause_values) + (k,),
)
elif search_strategy == SingleStoreDB.SearchStrategy.WEIGHTED_SUM:
cur.execute(
"""SELECT {}, {}, __score1 * %s + __score2 * %s as __score
FROM (
SELECT {}, {}, {}, MATCH ({}) AGAINST (%s) as __score1
FROM {} {}) r1 FULL OUTER JOIN (
SELECT {}, {}({}, JSON_ARRAY_PACK(%s)) as __score2
FROM {} {} ORDER BY __score2 {} LIMIT %s
) r2 ON r1.{} = r2.{} ORDER BY __score {} LIMIT %s""".format(
self.content_field,
self.metadata_field,
self.id_field,
self.content_field,
self.metadata_field,
self.content_field,
self.table_name,
where_clause,
self.id_field,
self.distance_strategy.name
if isinstance(self.distance_strategy, DistanceStrategy)
else self.distance_strategy,
self.vector_field,
self.table_name,
where_clause,
ORDERING_DIRECTIVE[self.distance_strategy],
self.id_field,
self.id_field,
ORDERING_DIRECTIVE[self.distance_strategy],
),
(text_weight, vector_weight, query)
+ tuple(where_clause_values)
+ ("[{}]".format(",".join(map(str, embedding))),)
+ tuple(where_clause_values)
+ (k * vector_select_count_multiplier, k),
)
else:
raise ValueError(
"Invalid search strategy: {}".format(search_strategy)
)
for row in cur.fetchall():
doc = Document(page_content=row[0], metadata=row[1])
@@ -503,10 +857,12 @@ class SingleStoreDB(VectorStore):
content_field: str = "content",
metadata_field: str = "metadata",
vector_field: str = "vector",
id_field: str = "id",
use_vector_index: bool = False,
vector_index_name: str = "",
vector_index_options: Optional[dict] = None,
vector_size: int = 1536,
use_full_text_search: bool = False,
pool_size: int = 5,
max_overflow: int = 10,
timeout: float = 30,
@@ -518,10 +874,99 @@ class SingleStoreDB(VectorStore):
2. Creates a new table for the embeddings in SingleStoreDB.
3. Adds the documents to the newly created table.
This is intended to be a quick way to get started.
Args:
texts (List[str]): List of texts to add to the vectorstore.
embedding (Embeddings): A text embedding model.
metadatas (Optional[List[dict]], optional): Optional list of metadatas.
Defaults to None.
distance_strategy (DistanceStrategy, optional):
Determines the strategy employed for calculating
the distance between vectors in the embedding space.
Defaults to DOT_PRODUCT.
Available options are:
- DOT_PRODUCT: Computes the scalar product of two vectors.
This is the default behavior
- EUCLIDEAN_DISTANCE: Computes the Euclidean distance between
two vectors. This metric considers the geometric distance in
the vector space, and might be more suitable for embeddings
that rely on spatial relationships. This metric is not
compatible with the WEIGHTED_SUM search strategy.
table_name (str, optional): Specifies the name of the table in use.
Defaults to "embeddings".
content_field (str, optional): Specifies the field to store the content.
Defaults to "content".
metadata_field (str, optional): Specifies the field to store metadata.
Defaults to "metadata".
vector_field (str, optional): Specifies the field to store the vector.
Defaults to "vector".
id_field (str, optional): Specifies the field to store the id.
Defaults to "id".
use_vector_index (bool, optional): Toggles the use of a vector index.
Works only with SingleStoreDB 8.5 or later. Defaults to False.
If set to True, vector_size parameter is required to be set to
a proper value.
vector_index_name (str, optional): Specifies the name of the vector index.
Defaults to empty. Will be ignored if use_vector_index is set to False.
vector_index_options (dict, optional): Specifies the options for
the vector index. Defaults to {}.
Will be ignored if use_vector_index is set to False. The options are:
index_type (str, optional): Specifies the type of the index.
Defaults to IVF_PQFS.
For more options, please refer to the SingleStoreDB documentation:
https://docs.singlestore.com/cloud/reference/sql-reference/vector-functions/vector-indexing/
vector_size (int, optional): Specifies the size of the vector.
Defaults to 1536. Required if use_vector_index is set to True.
Should be set to the same value as the size of the vectors
stored in the vector_field.
use_full_text_search (bool, optional): Toggles the use a full-text index
on the document content. Defaults to False. If set to True, the table
will be created with a full-text index on the content field,
and the simularity_search method will all using TEXT_ONLY,
FILTER_BY_TEXT, FILTER_BY_VECTOR, and WIGHTED_SUM search strategies.
If set to False, the simularity_search method will only allow
VECTOR_ONLY search strategy.
pool_size (int, optional): Determines the number of active connections in
the pool. Defaults to 5.
max_overflow (int, optional): Determines the maximum number of connections
allowed beyond the pool_size. Defaults to 10.
timeout (float, optional): Specifies the maximum wait time in seconds for
establishing a connection. Defaults to 30.
Additional optional arguments provide further customization over the
database connection:
pure_python (bool, optional): Toggles the connector mode. If True,
operates in pure Python mode.
local_infile (bool, optional): Allows local file uploads.
charset (str, optional): Specifies the character set for string values.
ssl_key (str, optional): Specifies the path of the file containing the SSL
key.
ssl_cert (str, optional): Specifies the path of the file containing the SSL
certificate.
ssl_ca (str, optional): Specifies the path of the file containing the SSL
certificate authority.
ssl_cipher (str, optional): Sets the SSL cipher list.
ssl_disabled (bool, optional): Disables SSL usage.
ssl_verify_cert (bool, optional): Verifies the server's certificate.
Automatically enabled if ``ssl_ca`` is specified.
ssl_verify_identity (bool, optional): Verifies the server's identity.
conv (dict[int, Callable], optional): A dictionary of data conversion
functions.
credential_type (str, optional): Specifies the type of authentication to
use: auth.PASSWORD, auth.JWT, or auth.BROWSER_SSO.
autocommit (bool, optional): Enables autocommits.
results_type (str, optional): Determines the structure of the query results:
tuples, namedtuples, dicts.
results_format (str, optional): Deprecated. This option has been renamed to
results_type.
Example:
.. code-block:: python
from langchain_community.vectorstores import SingleStoreDB
from langchain_community.embeddings import OpenAIEmbeddings
from langchain_openai import OpenAIEmbeddings
s2 = SingleStoreDB.from_texts(
texts,
OpenAIEmbeddings(),
@@ -536,6 +981,7 @@ class SingleStoreDB(VectorStore):
content_field=content_field,
metadata_field=metadata_field,
vector_field=vector_field,
id_field=id_field,
pool_size=pool_size,
max_overflow=max_overflow,
timeout=timeout,
@@ -543,6 +989,7 @@ class SingleStoreDB(VectorStore):
vector_index_name=vector_index_name,
vector_index_options=vector_index_options,
vector_size=vector_size,
use_full_text_search=use_full_text_search,
**kwargs,
)
instance.add_texts(texts, metadatas, embedding.embed_documents(texts), **kwargs)

247
libs/community/tests/integration_tests/vectorstores/test_singlestoredb.py
View File

@@ -1,4 +1,5 @@
"""Test SingleStoreDB functionality."""
import math
import os
import tempfile
from typing import List
@@ -67,11 +68,76 @@ class RandomEmbeddings(Embeddings):
return [np.random.rand(100).tolist() for _ in uris]
class IncrementalEmbeddings(Embeddings):
"""Fake embeddings with incremental vectors. For testing purposes."""
def __init__(self) -> None:
self.counter = 0
def set_counter(self, counter: int) -> None:
self.counter = counter
def embed_query(self, text: str) -> List[float]:
self.counter += 1
return [
math.cos(self.counter * math.pi / 10),
math.sin(self.counter * math.pi / 10),
]
def embed_documents(self, texts: List[str]) -> List[List[float]]:
return [self.embed_query(text) for text in texts]
@pytest.fixture
def texts() -> List[str]:
return ["foo", "bar", "baz"]
@pytest.fixture
def snow_rain_docs() -> List[Document]:
return [
Document(
page_content="""In the parched desert, a sudden rainstorm brought relief,
as the droplets danced upon the thirsty earth, rejuvenating the landscape
with the sweet scent of petrichor.""",
metadata={"count": "1", "category": "rain", "group": "a"},
),
Document(
page_content="""Amidst the bustling cityscape, the rain fell relentlessly,
creating a symphony of pitter-patter on the pavement, while umbrellas
bloomed like colorful flowers in a sea of gray.""",
metadata={"count": "2", "category": "rain", "group": "a"},
),
Document(
page_content="""High in the mountains, the rain transformed into a delicate
mist, enveloping the peaks in a mystical veil, where each droplet seemed to
whisper secrets to the ancient rocks below.""",
metadata={"count": "3", "category": "rain", "group": "b"},
),
Document(
page_content="""Blanketing the countryside in a soft, pristine layer, the
snowfall painted a serene tableau, muffling the world in a tranquil hush
as delicate flakes settled upon the branches of trees like nature's own
lacework.""",
metadata={"count": "1", "category": "snow", "group": "b"},
),
Document(
page_content="""In the urban landscape, snow descended, transforming
bustling streets into a winter wonderland, where the laughter of
children echoed amidst the flurry of snowballs and the twinkle of
holiday lights.""",
metadata={"count": "2", "category": "snow", "group": "a"},
),
Document(
page_content="""Atop the rugged peaks, snow fell with an unyielding
intensity, sculpting the landscape into a pristine alpine paradise,
where the frozen crystals shimmered under the moonlight, casting a
spell of enchantment over the wilderness below.""",
metadata={"count": "3", "category": "snow", "group": "a"},
),
]
@pytest.mark.skipif(not singlestoredb_installed, reason="singlestoredb not installed")
def test_singlestoredb(texts: List[str]) -> None:
"""Test end to end construction and search."""
@@ -505,3 +571,184 @@ def test_singestoredb_add_image2() -> None:
output = docsearch.similarity_search("horse", k=1)
assert "horse" in output[0].page_content
drop(table_name)
@pytest.mark.skipif(not singlestoredb_installed, reason="singlestoredb not installed")
def test_singlestoredb_text_only_search(snow_rain_docs: List[Document]) -> None:
table_name = "test_singlestoredb_text_only_search"
drop(table_name)
docsearch = SingleStoreDB(
RandomEmbeddings(),
table_name=table_name,
use_full_text_search=True,
host=TEST_SINGLESTOREDB_URL,
)
docsearch.add_documents(snow_rain_docs)
output = docsearch.similarity_search(
"rainstorm in parched desert",
k=3,
filter={"count": "1"},
search_strategy=SingleStoreDB.SearchStrategy.TEXT_ONLY,
)
assert len(output) == 2
assert (
"In the parched desert, a sudden rainstorm brought relief,"
in output[0].page_content
)
assert (
"Blanketing the countryside in a soft, pristine layer" in output[1].page_content
)
output = docsearch.similarity_search(
"snowfall in countryside",
k=3,
search_strategy=SingleStoreDB.SearchStrategy.TEXT_ONLY,
)
assert len(output) == 3
assert (
"Blanketing the countryside in a soft, pristine layer,"
in output[0].page_content
)
drop(table_name)
@pytest.mark.skipif(not singlestoredb_installed, reason="singlestoredb not installed")
def test_singlestoredb_filter_by_text_search(snow_rain_docs: List[Document]) -> None:
table_name = "test_singlestoredb_filter_by_text_search"
drop(table_name)
embeddings = IncrementalEmbeddings()
docsearch = SingleStoreDB.from_documents(
snow_rain_docs,
embeddings,
table_name=table_name,
use_full_text_search=True,
use_vector_index=True,
vector_size=2,
host=TEST_SINGLESTOREDB_URL,
)
output = docsearch.similarity_search(
"rainstorm in parched desert",
k=1,
search_strategy=SingleStoreDB.SearchStrategy.FILTER_BY_TEXT,
filter_threshold=0,
)
assert len(output) == 1
assert (
"In the parched desert, a sudden rainstorm brought relief"
in output[0].page_content
)
drop(table_name)
@pytest.mark.skipif(not singlestoredb_installed, reason="singlestoredb not installed")
def test_singlestoredb_filter_by_vector_search1(snow_rain_docs: List[Document]) -> None:
table_name = "test_singlestoredb_filter_by_vector_search1"
drop(table_name)
embeddings = IncrementalEmbeddings()
docsearch = SingleStoreDB.from_documents(
snow_rain_docs,
embeddings,
table_name=table_name,
use_full_text_search=True,
use_vector_index=True,
vector_size=2,
host=TEST_SINGLESTOREDB_URL,
)
output = docsearch.similarity_search(
"rainstorm in parched desert, rain",
k=1,
filter={"category": "rain"},
search_strategy=SingleStoreDB.SearchStrategy.FILTER_BY_VECTOR,
filter_threshold=-0.2,
)
assert len(output) == 1
assert (
"High in the mountains, the rain transformed into a delicate"
in output[0].page_content
)
drop(table_name)
@pytest.mark.skipif(not singlestoredb_installed, reason="singlestoredb not installed")
def test_singlestoredb_filter_by_vector_search2(snow_rain_docs: List[Document]) -> None:
table_name = "test_singlestoredb_filter_by_vector_search2"
drop(table_name)
embeddings = IncrementalEmbeddings()
docsearch = SingleStoreDB.from_documents(
snow_rain_docs,
embeddings,
distance_strategy=DistanceStrategy.EUCLIDEAN_DISTANCE,
table_name=table_name,
use_full_text_search=True,
use_vector_index=True,
vector_size=2,
host=TEST_SINGLESTOREDB_URL,
)
output = docsearch.similarity_search(
"rainstorm in parched desert, rain",
k=1,
filter={"group": "a"},
search_strategy=SingleStoreDB.SearchStrategy.FILTER_BY_VECTOR,
filter_threshold=1.5,
)
assert len(output) == 1
assert (
"Amidst the bustling cityscape, the rain fell relentlessly"
in output[0].page_content
)
drop(table_name)
@pytest.mark.skipif(not singlestoredb_installed, reason="singlestoredb not installed")
def test_singlestoredb_weighted_sum_search_unsupported_strategy(
snow_rain_docs: List[Document],
) -> None:
table_name = "test_singlestoredb_waighted_sum_search_unsupported_strategy"
drop(table_name)
embeddings = IncrementalEmbeddings()
docsearch = SingleStoreDB.from_documents(
snow_rain_docs,
embeddings,
table_name=table_name,
use_full_text_search=True,
use_vector_index=True,
vector_size=2,
host=TEST_SINGLESTOREDB_URL,
distance_strategy=DistanceStrategy.EUCLIDEAN_DISTANCE,
)
try:
docsearch.similarity_search(
"rainstorm in parched desert, rain",
k=1,
search_strategy=SingleStoreDB.SearchStrategy.WEIGHTED_SUM,
)
except ValueError as e:
assert "Search strategy WEIGHTED_SUM is not" in str(e)
drop(table_name)
@pytest.mark.skipif(not singlestoredb_installed, reason="singlestoredb not installed")
def test_singlestoredb_weighted_sum_search(snow_rain_docs: List[Document]) -> None:
table_name = "test_singlestoredb_waighted_sum_search"
drop(table_name)
embeddings = IncrementalEmbeddings()
docsearch = SingleStoreDB.from_documents(
snow_rain_docs,
embeddings,
table_name=table_name,
use_full_text_search=True,
use_vector_index=True,
vector_size=2,
host=TEST_SINGLESTOREDB_URL,
)
output = docsearch.similarity_search(
"rainstorm in parched desert, rain",
k=1,
search_strategy=SingleStoreDB.SearchStrategy.WEIGHTED_SUM,
filter={"category": "snow"},
)
assert len(output) == 1
assert (
"Atop the rugged peaks, snow fell with an unyielding" in output[0].page_content
)
drop(table_name)